Technology

Dear friends and partners!
We are delighted to inform you about one of the most important events of this year: we have released a new set of products, combined into one platform: JS Platform. Most of the products are already available worldwide.
We challenged rather difficult task: to find and develop fundamentally new technologies, however this time we managed to outdo ourselves and released new generation of products, providing significant additional effect in comparison to the previous versions. For the last few months we managed to complete over 700 Data science, Risk management and IT tasks in order to release our new product.
The new product generation has a range of significant changes from the perspective of efficiency for our partners. Here are the main updates, which you will find in JS Platform:
<ol>
<li>15+ new parameters and variables in JuicyScore and JuicyID;</li>
<li>Antifraud scoring with Gini increase;</li>
<li>45+ new anomalies and randomizers tests, which also let to detect various fraudulent digital instruments;</li>
<li>For the first time, a separate group of tests will be allocated for remote access detection on a user’s device;</li>
<li>The system of estimated income verification has been improved significantly for mobile SDK users;</li>
<li>We elaborated totally new approaches to self-learning models and their efficiency increase;</li>
<li>Offline to online solution has been developed. It helps to improve risk assessment significantly without affecting the conversion rate;</li>
<li>New device authentication technologies have been developed, including device randomization detection;</li>
<li>We created on-premise/local variant of JuicyBio, later on we are going to focus only on JuicyBio on-premise development. This solution is now undergoing the last stages of testing;</li>
<li>Service operating speed has been increased by 50%+</li>
</ol>
Despite the difficult situation all over the world as well as the essential changes in the work of all the financial market players, Juicy Team keeps up high development speed in order to extend the functionality of our products and generate more value to our clients and partners.
We permanently conduct researches to identify new attributes, work on strengthening of our expert valuation in digital risk management to help our clients and partners in their day-to-day work, fraud prevention and risk assessment. Thanks to all of you for your support and sharing your ideas!
Yours JuicyTeam

Dear friends and partners! We are delighted to inform you about one of the most important events of this year: we have released a new set of products, combined into

medium_API-14-1.jpg

small_API-14-1.jpg

thumbnail_API-14-1.jpg

API-14-1.jpg

Fintech

Technological agility helps fintech leaders to stand out from competition and increase the market share. Strong and reliable partners make this growth even more sustainable and significant. ID Finance in Mexico (MoneyMan brand - moneyman.com.mx) has been cooperating with JuicyScore, risk management and antifraud leader, for the last 3 years.

In comparison with other solutions JuicyScore has a number of benefits, such as stable Device ID and highly informative data vector, which allowed JuicyScore to become a part of the Company's decision making strategy. Moreover JuicyScore let to strengthen risk management technological stack and company's scoring models.

_Which variables performed best on being plugged into the model?_

Dmitry Mikolauskas, Risk Director at ID Finance Mexico has shared his experience of working with JuicyScore team: “We managed to increase our model separating capacity 1,4 times, adding to the predictive model combinations of variables related to user behaviour on the web-site. For example, cursor distance covered, cursor movement speed, time spent on a web-page, screen idle time, and device quality segmentation.”

Dmitry Mikolauskas also noted “According to A/B test results of the predictive model, which included JuicyScore latest generation technologies, the updated version showed significant effect, which let us to increase new applications approval rate 1,5 times, given that risk management metrics remained at the same level.”

Group of companies ID Finance is currently cooperating with JuicyScore in Mexico and a number of other countries, for JuicyScore had already showed a great performance on the markets of Asia and Europe. JuicyScore distinctive feature is unique devicometry technologies, that help to build probabilistic device fingerprint and device authentication as well as does not store, process or even uses online users direct identifiers. JuicyScore solutions really helps to thoroughly deal with the risks, occuring on a device level or within the frames of an internet session.

Alexander Akhlomov, Chief Product Officer at JuicyScore believes that despite the fact that LatAm bank lending market may be considered to be rather developed, millions of people living in this region still do not have access to traditional financial products. ID Finance offers financial products range designed for financial access improvement of an abundance of the so-called “unbanked” on “underbanked” population - i.e. those people, who do not have access to banking products or have a range of limitations. By means of technological development the company offers tailored solutions for the consumers and has strongly taken place of one of the leading fintechs in the region.

_If an online business does not store and process the direct user identifiers, what are the main advantages of such approach?_

Natural persons' direct identifiers include identity document information, full name, full phone numbers, full bank cards numbers, photos, email addresses and other data sources, which allow to directly identify an individual or build a direct system of communications with a user. Despite huge investments of companies in IT infrastructure and organizational measures, in some cases direct identifiers, contacts and sensitive data breach may happen, which usually leads to reputational and financial losses for all the parties.

_About MoneyMan MX_

ID Finance Mexico is an international fintech company offering term loans in various countries around the world. The MoneyMan brand, part of the group of companies, offers its customers a PDL product of up to MXN 15,000 with maturities of up to 90 days. The company deals with the issues related to financial inclusion in Latin America, being one of the key players in the region offering alternative lending products.

Technological agility helps fintech leaders to stand out from competition and increase the market share. Strong and reliable partners make this growth even more sustainable and significant. ID Finance in Mexico (MoneyMan brand - moneyman.com.mx) has been cooperating with JuicyScore, risk management and antifraud leader, for the last 3 years.

31 Aug 2023_MoneyMan MX leverages JuicyScore to increase approval rate x 1.5 times.png

MoneyMan MX leverages JuicyScore to increase approval rate x 1.5 times

Automatisation in Risk Management has long been a necessity - risks in online business keep growing immensely, while manual data processing is unable to guarantee effective protection from online fraud. Outdated technological approach may do harm to both - online business finances and reputation.

We are really proud to announce our new product - automatic risk management reporting and benchmarking. This product will allow experts to compare fraud risk indicators with aggregated data from other companies in the same region as well as to receive very clear feedback and guidelines to improve decision-making system and application flow segmentation setup.

This product also enables risk experts to assess risks and benchmark traffic based on our cutting-edge methodology and to gain insights into company's current problem points. New product is available to all our clients in their personal accounts.

Our approach is based on the group of certain metrics - First Attempt Fraud, Secondary Attempt Fraud and High Risk Applications - all together they form full group of events in terms of risk categories and risky applications with a regional breakdown. On applying this methodology it's easy to conduct segmentation on the whole application flow and highlight risk zones.

_**Practical application of the product**_

_First Attempt Fraud prevention category_ indicates the level of traffic “toxicity”, flowing to credit organization online resource including technological fraud, network anomalies and other high risk markers. If this indicator is above the market average for your segment, it means that your current anti-fraud policy has a significant potential to grow.

![](/uploads/MNLT_pics_001_b6a7227e69.jpeg)

_Secondary Attempt Fraud Prevention category_ indicates the proportion of repeated devices in application flow, i.e. those devices, which have been spotted previously on the creditor's resource and no special tools for device fingerprint distortion have been used. If Secondary Attempt Fraud metrics are higher than market average, it means that the company should use anti-fraud technologies at the device authentication level (in particular, repeated devices parameters) in more effective way. In order to handle such traffic, it's worth using various verification tools related to alternative data sources. In combination with other markers of JuicyScore response vector, for example, consistency of application data checks, confirming that the application data belongs to this particular user etc. you can work with such traffic easily and effectively.

![](/uploads/MNLT_pics_002_04355f836f.jpeg)

_High Risk Applications category_ in most cases is the most extensive one in terms of the number of observable parameters and share over the traffic flow. If HRA metrics in the IDX5 section are 10% or more lower than market average, it means that the initial traffic is of lower quality in the context of disposable income. Device quality correlates quite strongly with disposable income indicator. You should pay attention to those methods of traffic filtering and sources of its generation in order to improve the situation in comparison with competitors.

![](/uploads/MNLT_pics_003_6f17a7f01a.jpeg)

This group also includes risk markers of various nature, so the most effective tool is to build a scoring model, including JuicyScore data with further cutoff levels determination, applicable actions and additional data sources. As an alternative solution a creditor may arrange sets of rules and filters in order to highlight high-risk areas. You also can use this information at the stage of traffic sources selection.

_You can discover more details in terms of risk markers and various parameters in product's paid mode._

All graphs are provided with a color code system, the colors reflect the current situation based on the last two columns.

- Green light turning into red means negative dynamics, deterioration of the situation (it is worth paying attention to the possible causes of its occurrence);
- Green light turning into green means that current situation is steadily better than market average;
- Red light turning into green means positive dynamics, situation is improving (continue to monitor the situation);
- Yellow color turning into yellow - steadily near the market level;
- In case both lights are red, the situation is consistently worse than the market average (you need to refer to recommendations);
- Any light turning into grey means that there is not enough data in order to analyse specified period.

![](/uploads/MNLT_Jun_001_9aa125b8d2.jpeg)

_**JuicyScore methodology fundamentals**_

The key positive factor of this approach is that all the high-risk sessions inevitably fall into one of the four groups, therefore, we can spot 100% of all possible online fraud attempts. Another advantage is that this approach does not depend on the country, technologies, local legislation etc., which means that one can compare various online resources and countries.

The key difficulty lies in the risk assessment itself, i.e. in finding suitable variables/markers/behavioral signs that would characterize the risk for a particular session and would allow it to be assigned to one of the groups, as well as the variety of available data and tools. The solution to this problem lies in the construction and design of a scoring system to counteract fraud.

![](/uploads/Space_of_the_devices_En_001_b456617a52.png)

You can learn more about our methods [here](https://juicyscore.ai/blog/en/rare-events-risk-assessment-how-data-science-can-help-business-thrive-en/?ref=juicyscore.com).

Naturally, in antifraud and risk management there is no single and universal approach that would solve any problem and give 100 percent result - every expert is in constant search of a balance between data value and a stable approach. However, with our new product, we will significantly simplify the internal risk assessment procedures in most companies: with the help of automation, you can be one step ahead of other companies that have not yet fully realized the need to move to automation.

JuicyTeam always provides its customers with access to the best tools, we keep our researches in 30 countries and dozens of companies, so we are confident that new methods and tools for automating risk management will help to reduce risks in online business significantly and improve the quality of decision making.

Automatisation in Risk Management has long been a necessity - risks in online business keep growing immensely, while manual data processing is unable to guarantee effective protection from online fraud.

Risk-1.png

Risk management automatic reporting: new product announcement

Risk

We are constantly searching for cutting-edge informative data, which provide our clients with the opportunity to assess credit and operational risks really effectively. We discover new attributes, which improve quality of decision-making process and also pay great attention to design mechanics and ensure transparency of the results.

This presents a major challenge: the assessment of markers, which have been found and the way to maximise their potential in models. This problem is connected to the fact, that most part of those markers is related to rare events group and their standard assessment is much harder than usual or is nearly impossible in terms of standard methods and procedures.

In 2015-2017 JuicyTeam had designed rare events risk assessment technology which allowed to deliver interpretable and stable result independent from geography and assessment stage within the frames of credit conveyor or decision making process.

_What are the main points of JuicyScore approach?_

Our methodology is divided in 2 parts:

1. Methods of search and localisation of new markers;
2. Methods of valuation of the markers in risk assessment process.

The first part is devoted to rare events search and determining based on precise user device authentication as well as determining of various parameters characterising user device, its environment and methods of its using. We collect more than 65 000 events or the so-called data points on user device during one online session, most part of those are aggregated in data sets on a device.

Our methodology is based on the division of the entire session space into 4 groups, which allows to prioritize the search and estimate the probability / density of new markers discovery:

- FAP (First attempt fraud) is a group where we identify fraud basing on intentions for devices that have not been spotted before within internet session framework;
- SAP (Secondary Attempt Fraud) is a group where we identify fraud basing on intentions for devices that have not been spotted before within the framework of the online resource;
- HRA (High Risk Applications) is a group that does not fall into the above mention two groups, but the risk in this group is rather high;
- Other sessions with low fraud risk (with the advent of new risk assessment technologies some sessions can be removed from this group and may be added to FAP/SAP/HRA groups).

The second part of our methodology is related to risk assessment technologies / rare events scoring or markers. This technology is based on the complex of markers assessment connected with physical meaning and its parameters which may be logically grouped in one of the directions. We allocate 10 such directions - indexed IDX 1-10.

What's the use of analytical work? Every single index (both - statistically and logically) covers all the significant markers in one direction and may be used in classical methods of mathematical statistics or data science. _Ok, but why IDX variable are so helpful in risk assessment?_

The essential part of IDX indexes is a combination of rare events and factors related to them, collected by means of Deep Machine Learning algorithms to one variable, which may be used either for modeling or for integration to financial institutions decision-making system.

It should be noted that IDX indexes were created as Gaussian variables. Such approach has a number of bonuses: on the one hand they remain statistically significant in any other model, and, on the other hand, they allow structuring the entire probability space no matter what type of fraud events we have to deal with.

We are going to share rather interesting data on \*\*\*\*10 indexes as well as best practices for their application in risk management. It should be noted that we provide a “clean sample”, which implies the data that did not reveal a high probability of fraud through technical manipulations with the device or significant anomalies in the Internet connection.

## IDX1: Stop markers

IDX1 is a combination of 50+ rare events, which show high probability of fraud through technical manipulation of the device. This variable includes the whole variety of device randomization tools, techniques of interference into "digital fingerprint" as well as determines the most dangerous markers of user high risk behaviour and network connection markers. The variable can be used both in rules and also as a component of a fraud prevention model to identify the most dangerous customer segments. The risk level raises along with the parameter value, high values may be used as filters for automatic denial.

The graph shows aggregated data for companies that have already been using IDX 1 in their decision making model.

![](/uploads/IDX_1_En_f59cf808f5.jpeg)

Apart from the use of randomizers, which are identified with IDX1 as well as with the separate stop factors - for example, copying of another device's session (vector's variable session clone), identification of anomalies in the header of the web session (vector's variable UserAgent Issue), indication of manipulation with the color palette (variable Canvas blocker) - also great attention should be paid to browser or operating system anomalies.

## IDX2 - User behaviour markers

This aggregated variable is a combination of various user behaviour markers on online business web resource. In JuicyScore vector there are dozens of markers which are in some way related to user behaviour and the main issue in terms of variable construction is to find such stable markers, which would allow, aggregated in one, to identify high risk segments regardless of online business operating geography.

Variable IDX2 is based on the factors related to various user behavior or device utilization categories. On the one hand, it incorporates many factors connected with frequency characteristics. For example, the number of applications or requests for financial product obtaining from one user or the same device with a certain time periodization or without it over the entire history. On the other hand, there are also such parameters included into this variable, which identify stability or, on the contrary, data variability used in a credit application. A wide variety of such data on the same device or related to one virtual user indicates a high operational risk.

At the same time high frequency of applications without data manipulation may indicate a higher credit risk (the so-called credit shopping, when a borrower applies for multiple credits to many financial institutions within a short period of time). The presence of both - high frequency and high variability of data on the same device or virtual user - is a strong sign of high operational risk.

On the graph below you can see an approximate generalized graph of the change in the level of risk based on ranges of the variable.

![](/uploads/IDX_2_En_2c31ad9cdb.jpeg)

Moreover, this variable also includes a number of factors regarded as high risk markers of user behaviour, which are not related to those two categories. For example, the way of form application filling, device utilization etc. Basically, we can see the combination of rare events of medium and high risk, which, integrated together in a certain way, may be used in a decision-making system and models, constructed by means of classical Gaussian methods.

## IDX 3: Device Markers

IDX3 is a combination of secondary risk markers and device anomalies, where every single anomaly may reflect possible risk, which should be taken into account during the borrower's verification. The combination of such markers highlights a high risk segment. Similarly to IDX1, the risk level is raising along with the variable value, and high values may be used for automatic denial.

![](/uploads/IDX_3_En_765ccefb10.jpeg)

Cross-tabulation of various indexes values may also provide some important information on risk level. For example, if IDX1 and IDX 3 are equal to 0, thus there is a high probability that no randomisation and virtualization are identified, and we are dealing with a physical device.

## IDX4: Internet connection markers

IDX4 is a combination of network parameters and anomalies, the higher values are may be used for fraud detection purposes.

This index takes into account such indicators as the type of IP used, device time zone and its match with the real local time zone, the use of DNS configurations, etc.

![](/uploads/IDX_4_En_e1d77460f5.jpeg)

## IDX 5 - Device quality index

Speaking about the operational risk assessing, the prior task is to reject applications of users with high risk of a loan defaulting and non-payment risk, while the main objective of credit risk assessment is to find those segments of users, which can be provided with financial products with the right and accurate parameters. Aggregate variable IDX5 falls into the second category. Using this variable financial institutions are able to improve credit risk segmentation of the incoming flow significantly. It is particularly important when there is a great lack of strong data institutions or quality of the data provided by them is rather low.

Device quality function is the level of its cost, which is affected by the following data categories: device category (desktop or mobile), its technical metrics aggregate (storage capacity, number of cores, storage quality etc.) and device manufacturer (famous brand of the device or a no-name). It is important to note that devices with certain technical characteristics anomalies are not included into this index in order to convey a higher orthogonality level with the other aggregate variables IDX.

Every device has a big range of technical metrics and parameters, which affect on its quality and may be used also in order to assess credit risk. That is why developing this index it was highly important to identify such metrics and model factors in order to ensure the stability of value range of every factor, keeping in mind that we also need to provide the stability of value range of the index itself and to enhance its sharing ability. Speaking about stability it is also important to note that it should be ensured in time and also among all the geographies of our clients’ businesses.

![](/uploads/IDX_5_En_031a2d2243.jpeg)

Within the frames of Device Quality Index value interpretation, the part of the flow with low values of this Index identifies a segment with high credit risk and low level of disposable income. The part of the flow with high values of this Index makes it possible to identify segments with a low level of credit risk.

## IDX 6 - Internet Infrastructure Quality

Internet infrastructure quality is a good component for credit risk assessment and social fraud detection, may be used as a parameter for credit score model. A value higher than or equal to 2 usually highlights more premium sub-segment in the application flow. This index includes such indicators as the IP of the country and the region, Internet infrastructure quality in the region, connection speed and quality etc.

![](/uploads/IDX_6_En_d5bab7cbc3.jpeg)

## IDX 7: Device Applications Quality

The variable is valid only for mobile SDK and represents an aggregated assessment of the applications installed on the applicant's device, it is recommended to be used to identify credit risk and social default risk. Please note that to evaluate this setting, you must configure the collection of the final list of applications as part of the SDK connection.

![](/uploads/IDX_7_En_abf4c589c3.jpeg)

## IDX8: Device Credentials Variability

Device credentials variability, aggregated assessment of applicant’s credentials on credit applications, an indicator of loan application data manipulation. The higher the value, the lower the risk. Thus, in this index, we have included such indicators as repeated multiple digits of the phone/user region, as well as a repeated device fingerprint.

![](/uploads/IDX_8_En_75e3114ea0.jpeg)

## IDX9: Device Applications Risk

Device applications risk is the aggregated assessment of the applications installed on the device. The risk grade is showing the level of risk of the devices installed and identified. Please note that in order to evaluate this setting you must configure the collection of the final list of applications as part of the SDK connection. We at JuicyScore perform detailed analysis of a wide range of applications that can have a significant impact on operational or credit risk (for example, remote access applications, malicious applications, applications with a toxic reputation, etc.).

We constantly expand our libraries with descriptions and characteristics of new applications. It is also important to note that we do not and have not sought to assess the risk of each of the available applications, as we consider this to be excessive.

![](/uploads/IDX_9_En_4694fab03f.jpeg)

The value of the variable shows the aggregate risk level of the applications installed on the applicant's device.

## IDX10: EI estimation

![](/uploads/IDX_10_En_9227fcd278.jpeg)

Income estimation index shows the grade of disposable income. The higher the value the higher the income grade, the lower the risk and vice versa. This parameter is highly recommended for use in the assessment of operational and credit risk.

_Why our methodology is so effective?_

In order to fight fraud easily and effectively, you need to have as many useful and functional tools as possible. However, everyone knows that in risk management there are no universal solutions or tools that would have the same efficiency and payback for all companies. Our solution has a number of advantages that distinguish it from others:

- Our methodology covers the entire probabilistic space of events;
- The method gives a multiple economic payback: depending on the value of the asset and the type of risk, as well as the place in the credit pipeline;
- Installing JuicyScore fully satisfies the need to obtain the maximum set of data for the Internet session, device and online behavior of the user.

The IDX indexes may also be set up as stop markers to cut off the flow in the red zone. Behind each red zone is the result of complex calculations, the results of numerous use cases studies in 30 countries, dozens of companies.

We pay great attention to the mechanics of our indexes design and we keep looking for the most useful and informative data that helps to manage credit and operational risks. JuicyScore allows to simplify the process of using the latest Data Science cutting edge technologies in risk management and anti-fraud.

If you would like to know more about our approach or to get a free consultation about risk level in your business please contact [help@juicyscore.com](mailto:help@juicyscore.com?ref=juicyscore.com) or [info@juicyscore.com](mailto:info@juicyscore.com?ref=juicyscore.com).

We are constantly searching for cutting-edge informative data, which provide our clients with the opportunity to assess credit and operational risks really effectively. We discover new attributes, which improve quality

Swans-1.png

Rare events risk assessment: how data science can help business thrive?

Remote access technologies are very popular in the IT industry. The scenario of remote access application is quite extensive: testing programs, solving technical issues, remote work setup. However over the past few years financial institutions have seen how these technologies may do harm to both - lenders and potential borrowers: the increased growth rate of online fraud is caused not only by technologically advanced fraudsters improving their tools, but also by expansion of remote access technologies use by a wide range of individuals, including technically advanced ones and much less tech-savvy.

Today we will talk about the best methods, which will help online businesses to deal with the problem of social engineering and remote access and also will shed some light on the issue of protection from these types of risk.

There is one glaring problem with remote access technologies in credit risks: they can be used by fraudsters to log in to a personal account in order to read SMS with a passcode or apply for a loan on user's behalf. In addition to that, there is an abundance of additional social problems: fraudsters usually attack socially unprotected population, for example, aged people.

Financial institutions constantly develop online channels for financial services rendering, expand the range of products and services, make their products more accessible to different categories of customers and also solve many social and economic problems associated with financial inclusion.

Being aware of this, fraudsters, in turn, improve their tools, use a combination of social engineering and remote access technologies to attack such categories of customers. Consequently, these types of attacks are often effective, so making financial institutions aware of the remote access tools on the borrower's device makes it possible to counteract such situations and reduce the number of such fraud cases.

_What is a remote access technology?_

Remote access technology refers to any set of IT tools used to connect, access or manage devices, resources and data stored on a local network from a remote geographic location. You may ask, so how this is different from using a cloud solution? The thing is that remote access technologies provide access to an on-premises environment rather than being hosted out of the office in a shared environment and they are accessible via the Internet. Three most common remote access technologies are Remote Desktop Services, remote access software and VPNs.

We can't but agree, remote access remains a very important tool for those companies, which have not yet migrated to the cloud or need access to on-premise computers or resources. With this access users can manage files and data stored on a remote device, it also allows them to continue to collaborate and keep a productive work from any place in the world.

_Social engineering: how it works?_

Social engineering is a fairly broad term that refers to a type of fraud when criminals influence a person through psychological manipulation in order to directly steal money or in order to obtain sensitive and personal data (for example, passport data, login and password from an online personal bank account) for the subsequent commission of a crime and theft of money. The main channels are social media and phone calls.

According to JuicyScore, 60-70% of professional social engineering scams follow this pattern: a user receives a call from an unknown person, who pretends to be an employee of a bank or any governmental authority. In some cases, number spoofing, call center imitation and IVR can be used. After that a scammer fraudulently lures the data, reporting false information, for example, that passwords and accounts are at risk and requests user's personal data: card details or an SMS code. Also he usually offers to download “official” bank application in order to protect user's funds. Such application is often nothing but a remote access program. On collecting victim's personal data, scammers enter personal account using remote access technologies and make a purchase, card2card transfer or apply for a loan on user's behalf.

The statistics of successful attacks performed in this way is shocking: according to our data, when a fraudster draws up a loan agreement with the withdrawal of funds to a debit card, the client gives away 2-3+ SMS or one time password codes on average.

_Main types of attacks_

There are several types of attacks associated with the use of remote access technologies. According to our data, the first two types are the most common:

- Use of special malicious plug-ins or "prepared" browsers;
- An attack made by infected computers (most often through virus mailings, containing malicious links with dangerous programs and applications);
- Passive social engineering - when a person gives away some information during a call or installs special applications, while fraudsters do the rest;
- Active social engineering - when a person does everything with ones own hands (including the last transfer of funds or applying for a loan), and the attackers only tell the victim what to do (often they are in the office or somewhere not very far the person - they tell the victim to install malware, download the application from a link, say password from SMS and so on). This is the most difficult case to spot and to investigate as well.

_How we can help to spot remote access?_

As you may already know, the best way to protect personal data is not to use it. JuicyScore team is able to determine the presence of remote access technologies during the session, to be precise, directly at the very moment of filling in an application for a financial service or product obtaining.

We tracked the dynamics of devices with signs of remote access we spotted recently and came to the conclusion that this indicator tends to grow.

![](/uploads/Eng_001_1_f9a3a7c9b6.jpeg)

We developed a few ways, which proved to be the best in spotting remote access technologies. They also can be effective in remote access fraudulent applications detection.

We collect as much information as possible that a browser can legally give out through different browser APIs, after that we analyze such data and identify typical signs of virtual machines or malicious plugins presence. Here are some of the features of our approach we use in order to spot remote access and social engineering:

- Dangerous plug-ins and randomizers specific to remote access spotting;
- Active dangerous applications and remote access applications on the device spotting;
- Injections during online sessions spotting;
- Specific anomalies in the traffic characteristics of the device and its connection spotting;
- Device parallel activity presence spotting.

Moreover, we are constantly discovering new tools for detecting social engineering technologies. Based on our data, we can say that the number of sessions with signs of remote access is about 2.5-3% of the total flow of applications. This fact may serve as a signal for online lenders that it is high time to pay attention to such applications and introduce additional verification.

For example, by means of repeated verification, creating pop-up warnings for users, temporarily blocking activities / conducting financially significant transactions, creating a list of allowed device prints for an existing client, comparing Device IDs or device fingerprints during registration and login, analyzing other behavioral markers. All these steps will significantly reduce the risk of fraud, which in this case may cause not only the risk of material losses for both parties, but also generally reduces trust of entire customer segments in the online channels of banks and other financial institutions.

Remote access technologies are very popular in the IT industry. The scenario of remote access application is quite extensive: testing programs, solving technical issues, remote work setup. However over the

Remote-access-1.png

Remote access and social engineering risks prevention best practices

Virtual machine may be a great instrument for dealing with issues related to data protection and safe program delivery, code testing, software performance research. However, the use of such technologies order to obtain financial services may indicate user's malicious intentions and, therefore, high risk for the business. 

We have spotted rising number of virtual machines in application flow recently, thus, we would like to share our statistics data on virtual machines percentage ratio in various regions as well as to highlight the most effective approaches to identify this threat.

_What is a virtual machine and how it is related to risk assessment technologies?_

The core of our fraud prevention approach is careful and accurate device authentication, taking into account various parameters of any device, its environment and use. Virtual machines detection has become very relevant issue recently, for we have spotted a higher degree of virtual devices use in the regions of our presence.

What is a virtual device or machine? It is any type of device (PC, tablet PC, smartphone etc.) created with special software or program code. In fact, such device doesn't differ from any physical computer/laptop/smartphone or even a server. It also has a processor unit, memory module, data and file storages, and it also can connect to the Internet if needed. However while real computers have physical storage system, memory modules and microprocessor chipsets, virtual machines or software-defined computers exist only as a code.

Virtual device make company's IT-infrastructure operations much easier and also increases productivity due to resource optimization. For example, several servers may be combined into one virtual machine and vice versa - one can split one powerful machine into several smaller servers, depending on the current needs of the company. However, the use of such technologies in financial products and services obtaining may indicate user's malicious intentions and, therefore, may lead to high risk for the business.

There are several types of virtual machines. The first one may be described as various software solutions, which bear signs of virtualization. For example, private mode in a browser or LockDown mode on a device. Such mechanisms are often used to protect users privacy. Solutions with signs of virtualization, as a rule, do not carry significant risks, but sometimes the risk of such software solutions can be above average.

The second type can be attributed to the most dangerous ones - the actual virtual machine - a system that emulates computer hardware (actually, it may be both software and hardware). It is necessary to distinguish these two types of virtual machines in order to deal with them in a more effective way.

## _Virtual machines average fraud level_

As part of the analysis of virtual machines detection, we reviewed the statistics on a selection of variables: to be more precise, we estimated the average risk not only for virtual machines, but for all records.

Based on our analysis, the average risk across all applications with the signs of virtual machines presence was on average 1.3-1.5 times higher than the average risk across all records. At the same time, those companies that do not pay attention and do not filter virtual machines have a default rate 2.5-3 times higher than average.

Also we assessed growth dynamics of the virtual machines use in various regions over the past 6 months.

![](/uploads/fin_Eng_003_cb243a8f76.jpeg)

Over the past six months we noticed an increase in the use of virtual machines. December surge is quite easy to explain: the amount of online-fraud attacks usually rises dramatically before the holidays along with the consumers’ need for shopping and, as a result, the need for additional credit means.

In addition to that, we analyzed the distribution of traffic by virtual machines over the past 2 months with a regional breakdown. On the graph below you can see the percentage of applications with signs of virtual machines use from the total flow of applications:

![](/uploads/fin_Eng_002_fb64162b21.jpeg)

## _What are the main approaches to virtual machines detection?_

The classic method to identify virtual machines is based on operating system characteristics analysis. However, this method is not available for a web application. In this case a number of solutions and methods are also available. Nevertheless they also a range of drawbacks, such as Type I and Type II significant errors:

1. Virtual machines detection through significant anomalies. The advantage of this approach is that it has the lowest degree of Type II error:

a. Rendering anomalies;

b. Anomalies of screen and graphic properties;

c. RAM properties and some other functions anomalies;

2\. Virtual machines detection using performance tests:

a. Graphic performance (for example, FPS, rendering graphics characteristics);

b. Operating memory performance;

c. Hard drive performance;

d. Sound card performance and other properties;

3\. Identification of virtual machines through analysis of the installed software integrity. This direction is the most promising and effective for web applications due to the fact that virtual machines significantly differ in content/functionality from real devices.

Naturally, in antifraud and risk management there is no single and universal approach that would solve any problem and give 100 percent result - every expert is in constant search of a balance between data value and a stable approach. Based on our experience, the most effective anti-fraud and risk management solutions should be able to detect risk in real time, have high data value in order to improve decision-making system, as well as analyze user behaviour and identify hidden correlations.

Virtual machine may be a great instrument for dealing with issues related to data protection and safe program delivery, code testing, software performance research. However, the use of such technologies

Virtual-machine-1.jpg

Risks of virtual machines and the best ways to detect them

Along with the opportunities global digitalisation has brought us risks. The emerge of a large number of online businesses inevitably leads to a fraudulent traffic increase. Based on our experience in dozens of countries we can spot more than a thousand patterns of fraudulent behaviour, which, unfortunately, are constantly being improved by scammers. However we also believe that exploring the new methods of risk assessment gives much better result in terms of online fraud fighting than another superficial set of risk markers, which may be used as filters for automatic denial.

Today we are going to share our new methodology benefits, which allow online businesses to identify high-risk and carry out segmentation throughout the entire flow of applications. We are also proud to announce the launch of our new product: automatic risk management reporting, which will be available to all our clients in Q1 2023.

The aggregated results show a comparison of fintech companies in different regions of the world, data are given as of November of this year.

Our methodology is based on the division of the entire session space into 4 groups, which are the following:

![](/uploads/Space_of_the_devices_En_001_2_5c76feaf6f.png)

The key positive factor of this approach is that all the high-risk sessions inevitably fall into one of the four groups, therefore, we can spot 100% of all possible online fraud attempts.

Another advantage is that this approach does not depend on the country, technologies, local legislation etc., which means that one can compare various online resources and countries.

![](/uploads/Fap_Sap_Hra_001_805b519f25.png)

Here you can see the share of FAP, SAP & HRA, which sums up the level of toxic / high-risk sessions by region.

![](/uploads/FAP_Eng_001_0d1f7fffe3.png)

In the First attempt fraud category we included a combination of 50+ rare events with a high probability of fraud related to a device technical manipulation, in particular, the use of randomizers, some browser or operating system anomalies. We also noticed that the number of First attempt fraud in Latin America exceeded the same indicator in other regions almost two times, which may be an illustration of the increase of single fraudsters attacks in this region over the past month.

![](/uploads/SAP_En_001_df3d96d1a5.png)

In Secondary attempt fraud metric we included repeated devices as well as repeated users. We also noticed that the number of Second attempt fraud in South and South-East Asia and in Eastern Europe remains at a relatively high level, which indicates that fraudsters have been using large amount of data from the fraud net recently.

![](/uploads/Hra_Eng_001_1_cce22acaac.png)

As you can see here, the level of high-risk applications in South Asia and Eastern Europe is the highest. Usually such figures occur in the countries with low income of the population and also may indicate the rise of a credit shopping in the region.

![](/uploads/Avg_device_Eng_001_c5bc02eb98.png)

The highest level of average inflow device quality has been spotted in the regions of South-East Asia and Eastern Europe. The lowest rate has been recorded in the countries of South Asia.

In this regard, we recommend online companies to take into account the following actions:

1. Fine-tune the process of regular reporting on key types of risk;
2. Give credence to preparation of strategies and risk policies for dealing with different situations. For example, if your company has a high FAP compared to the average in the region, then you should look for a wide range of stop factors for identifying this type of risk; if you have a low flow value in terms of device quality compared to the region, then it will be useful to reconsider the current customer acquisition strategy;
3. Make sure that the company has all the necessary technologies in order to detect the risk of fraud in real time - hundreds of fraudsters can attack the resource of a financial institution in a short period of time;
4. Check if the data information value is sufficient to improve the quality of models and decision-making systems;
5. Restrict the use of your users' personal data. Please note: often personal data is not required for auxiliary and vendor tasks - it is often possible to find an alternative solution based on anonymized sessions or tokens;
6. Determine for yourself a work option in case of a partial failure of the online infrastructure and a conservative work plan in case of a massive fraud attack (reducing the approval level for applications, introducing additional methods of user verification etc.);
7. Pay special attention to the user's behavior on your online resource in order to identify hidden correlations.

We are planning to create the system of consolidated reporting for all our clients in the nearest future.

Along with the opportunities global digitalisation has brought us risks. The emerge of a large number of online businesses inevitably leads to a fraudulent traffic increase. Based on our experience

JS-1.jpg

Fraud traffic analysis in different regions: short study

For many years we have been working really hard to ensure our products bring maximum benefit in terms of borrowers evaluation and risk assessment. In many financial institutions, micro lending companies as well as the companies, which are not related to financial sector, significant part of business is offline. Estimation of such potential clients is related to a number of difficulties - lack of data, “thin” customer credit profile etc. Until now it was not possible to conduct a risk assessment through the device for such clients. And today we present a new solution we have developed: an offline to online transition tool, which helps to assess the risk of an offline applicant and build their digital profile without significantly affecting the conversion rate.  
  
The purpose of creating JS App - Device Risk Analytics is to provide customer device scoring as a service for offline businesses, including customer assessment supplement via conventional offline sources. The product will also be useful for companies that do not have their own IT infrastructure, as well as for online businesses creating an MVP and would like to save money on launching it.  
  
_What industries_ _JS App - Device Risk Analytics is designed for and in what cases it is applicable?_  
  
Solution may be effectively used in any business with offline offices, branches or stores and where potential client underwriting or risk assessment are required.  
  
In order to get a better understanding in what industries JS App - Device Risk Analytics can be applied, we would like to demonstrate a few use cases, when a person who is rather subtle in financial issues would rather use an offline channel in order to obtain a financial product.  
  
Let’s imagine a situation when a person enters the bank or micro lender’s office, which is located not far from home or office in order to open an account, a card or to get a loan - we can’t but agree that this is rather usual scenario, because many people are comfortable talking to a real person, who can help to solve all the issues at once (such situation also may occur, speaking about certain categories of borrowers, for example, elder people).  
  
Another example refers to a point of sale, where a consumer purchases and pays for goods and services or any offline store, telecom or electronics store. Such people initially do not have the goal of obtaining a financial product, but may decise to buy something if a POS loan, credit card or any BNPL product are available. This also may be a person who wants to buy a car or to get an insurance certificate.  
  
In all the above use cases, service provider collects a potential client's data, does verification and underwriting, using device risk assessment only as a part of verification process. But we believe that device data and a strong digital profile can play more important role and give much more value in a decision making process.   
  
First of all the product will help greatly in risk assessment of clients with the so-called “thin” credit history, for example, young people, retired people, residents of remote settlements, citizens of other countries as well as those who have never used credit services. Depending on the country, the proportion of such citizens can start from 30%.  
  
In many cases device information available for risk assessment through our JS App - Device Risk Analytics solution will have a significant impact both on the approval rate (starting from 10% of relative growth) and expected Gini increase (assuming that in markets with a developed credit bureaus structures, share of credit history effect is about 50-75%, and the lack or absence of digital profile data can strengthen the decision-making model by dozens of percents in relative terms and by 5-15% Gini in absolute terms).  
  
_UX for JS App - Device Risk Analytics user_  
  
In order to start a borrower should download our application either from Apple Store or Google Play - it is available for iOS and Android (soon will be available on Huawei AppGallery).  
  
When the application is downloaded a client is able to become familiar with security and data processing policy, grant access to camera (only when using an app) in order to scan a qr code and granting access to geolocation. At the same time a client can see the logo of the financial company were he is applying for a financial service, since the JS App - Device Risk Analytics is a white label (some sort of partnership when a company can provide a service to its customers on behalf of its brand using its partner company's product).

![](/uploads/2022_11_09_23_12_51_88d164b4da.png)

_JS App - Device Risk Analytics solution consists of:_

- Mobile application for iOS/Android - both in the format of ready-for-download application from the marketplace as well as in SDK format for embedding into the native application of a financial institution;
- A ready-made module for generating a qr-code and viewing the results - the qr-code will be tied to a specific financial institution and a specific client application.

_JS App - Device Risk Analytics solution advantages:_

- Risk assessment significant improvement (especially in cases where there are not enough classical sources - credit bureau data or a profile data). With JS App - Device Risk Analytic_s_ it becomes much easier to build a complete digital profile for each client, and not just for those who come to the online channel;
- The ability to set up communication with customers more accurately by analyzing those who have previously visited the site, linking the entire set of devices to each client profile;
- No need to create your own infrastructure for device evaluation and fingerprinting - JuicyScore provides a turnkey solution + support and regular (monthly) product updates;
- Ease of integration:

 \- Device risk assessment is seamless for the client (the process takes no more than 30 seconds), practically does not affect the conversion and at the same time significantly improves the quality of the decision made on the side of the financial institution;

 \- While most of similar solutions use personal data, our JS App - Device Risk Analytics solution does not request access to personal data, passport or other identity documents.

For many years we have been working really hard to ensure our products bring maximum benefit in terms of borrowers evaluation and risk assessment. In many financial institutions, micro lending

JS-App_pic.jpeg

Device risk analytics for offline business: New JS App advantages

Dear friends and partners!

We are delighted to inform you about one of the most important events of this year: we have released a new set of products, combined into one platform: **JS Platform**. Most of the products are already available worldwide.

We challenged rather difficult task: to find and develop fundamentally new technologies, however this time we managed to outdo ourselves and released new generation of products, providing significant additional effect in comparison to the previous versions. For the last few months we managed to complete over 700 Data science, Risk management and IT tasks in order to release our new product.

The new product generation has a range of significant changes from the perspective of efficiency for our partners. **Here are the main updates, which you will find in JS Platform**:

1. 15+ new parameters and variables in JuicyScore and JuicyID;
2. Antifraud scoring with Gini increase;
3. 45+ new anomalies and randomizers tests, which also let to detect various fraudulent digital instruments;
4. For the first time, a separate group of tests will be allocated for remote access detection on a user’s device;
5. The system of estimated income verification has been improved significantly for mobile SDK users;
6. We elaborated totally new approaches to self-learning models and their efficiency increase;
7. Offline to online solution has been developed. It helps to improve risk assessment significantly without affecting the conversion rate;
8. New device authentication technologies have been developed, including device randomization detection;
9. We created on-premise/local variant of JuicyBio, later on we are going to focus only on JuicyBio on-premise development. This solution is now undergoing the last stages of testing;
10. Service operating speed has been increased by 50%+

Despite the difficult situation all over the world as well as the essential changes in the work of all the financial market players, Juicy Team keeps up high development speed in order to extend the functionality of our products and generate more value to our clients and partners.

We permanently conduct researches to identify new attributes, work on strengthening of our expert valuation in digital risk management to help our clients and partners in their day-to-day work, fraud prevention and risk assessment. Thanks to all of you for your support and sharing your ideas!

Yours JuicyTeam

Dear friends and partners!  We are delighted to inform you about one of the most important events of this year: we have released a new set of products, combined into

JS Platform: 5 in 1. Risk management automation

The challenges of online traffic filtering and fraud protection arose in parallel with the spread of the Internet and have gone beyond the usual solutions for online business. A significant role in fraud protection at the moment is played by device reputation and the anomalies related to it. In present material we would like to shed light on some of the most effective ways of solving these problems and also to present our vision of Device ID usage trends.

The main purpose of calculating or determining Device ID remains device authentication in order to solve various applied tasks. For example, accurately determined Device ID helps to prevent repeated applications from a device with signs of fraud risk. The Device ID is quite an effective tool, at least for the reason that changing the real Device ID is tantamount to buying a new device, which leads to additional financial and operational barriers for fraudsters or toxic users.

It would seem that accurate determination of a Device ID is a fundamental tool in the fight against the risk of fraud and should be a fairly simple solution, but there are a number of serious technical aspects that need to be considered:

1. **Device ID verification accuracy.** This problem is related to the fact that practically every device can be modified in terms of factory settings and authentication details or the parameters, relevant for Device ID authentication;
2. **Online privacy.** This aspect is connected with the fact that for further sustainable development of the biggest tech companies in terms of digitisation users privacy solution is highly required. At the moment we can see that the problem of “passive privacy” has been solved to some extend - in order to do that different measures have been taken, to name a few, MAC-addresses routers removal from public network, third-party cookies, [Google Privacy Initiative](https://blog.google/products/android/introducing-privacy-sandbox-android/?ref=juicyscore.com) ([https://privacysandbox.com/](https://privacysandbox.com/?ref=juicyscore.com)), Apple Privacy Requirements, [FireFox third-party cookies initiative](https://support.mozilla.org/en-US/kb/third-party-cookies-firefox-tracking-protection?ref=juicyscore.com) and some others. It is quite obvious, that digital risk management technologies should be in line with trend of privacy requirements development.

Juicy Team believes that in order to solve the above mentioned problems and face the contemporary challenges online businesses need to develop technologies constantly and also to change the approaches, which existed for the last 15-20 years, since many of them are gradually loosing their efficiency. For the last 6 years (since 2016) we have been working on probabilistic Device ID authentication — it is a 3-rd generation Device ID beyond current privacy regulation rules.

## 1st Generation Web ID/ Device ID.

ID based on virtual user data.

Device fingerprint of the first generation may not be regarded as a device fingerprint in the modern sense. Fundamentally it is based on the characteristics of the device, but to a greater extent on the traces user leaves on the Internet - to name a few the most common, email or mobile phone number. This also may be a reversible hash and in some cases even a clear text.

The main flaws of such fingerprinting were:

- Real user data compromise risk;
- The ease of manipulation - a user can create and operate with an infinite number of different email addresses and a significant number of different phone numbers in various combinations. In addition, forging several fields of personal data is not such a difficult task for an unscrupulous online business client;
- This generation of Device ID was created without taking into account modern privacy requirements.

All these reasons created a precondition for other technologies development on the Internet.

## 2nd **Generation Web ID/ Device ID. Device statistical ID**

Within this generation of IDs, which began to develop actively in the last 10-15 years, we distinguish IDs related to statical device data (MAC address, EMEI, browser hash (browser hash is often called Device ID), persistent sessions and etc.). Let's take a closer look at some of them:

**Browser Hash**

This fingerprint detection technology is based on the analysis of statistical components of browsers such as model, browser and operating system versions, system language, screen resolution, time zone, clock data down to millisecond as well as a list of standard fonts installed on the device. Most of the methods are available for all browser modes, as they are necessary for the browsers normal operation. Among the various directions of this method we distinguish classical device fingerprinting, canvas fingerprinting, webgl fingerprinting, audio fingerprinting and a number of others.

The main problem is instability, as browser hash masks change when manipulating a small number of the parameters above as well as in case of browsers changing.

The second important problem is low accuracy, since the browser hash itself does not allow you to authenticate the device with a high degree of probability.

**Persistent sessions**

One of the first device detection technologies was Evercookie or Persistent cookie technology. Its essence lies in the fact that this type of cookie does not just store information in one data warehouse, such as an http cookie, but uses all available storages of modern browsers - modern HTML 5 standard, Session Storage, Local Storage and others. The ETag header is also used - this is an http header, very short, but you can encode any information in it, and if Java is installed, Java presistence API is used. In addition to this, the PNG Cookies mechanism is used (encoding a piece of information as a small PNG file) and playbacks it through the Canvas API.

Despite all this, persistent-cookies methods practically do not work under an incognito or in private modes of all the contemporary browsers.

**Classic Device Fingerprinting**

Code libraries checks user's browser for all specific and unique settings and data for this browser and the device as well, the data is reassembled into a string and cached by a certain algorithm. The basis of hash key formation is UserAgent. Browser language, time zone (offset from UTC), fonts, color palette, separate platform-dependent constants and other data specific to the user and platform are added to it, sessions of Internet giants can also be used. Information about plugins installed as well as all multimedia types or main types that support this plugin is used in order to increase uniqueness. Canvas Fingerprinting technologies are also used - certain text is drawn on a hidden canvas element with certain effects applied to it. And then the resulting image is serialized into a byte array and converted to base64.

The main flaws of this approach include the frequent changeability of the UserAgent in modern browsers, specific features of hardware implementation of devices by some manufacturers as well as the features of outdated browsers, lack of integration with Flash and Silverlight and some others.

According to a study by the Electronic Frontier Foundation (a part of the Panopticlick project, [https://coveryourtracks.eff.org](https://coveryourtracks.eff.org/?ref=juicyscore.com)), the uniqueness of a fingerprint is about 90-94%. Due to the collection and analysis of a large number of parameters and settings of device hardware and software, it’s possible to ensure the necessary level of entropy and uniqueness of each digital fingerprint. Existing fingerprinting-based device detection technologies have caused a significant breakthrough in terms of incoming traffic assessment, web resource audience assessment etc.

The main problems of classical fingerprinting are related to the following aspects:

1. Fingerprinting relies on a limited set of root parameters, the number of which is decreasing over the time (browsers’, marketplaces and software platform developers policies tightening as well as new regulatory requirements). For example, reducing the readable UserAgent field has resulted in a significant reduction in the effectiveness of a number of fingerprinting technologies. Reducing the set of initial available parameters for device fingerprinting inevitably leads to an increase in collisions.
2. Digital fingerprint calculation technology. Basically, there are two approaches: front-end and back-end fingerprinting. On the front end it can be easily assembled and read. The downside is that the scammer will also be able to read and make changes to the Device ID, which seriously reduces the potential toolkit for building other ways to identify and match devices, while the Device ID calculation technology on the back end is not visible at all.
3. Device fingerprint stability regarding the same real device (especially if we are talking about a trustworthy client) is not sufficient for the effective use of these technologies in industries with higher accuracy requirements.
4. In an extension of the above problem, the opportunity to use such fingerprint in account-centric systems as well as to calculate any attributes in relation to this accounting unit and use such data for decision making may sometimes be limited.

## 3rd **Generation Web ID/ Device ID. Probablistic Device ID**

**JuicyScore approach key points**

Our solution is based on the following approaches that allow us to build a more robust probabilistic device ID: JuicyDeviceID with a uniqueness level of 95-99%+, depending on the required probability setting, service response time tolerance / JuicyDeviceID calculation time and various secondary aspects of international online markets (density and quality of the Internet infrastructure, most common devices of a certain model etc.):

1. A wide range of additional methods;
2. Server-network methods: Juicy\_TCP/IP fingerprinting, AI\_TLS and a number of other technologies;
3. Behavioural patterns and load tests: these approaches are related to the fact that the behavioral use of devices and their performance is unique. Even the devices, which were released on the same day and on the same assembly line, will have different behavioral characteristics;
4. Unique ways of building robust device fingerprints: we constantly analyze various anomalies that lead to loss of stability, discuss their “normality” with industry experts and take these aspects into account during our work;
5. Artificial intelligence: we use our own proprietary artificial intelligence best practices and know-hows in order to determine to which extend two devices are similar to one another, basing on 50k+ parameters and their combinations. These methods make it possible to take into account possible anomalies on the device and provide convergence within the framework of the problem of effective device authentication;
6. From this perspective we have a very fast feedback loop when new fraud patterns appear. We can look at every device that showed signs of fraud risk a month ago, find new characteristics and anomalies and analyze the exact risk of fraud as well as all the details;
7. Architecture uniqueness: Our JuicyDeviceID calculation architecture will retain its uniqueness even with the mass adoption of Web3\. Its main advantage is flexibility, which, in conjunction with the speed of development, shows a remarkable result;
8. Significant increase in user’s privacy and security by means of removing the alternative to synchronize probabilistic device\_ids at the architecture level, avoiding the use of direct user identifiers due to the limited time of use / limited time of stability of the probabilistic device\_id as well as by means of mandatory informing of the users about the collection and evaluation of probabilistic device\_ids .

Also the advantages of our method include security: all device calculations are performed at the server level, not at the level of the user's device. Thus, we reduce the impact of possible manipulations, preserve privacy and reduce the load on user's device. Moreover, it allows us to constantly detect various randomizers / anti-detect plugins, virtual machines as well as maintain privacy, since in this mode we can require online sites to follow the data processing policy, inform users and ensure the operation of the service only in order to prevent fraud and reduce operational risks.

## **What is the main point of stable Device ID value?**

**JuicyDeviceID is an effective way to solve application fraud prevention and detect multi-accounting:**

- Duplicate accounts and multi-accounting risk reduction: when a fraudster uses the same device or multiple devices to create multiple personal accounts. Device ID helps to reduce risky accounts and improve unit-economy significantly.
- Protection of personal accounts of users from unauthorized access by detecting unfamiliar devices or devices with a new Device ID.
- A strong Device ID significantly strengthens the account-centric systems on the online business’ side and allows to embed such ID to the decision-making system (filters, rules, models, reporting). Different variable markers and attributes in relation to such ID have great predictive power, allow to determine the characteristics of the device, Internet connection and user behaviour and therefore to increase the quality and separating power of various models on the online business side;

## **Device ID vs methods of virtualization and randomisation**

A strong and stable ID device is a necessary, but not always sufficient tool for an effective risk management in online business. According to our experience and best practices in various markets, the most dangerous cases are associated with the use of professional device randomization methods or the so-called randomizers or anti-detect tools. Any business needs to have a set of technologies to define randomization, virtualization and remote access.

We have developed 300+ randomization detection technologies and are constantly improving our technological stack.

**Conclusion**

Any online business that works with any assets is in a great need of a stable Device ID. The most common use case for Device ID is to prevent or identify the risk of user fraud. In order to improve business performance many companies prefer to use not one, but a multiple solutions aiming to use their synergy to add more value. Protection across personal data verification, for example, such as data from the credit bureaus, telecom operators, social networks, as well as verification of digital fingerprints of devices and anomaly detection, gives the best result in terms of the synergy of these two concepts, which positively affects the final level of risk in the portfolio and ROI. Thus, companies have to admit that the use of multiple solutions plays a crucial role in growing a successful and sustainable business.

The challenges of online traffic filtering and fraud protection arose in parallel with the spread of the Internet and have gone beyond the usual solutions for online business. A significant role in fraud protection at the moment is played by device reputation and the anomalies related to it. In present material we would like to shed light on some of the most effective ways of solving these problems and also to present our vision of Device ID usage trends.

Gadgets2-1.jpg

The future of Device ID in the digital world

Covid-19 pandemic has fundamentally affected the development of many online and offline businesses and healthcare has become an area of massive interest. [Experts believe](https://timesofindia.indiatimes.com/blogs/voices/the-rise-of-bnpl-services-in-india-and-how-it-is-revolutionising-healthcare-sector/?ref=juicyscore.com) that embedded finance platforms for healthcare in India are much needed given that out of pocket expense is very high and insurance penetration is among the lowest in the world. According to experts, Indian healthcare expense is estimated to be around $120 billion per annum and is growing at the rate of 15 % every year. The government has already shown its support encouraging startups to create leading-edge solutions in order to provide people with affordable high-quality medical services.

Embedded finance platforms in India are expected to be one of the largest contributors to retail payments in the country due to tremendous usage and high frequency. Today the experts from the first healthcare finance solutions in India SaveIN will share their experience of cooperation with [JuicyScore](https://juicyscore.ai/en/about/?ref=juicyscore.com), one of the leading risk management and anti-fraud solutions provider.

_How embedded finance solutions work?_

Embedded finance healthtech solutions allow customers to make avail healthcare products and services at clinics and pay for the same in simple zero/low cost EMIs. Basically, the customer can split the cost of treatment or service , while embedded finance player pays the healthcare partner upfront, net of commission/fee/subvention, on behalf of the customer who is borrowing money for availing such service. The customer, in turn shall return the full amount to the embedded finance company in the prescribed time period

In case of SaveIN , It caters to healthcare needs by offering financing and credit options to patients at zero-cost equated monthly instalments (EMIs) across its network of healthcare providers. The fintech startup is creating a network of healthcare providers to offer embedded finance and pay-later options at merchants and clinics in healthcare business across the country and plays a very important function of developing an innovative model of online lending in healthcare.

_Cooperation with JuicyScore_

The companies launched a pilot project in the first quarter of 2022\. The robust and stable device ID and JuicyScore data vector informativeness has significantly strengthened the company's risk technology stack and has already proven to be effective in identifying a number of aspects related to fraud risk prevention and score models significant improvement.

The scoring model was developed on the basis of issued loan applications, which may serve as an illustration how JuicyScore data can be used to develop new predictors and score models based on data received from the client.

SaveIN intends to use JuicyScore alternate data to better understand and control fraud and credit risk at very early stage of the customer journey. The variables provided by JuicyScore will help SaveIN to filter out high risk applications in the funnel.

During the pilot, device behaviour markers were widely used - an aggregated index from JuicyScore standard output data vector, which includes risk markers, combination of user behaviour and device markers characteristics. For example, a combination of secondary risk markers and device anomalies, where each anomaly individually may indicate a possible risk that should be taken into account when verifying a borrower, and their combination, if triggered simultaneously, which highlights a high-risk area.

In addition, other IDX variables, created with deep machine learning algorithms were widely used. For example, a variable that is a combination of 50+ rare events, which show high probability of fraud through technical manipulation of the device. This variable includes the whole variety of device randomization tools, techniques of interference into "digital fingerprint" as well as determines the most dangerous markers of user high risk behaviour and network connection markers. The variable can be used both in rules and also as a component of a fraud prevention model to identify the most dangerous customer segments. The risk level raises along with the parameter value, high values may be used as filters for automatic denial.

_Why JuicyScore data is so valuable for embedded finance platforms?_

Due to the combination of digital risk management and antifraud solutions, as well as high information value along with fair pricing approach, our clients achieve 10x ROI on the cash flow, adding 5+ Gini points to decision engine score models. Wide data vector based on proprietary technologies of data collection, processing and engineering in combination with embedded external services help our clients to grow much faster than most of the competitors.

**JuicyScore**

JuicyScore company was established in 2016\. It provides technological solutions for fraud prevention in online-business and financial products risks reduction, it also provides customized score, analytics and consult on score modeling and decision making system in 17 countries all over the globe. All the solutions are customized due to the clients need providing analysis of data flow and score. The main feature of JuicyScore is that personal or sensitive data as well as direct user identifiers are not used, which has certain positive impact on the clients' reputation. The company is compliant to EU legislation on personal data (GDPR) all the device authentication technologies and data algorithms are proprietary, the company has 4 patents.

**SaveIN**

SaveIN has a vision of creating India’s largest integrated healthcare ecosystem powering access, quality and affordability in private healthcare.

SaveIN is a healthcare focused financial technology platform that aims to facilitate timely and quality healthcare for Indians with enhanced affordability. The company has created class leading embedded finance solution for healthcare providers and has created a pan-India network of thousands of healthcare providers while growing rapidly since its launch in January 2022.

Globally, SaveIN mirrors the business models of US-based Sunbit, Walnut and Scratchpay.

Covid-19 pandemic has fundamentally affected the development of many online and offline businesses and healthcare has become an area of massive interest.  Experts believe   that embedded finance platforms for healthcare

SAVEIN-1.jpg

JuicyScore helps to reduce risks: SaveIN, India's first embedded finance platform for healthcare use case

It is really important for any online business to be sure that behind every new account there is a real new user of a product. However, it is not always possible to verify and confirm a new user online. On the one hand, this facilitates the customer experience and increases conversion, on the other hand, it allows to create multiple accounts related to one person, which, in turn, leads to disruption of the unit economy and financial losses. Today we are eager to tell our audience about the most effective ways how to deal with multi-accounting issues.

## What is multi-accounting?

Multi-accounting implies a situation when multiple accounts are created by the same user (usually a natural person) on an online business website. Technically, multi-accounting is not considered to be an illegal activity, however, in many cases creation of multiple accounts violates the rules of online business and the terms of web resource usage and also leads to significant costs.

Problems related to multi-accounting are relevant for many industries, let’s take a look at some basic examples.

- **Lending.** In most cases it is impossible to create multiple accounts in lending industry, since all the personal accounts are always connected with a passport data, full name, phone number and, in most cases, the lender identifies a borrower in different ways. In this regard, the problem of multi-accounting lies in the area of connected devices of one group of people (for example, among family or friends), and the main risk is associated with an increased probability of delinquency or default due to a potentially high credit burden, which is artificially “spread” across several accounts.
- **Insurance.** First of all we are talking about auto insurance, when one broker or agent creates a large number of accounts and tries to issue many insurance policies, often using unverified or unreliable data.
- **E-commerce.** With the development of digitization and online payment system, especially during the pandemic, the risk of chargebacks has also increased significantly. According to the latest research of [Juniper Research](https://www.juniperresearch.com/researchstore/fintech-payments/online-payment-fraud-research-report?utm%5Fsource=juniperpr&utm%5Fcampaign=pr1%5Fonlinepaymentfraud%5Ffinancial%5Ffintech%5Fapr21&utm%5Fmedium=email), in 2021 e-commerce industry suffered $20 billion losses due to online fraud. In addition, there is a problem of a large number of fake or non-existing accounts that are considered to be a part of company’s portfolio, however no payments are carried out.\*
- **Betting**. Many bookmakers promise bonuses to newly registered players in order to attract more users. No doubt that in order to create multiple accounts, scammers use different credentials: full name, email, phone numbers, social media accounts, etc. Online scammers use accounts for the purpose of bonus hunting, receiving free bets, as well as to continue betting after the old account is blocked or when the betting limit is reached.

## Multi-accounting mechanics

The ways of creating multiple accounts depend on technological background of online scammers and the tools they use. The easiest way to create a multi-account is to re-register it from the same device. This is especially common when, due to the specifics of the industry or business process, a user can utilize a large set of different email addresses or mobile phone pool numbers (which is more complicated and expensive, but also possible).

![](/uploads/2022_06_08_14_33_57_b498b1931b.png)

As we can see, the main account creation scenario is when different users register new accounts on a unique device and each person logs in to one’s new account.

![](/uploads/2022_06_08_14_36_06_c673105e8c.png)

One of the misuse scenarios is this case is when a personal account is created from one device, while a person logs in from another device. Both actions are performed by the same person.

![](/uploads/2022_06_08_14_36_20_7fdd7e069e.png)

Also, one of the most important issues is the use of randomizers between steps. Was the account registered from a real device or a virtual machine was used in order to create it?

## The use of randomizers

More complex cases arise when a potentially risky user begins to use various technical tricks, device manipulation, randomization tools, deepfakes and device virtualization.

One of the newest and fastest growing types of online fraud is the use of sophisticated software also known as randomizers. The purpose of such software is to thwart existing digital device profiling technologies and bypass the system by misrepresenting the same device a fraudster uses to apply for a loan online as a new one every time.

Thus, a scammer can draw upon the credit from the same device for an unlimited number of times, changing only the borrower's data, while the security system of a financial institution will recognise each application as a unique.

![](/uploads/2022_06_08_14_38_14_899acf48f9.png)

Speaking of _randomization_ we imply the use of any software or code used for adding various anomalies for device fingerprint obfuscation or user activities concealment. A fraudster can randomize a physical device or add some interference to virtual device. One can randomize both a real device and add breakdown to a virtualized or fake device. In addition, in order to hide multi-accounting, scammers often use proxy servers or register each new account under a new IP address. It is also possible to use stolen credentials or create fake IDs based accounts.

## What problems does multi-accounting imply?

The scale of multi-accounting can vary from fairly harmless, when a regular user decides to take advantage of the promotion campaign of their favourite online stores, to bonus hunters, who have rather high technological skills and may cause great damage to the business by their malicious actions.

Practically, the whole strategy of unfair competitors can sometimes be built on multi-accounting: fake reviews or ratings, ads-clicking or fake bad (or good) reviews - all this can be the result of attempts to promote their products and damage the reputation of competitors, for example, in the case of using click farms in order to generate massive Internet traffic with bad faith. In addition, a large number of fake accounts on any web site, classified or social media can lead to a disruption of the unit economy as a whole. So, due to the erroneous assessment of marketing indicators, such as MAU, DAU, ARPU, misperceptions may occur, for example, about the value of the company itself.\*\*

## What are the best ways to solve this problem?

One of the most effective ways to combat multi-accounting is to use solutions based on device fingerprinting technologies, which allow to identify fraudsters quickly using device authentication and matching technologies:

- A set of device authentication technologies allows to create reliable and stable device identifiers over time - depending on how the business processes are built, you can either add a verification step, reject application and at the same time make the UX simpler for those who constantly use the same device;
- Identification of randomization signs, significant manipulations with the technical and software components of the devices. All these signs of risky behaviour distort device fingerprint, however, the very fact of their usage during an account creation or logging in should lead to a proper response of online business;
- Identification of complex proxying signs, network connection manipulation etc. All these signs may not ba an attempt to manipulate the device, but may seriously damage the process of account usage and management.

## Is your company protected from multi-accounting? Short check-list.

- It is worth making sure that the registration of an account, as well as the first login are made from the same device;
- Also a company needs to check that from the same device not more that three accounts are registered. On the one hand, this will help to ensure the customers identity, on the other hand, it will reduce the second type error when accounts for several family members are created from a family computer;
- Also you need to be sure that the same account is not used on more than 1-2-3 different devices in a limited period of time;
- You need to establish additional verification rules in terms of excessive privacy cases on the device - whether any special browser or software settings were used in order to hide device parameters and distort its digital fingerprint;
- A company needs to determine an acceptable level of network anomalies, for example, using the same IP address for different devices for a limited period of time, using foreign IP addresses and foreign network settings, shifting the time zones of the device and IP address. This will reduce cases of accounts misuse, as well as the cases of multiple entries on the same device;
- In addition to that, you should always remember that any company must have the resources to conduct tests for randomizers, conduct tests for virtual machines and private mode.

These are some of the basic guidelines for reducing multi-accounting cases; these rules should be reviewed regularly in order to check and monitor their effectiveness. Also creating and adding new rules to the system will also help. Another approach that works is to build a simple trustworthiness score, possibly binary or semaphore-based, which will allow to manage account creation in a more flexible way and reduce type 2 errors.

Summing it all up, a flexible and secure verification system is required in order to prevent multi-accounting. By doing this, regular users will be satisfied with the company's services or products, while fraudsters and bonus hunters won’t be able to cause any harm to an online business. When a company finds some special tools and technologies to distort the digital fingerprint during account creation, this fact may become a good reason for additional verification. One of JuicyScore's most effective technologies is device comparison and matching: a stack of device authentication technologies allows to build robust and stable in time device ID resistant to various anonymising techniques, to identify randomisation and anonymisation or the same virtual devices. Using these technologies you can spot multiaccounting, data manipulation and multiple accounts. With JuicyScore data you can avoid conversion reduction, improve customer journey and deliver the best user experience.

---

\*Pay Pal case is rather notable. Company’s shares instantly lost a quarter of their value after closing 4.5 million fake accounts. Accounts were created by users in order to receive additional bonuses for new accounts registrations. According to the company's management, the day before they decided to launch that marketing campaign, which turned out to be really unsuccessful, they offered from 5 to 10 dollars to every new user for new account registration. As a result, a large number of bots rushed to the servers of the payment giant, while the company was unable to trace them. Pay Pal had to revise its growth forecasts for this year, as well as reduce the medium-term goal of new audiences outreach.

\*\*In particular, Elon Musk announced that deal with Twitter was temporary on hold due to the revealed facts of multiple fake and spam accounts. Initially, the company stated that the number of fake accounts does not exceeded 5%, however, later it became clear that fake accounts can sum up to 20% of the total number of social network users, which certainly affects the valuation of the entire company.

It is really important for any online business to be sure that behind every new account there is a real new user of a product. However, it is not always

Multiaccounting-2.png

Multi-accounting - a modern-day plague for online business

BNPL market is now growing by leaps and bounds and investors keep showing that they are really interested in it. According to [Grand View Research](https://www.grandviewresearch.com/press-release/global-buy-now-pay-later-market?ref=juicyscore.com), by 2028 market size is expected to reach 20.40 billion USD.

_How BNPL solutions work?_

BNPL (Buy now, pay later) basic scenario is the following: a client purchases goods on the web-site or in the store and pays in installments over the next 2-12 months. A wide range of goods can be purchased this way, ranging from basic consumer goods, such as food, clothes or children’s goods up to brand-name goods and jewellery.

Despite a slight marginality reduction, retailer has the opportunity to push up the sales substantially as well as to improve access to the goods offered. The average check and resale chances also increase immensely.

BNPL solutions are a great alternative to credit cards, which allow customers to make a purchase and pay for it in instalments later. Revo Technologies (the owner of [Mokka](https://mokka.ro/?ref=juicyscore.com) brand) is the largest BNPL company, operating in Russia and Central Europe since 2012\. Today the experts shared their experience of cooperation with [JuicyScore](https://juicyscore.com/en/about/?ref=juicyscore.com), one of the leading risk management and anti-fraud solutions provider.

According to Sergei Bogdanov, Chief Risk Officer at Revo Technologies, “We chose JuicyScore, because it had already showed a great performance on the market and highly experienced team helped us to adapt the data to our processes and product specific features.”

_How easy was the integration?_

Integration was set up in two stages. We put the finishing touches to the widget, where we also included JavaScript from JuicyScore. Via this widget we managed to collect information necessary for JuicyScore service operation right on our iframes, integrated with retailer’s platform, at the moment of making a purchase. For Mokka mobile application JuicyScore SDK integration was set up in order to collect primary device and behavioural data. In web and mobile channels this information is sent to JuicyScore API and we get the response at once - statistics on device and user behaviour.

_What results have been achieved?_

According to Sergei Bogdanov, JuicyScore data helps to achieve significant savings. For instance, in 2021 we managed to save 500 thousand USD. JuicyScore allows to cut off 2-3% of application flow with average risk of 30-40%.

JuicyScore helps to mitigate operational risks really well, in this regard we use stop factors related to frequency characteristics. For example, the number of applications or requests from the same device or virtual user with a certain time periodization, which had been observed previously as well as the variables, verifying the probability of application sending from foreign IP addresses (by verification of the real IP address and/or DNS server location), - it considerably helps to mitigate fraud risk and the number of defaulted debts.

Apart from that, markers of additional verification are widely used. For example, the marker of data manipulation (email login or phone number in reduced format) in comparison to the previous application, sent from the same device. The use of suspicious plugins and specific browser settings installed on the device also refers to additional verification and applied in the model.

Some variables allow to set up a score value for conservative fraud policies or more aggressive ones in terms of risk. Besides, those variables help in such cases when credit bureau data is not enough or data provided is rather low quality.

On addition to the use case provided, we would also like to highlight a few BNPL trends that characterise the market.

_What BNPL-trends are going to affect business in the nearest future?_

Despite that fact that BNPL-platforms currently occupy only 2 percent of lending market, it is quite obvious that BNPL industry will continue to grow, since the need in goods is essential.

In the wake of e-commerce grow (in 2021 market grew by 32% in Russia and exceeded 4 trillion rubles), the need for BNPL solution also grows immensely. One more factor, which affects the growth is the increasing popularity of such services among the young people, who sometimes do not have access to alternative banking solutions due to insufficient credit history.

On the one hand, BNPL market will face rather tough competition, on the other hand, leading companies and key players, which pay great attention to mobile application development and tools for payments tracking designing as well as provide customers with new high-quality solutions for b2c and b2b, are going to win the market in the long run.

Juicy Team continues to maintain a high pace of development and delight our customers and partners with new product functionality, for one can ensure high competitiveness and business safety only through efficient and hard work in a dynamic digital world.

BNPL market is now growing by leaps and bounds and investors keep showing that they are really interested in it. According to Grand View Research  , by 2028 market size is

revo_Eng.jpg

“Revo Technologies” BNPL company success story along with JuicyScore

One of the fundamental principles which guide our working process is a constant search for useful and informative data, which help our clients and partners to achieve better results in credit and operational risks assessment. In fact, revealing and adding new features to our products not only allows to improve decision-making system, but also to ensure transparency and interpretability of results. That is why we pay so much attention to attributes architecture mechanics in our API vector response. Today we are eager to share the details about new user behaviour markers and device quality assessment principles as well as to demonstrate how JuicyScore data can significantly improve credit risk segmentation of the incoming flow.

One of the main approaches implemented in JuicyScore products lies in aggregate IDX variables. Their essential part is a combination of rare events and factors related to them, collected by means of Deep Machine Learning algorithms to one variable, which may be used either for modeling or for integration to financial institutions decision-making system. Today we are going to review two independent variables of our data vector, paying great attention to their mechanics as well as to the factors they are based upon.

**IDX2 - User behaviour markers**

This aggregated variable is a combination of various user behaviour markers on online business web resource. In JuicyScore vector there are dozens of markers which are in some way related to user behaviour and the main issue in terms of variable construction is to find such stable markers, which would allow, aggregated in one, to identify high risk segments regardless of online business operating geography.

Variable IDX2 is based on the factors related to various user behavior or device utilization categories. On the one hand, it incorporates many factors connected with frequency characteristics. For example, the number of applications or requests for financial product obtaining from one user or the same device with a certain time periodization or without it over the entire history. On the other hand, there are also such parameters included into this variable, which identify stability or, on the contrary, data variability used in a credit application. A wide variety of such data on the same device or related to one virtual user indicates a high operational risk. At the same time high frequency of applications without data manipulation may indicate a higher credit risk (the so-called credit shopping, when a borrower applies for multiple credits to many financial institutions within a short period of time). The presence of both - high frequency and high variability of data on the same device or virtual user - is a strong sign of high operational risk.

Moreover, this variable also includes a number of factors regarded as high risk markers of user behaviour, which are not related to those two categories. For example, the way of form application filling, device utilization etc. Basically, we can see the combination of rare events of medium and high risk, which, integrated together in a certain way, may be used in a decision-making system and models, constructed by means of classical Gaussian methods.

As a result we have a variable, which has integer values from 1 to 6 and allows to build credit risk segmentation of the incoming flow and has stable zones to be used in models or stop factors. For example, the interval between 1 and 2 may be regarded as a low risk zone, which is often better in terms of risk level than the average. Within this segment we can further look for additional segments to be approved. Intervals between 5 and 6 are high risk zones. Usually they are comparatively small and may be used as filters or rules for application rejection. You can see relative risk level graph presented below.

![](/uploads/2022_04_11_11_27_31_124d060caf.png)

_As we can see from the graph, variability indicator 1-2 stands for low risk, 3-4 - medium risk, when additional verification/validation should be carried out, indicator 5 stands for high risk level and in such cases we recommend to reject such applications._

![](/uploads/image_6_92b3d3291a.png)

IDX2 values distribution by the region

**IDX5 - Device quality index**

Speaking about the operational risk assessing, the prior task is to reject applications of users with high risk of a loan defaulting and non-payment risk, while the main objective of credit risk assessment is to find those segments of users, which can be provided with financial products with the right and accurate parameters. Aggregate variable IDX5 falls into the second category. Using this variable financial institutions are able to improve credit risk segmentation of the incoming flow significantly. It is particularly important when there is a great lack of strong data institutions or quality of the data provided by them is rather low.

_What approach is used to construct this variable?_

Device quality function is the level of its cost, which is affected by the following data categories: device category (desktop or mobile), its technical metrics aggregate (storage capacity, number of cores, storage quality etc.) and device manufacturer (famous brand of the device or a no-name). It is important to note that devices with certain technical characteristics anomalies are not included into this index in order to convey a higher orthogonality level with the other aggregate variables IDX.

Every device has a big range of technical metrics and parameters, which affect on its quality and may be used also in order to assess credit risk. That is why developing this index it was highly important to identify such metrics and model factors in order to ensure the stability of value range of every factor, keeping in mind that we also need to provide the stability of value range of the index itself and to enhance its sharing ability. Speaking about stability it is also important to note that it should be ensured in time and also among all the geographies of our clients’ businesses.

Within the frames of Device Quality Index value interpretation, the part of the flow with low values of this Index identifies a segment with high credit risk and low level of disposable income. The part of the flow with high values of this Index makes it possible to identify segments with a low level of credit risk.

![](/uploads/2022_04_11_11_28_00_b8d32cc39e.png)

_As we can see from the graph, variability indicator 0-1 stands for high credit risk, when additional verification/validation should be carried out or rejection required. Index indicator above 4 stands for low credit risk level, which means that it’s possible to find applications for additional approvals in this segment._

![](/uploads/image_7_f20a9a7532.png)

IDX5 values distribution by the region

This parameter is especially valuable for countries in South-East Asia and Africa regions in case there is a lack of reliable and regulated data sources.

The definitive and undoubted advantage of using index variables is universal applicability. They can be incorporated into the rules and filters or use separately as a component of operational or credit risk assessment model in order to identify most risky or, on the contrary, low risky segments. Basically, every index is a self-sustained model. The markers reviewed in this article are only a small part of the informative extension that we offer within the frames of our API.

JuicyTeam understands the importance of constant improvement and expanding the set of markers and stop factors which help to reduce costs and significantly reduce losses caused by fraud. We always provide our customers and partners with the best set of tools to cope with fraudulent applications.

---

Other related articles. We covered our approach to variables construction in other materials - [Deep Machine Learning: on the path to the truth](/blog/en/deep-machine-learning-on-the-path-to-the-truth/), where we reviewed some successful examples of methods which can be applied to solve practical problems to prevent fraud risk. Also we provided detailed information about IDX1 - a combination of stop factors and IDX3 - a combination of secondary risk markers and device anomalies, designed according to similar mechanisms, but aimed at slightly different range of issues, in our material [Device analysis true value - JuicyScore secret methodology](/blog/en/device-analysis-true-value-juicyscore-secret-methodology-en/).

One of the fundamental principles which guide our working process is a constant search for useful and informative data, which help our clients and partners to achieve better results in

IDX-1.jpg

Credit risk management tools: JuicyScore indexes review

The year 2021 gone by and speeded up the processes that many companies were trying to implement before and also shed light on some new challenges and opportunities. We have witnessed the wide range of technological breakthroughs, which let many players to optimize costs and implement new solutions. Today JuicyScore experts are eager to tell about the key fintech trends to watch in 2022.

According to Insider Intelligence, during the first 3 quarters of 2021 fintech companies raised over 94 billion dollars, almost the same number as past two years combined. However many experts believe that many technological breakthroughs are yet to come. In the beginning of 2022 we can see that Fintech market growth is far from declining, this year fintech industry may astonish us all by new achievements.

![](/uploads/2022_02_16_15_06_30_72f045570a.png)

**Web 3.0 and financial services**

Web 3.0 has attracted the attention of experts in recent years. Users are not only interacting with apps, but apps and devices also interact with each other, creating a more open and intelligent Internet. In addition, data in Web 3.0 is much more interconnected.

However, talking about Web 3.0 in the context of financial services, most people are referring to the next generation of the Internet, which will be decentralized and will run on a distributed infrastructure built on open standards (including blockchain), while the current version of the Internet relies heavily on various mediators, who also usually control the flow of information. Thanks to the extremely wide opportunities given by blockchain programming, with the use of smart contracts that allow to implement rather complex logic, even the most elaborate financial transactions and e-commerce can be implemented by means of Web 3.0.

In 2022 experts expect traditional financial institutions, financial service providers, and even some governments to test Web 3.0 technologies to improve and simplify financial service rendering. However they also expect that the regulatory framework will continue to evolve and adapt to circumstances where there are no mediators with registration or license obligations.

Development of new generation banking platforms may be reviewed as an example. In particular, a platform which provides compliance with regulations, security and issues instant loans. Also trading platforms based on blockchain technology are also being developed currently, which facilitate making payments abroad. Another example is distributed digital global ledgers through which financial institutions could provide every user with one’s own records and a user could have secure access to this information online.

**Changes in workflow and labour management**

The remote work format is already familiar to many online businesses. The consequences of the pandemic no longer seem so irrecoverable - the most successful companies have adapted to new conditions and are gradually developing new ways to organize workflow processes in order to make remote or hybrid work format more efficient.

Apart from that online companies are interested in changing the approaches to product development in order to achieve better results. In particular, ValOps/DevOps are replacing Agile, since DevOps and ValOps-commands reduce the product development time. At the same time, the high speed of development is also followed by sustainability and quality, so the companies’ management now rely not only on speed, but also on process automation.

Also, many online companies are coming to the conclusion that a reasonable approach to work-life balance, in particular, reducing the workweek from 5 to 4 days, maintaining the same scope of work, also motivates employees and attracts prospective ones.

Another trend is related to staff turnover - this lets to accomplish maximum efficiency and at the same time encourages workers to leverage their skills in order to achieve vertical growth. Also one of the consequences of this trend is a gradual increase in the pay gap between the "best" and "worst" employees.

**The rise of global cloud and private networks**

According to Statista, in 2021 50% of the volume of all the essential corporate data was stored in a cloud services. The amount of such services as well as the data volume are growing immensely because of online companies, which therefore expect to boost business agility. At the same time the pandemic has become some kind of a trigger for the trend increasing, for such services play a significant role in remote work organization. Many companies consider cloud services to be key factor of business development.

However it should be noted that the development of global private networks in the first place is related to the idea of secure and sustainable internet. Online companies are trying to rise the security and reliability of data storage.

Notably, speaking of the development of private networks the same trend is likely to happen at the national level in various countries. The global cloud private networks market size [was estimated](https://www.globenewswire.com/en/news-release/2021/12/21/2356098/28124/en/Global-Cloud-Virtual-Private-Network-Market-Research-Report-2021-to-2026-by-Component-Connectivity-Type-Organization-Size-End-user-and-Region.html?ref=juicyscore.com) at $18.08 billion in 2020 and is expected to reach $75.64 billion by 2026 with a CAGR of 26.9%.

**Further development of AI and ML technologies and regulation rules**

Information is highly valuable resource in the 21th century. Its volume is growing every year and various methods are used to process these data arrays. Being a subtype of machine learning, DML’s (Deep Machine Learning) main feature is the use of machine learning methods and neural networks to solve real problems similar to human ones. DML searches for deep intermediate relationships between factors. Each element of the found dependence should be checked for stability and could be used to solve the problem of the next level - a particular hierarchy of attributes that were obtained by one or another statistical algorithm is built in the system, and each new layer has data about the previous one. From a practical point of view, to solve the top-level problem, an ensemble of models is used, each of which solves one of the problems below in the hierarchy. Related business tasks may include fraud / spam detection; speech / handwriting recognition; translation and imitation of many other human cognitive functions.

Artificial intelligence (AI-based) tools, which are actively used by fintech today, have already had a huge impact not only on the financial sector, but also on the whole world, they solve the problem of financial inclusion that has become very acute in recent years.

It does not take a Crystal ball to forecast that AI and DML technologies will continue their development. Growing trend for data collection and access regulation (both personal and user data) will inevitably affect data processing and data analysis methods which subsequently result into AI regulation. Why is it inevitable? Many algorithms both existing and under development require data collection for modelling and model renewal. In case this data has been collected without needed user permission it may impose risk up to full data usage restriction. That means that data processing algorithms and models based on such data will be also restricted. One of the ways to balance such risk may be found in preliminary data depersonalisation, generalisation or “roughing” before this data will be used for modelling and algorithms design. This approach will allow to comply with data owner rights even if it cause models worsening at the first stage. A spectacular example is endless discussions around Google Topics/ FLoC API.

Many experts say that in 2022 there will be a pressing need in regulating this area. Indeed, many companies continue to invest in AI and develop new methods without a solid legal foundation, large number of big companies have so far relied on the principles of self-regulation, which supposed to be more or less in line with the latest practices. Lack of AI regulation is not necessary a bad thing, as it gives companies more freedom to explore and test new AI features without constant pressure from regulator entities. However, currently there are no existing precedents as well as established approaches to legal regulation of this situation. The major frameworks will be adopted in the US and Europe in the second half of this year.

**Quantum encryption and quantum computing**

Assumptions that quantum technologies are coming around the corner seemed to many experts rather bald, however within the last 5-7 years everything changed completely. The basis of quantum technology is quantum particle (qubit), which can be in superposition or, in other words, maintain multiple possible states simultaneously, opposed to bits, which may store only a 1 or 0\. Thus, quantum computing let us to make not sequential computing, but parallel, processing huge amount of data at the same time. Earlier experts believed that quantum computing can help us to solve such problems, which we can not define yet, however some projects already do exist - for example, giants like Google or Toshiba have already showed to the world the first prototypes of quantum computers. China, Europe and Russia also support similar projects.

_What problems can be solved using quantum computing?_

First of all quantum computing may be used in financial sector in order to solve optimization problems or quantum communication maintenance. The training of classical neural networks and artificial intelligence also needs optimization, in its turn, Machine learning complements this development.

In view of this trend, there is also a question related to information security methods. Most of the cyber defense infrastructures which are currently used, namely public-key cryptography, can soon be replaced with quantum cryptography, which will be based on quantum entanglement (if some bit of information has been seen by anyone who was not supposed to see it, the properties of such bit will change), or by post-quantum cryptography, which is essentially a software based on a new generation of mathematical quantum algorithms.

In the nearest future as a development of this trend companies, working under such problem, will increase the number of qubits, improving the effectiveness at the same time. Also it should be noted that cloud access platform for quantum computers should be consonant with the other platforms, so we are going to witness synchronization of development processes in different countries.

**Embedded finance and Open Banking**

It's no surprise that many experts forecast the continued success of embedded finance in 2022\. The rising demand for it has been a long time coming, so many financial institutions are increasingly offering banking as a service (BaaS). In order to make it real, online companies need new technologies and SDKs being available due to the fact that usually such services are distributed via opened APIs, also this trend would require strong antifraud and risk management measures to be taken.

Many experts say that a few previous years we have witnessed a lot of innovations in terms of fintech digital transformation. Indeed, remote work, new digital platforms and user experience somewhat confused a lot of users and led to certain limitation and problems. However it seems that 2022 is going to be more stable in terms of normal business activities. Innovation for innovation's sake no longer makes so much sense, a certain balance has been established and so we are going to see many changes in fintech in the nearest future.

Moreover, BNPL is also a part of such trend. Practically it is an alternative to credit cards, which lets the client to pay for purchases dividing payments. This allows trading platforms to become more accessible and providers can earn a percent from such payment or installment.

**Money+technologies equal to a new start**

A year gone by and speeded up the processes that many companies were trying to implement before and also shed light on some new problems.

Speaking of BNPL, not only the simplicity and speed of purchase are important for the end user, but also its safety. However, one of the main vulnerabilities of many BNPL companies is that they do not conduct official credit checks, using internal algorithms based on the information they have instead. The most effective way to solve this problem is device risk assessment and analysis of non-personal data, as well as the assessment of Internet connection and user behavioral data.

Apart from that, continuing the trend of non-personal data as a part of loan pipeline for neobanks and alternative lending, we offer the JuicyID product, which allows companies to strengthen the external perimeter of the client's personal account on the web portal or mobile application of a financial institution, identify a large set of dangerous device anomalies, highlight related devices of the same user, and, as a result, reduce losses. In addition, JuicyID can be effectively used as part of the protection of the personal account of creditors - the JuicyID data vector allows to build rules and models to protect financial and quasi-financial information, flexibly configure access authorization rules depending on the risk level of the device or user.

Machine learning methods are also a powerful tool that allows, with an understanding of the nature of the data and the correct application of modeling methods, to achieve practical solutions, such as predicting various kinds of risks. For example, JuicyScore is one of the most effective ways to increase the information value. We believe that it is necessary to strike a balance between secure, authorized access to data and the presence of a regulatory and methodological basis for the possibility of effective work of data consumers. When applying the regulation of data processing operations, a differentiated approach should be implemented, which would correlate with the level of risk for individual categories of data and the potential damage that may arise from malicious users.

Juicy Team will continue to maintain a high pace of development and delight our customers and partners with new product functionality, for you can ensure high competitiveness and business safety only through efficient and hard work in a dynamic digital world.

The year 2021 gone by and speeded up the processes that many companies were trying to implement before and also shed light on some new challenges and opportunities. We have

fintech-1.png

Fintech 2022

Online companies are constantly searching for the new methods of customer acquisition as well as new ways of services provision on the financial market. However, speaking about online business in general and online lending in particular, it is more important to find the balance between business volume which you can acquire via particular channel as well as its quality and the risks related to it.

Taking into consideration that the marketing expenses is one of the most significant shares of any online business budget, this also implies that such expenses are the most sensible for P&L. So we would like to tell you about the various client acquisition approaches and risks, related to them.

In general all the fundamental ways of traffic attraction can be divided into two large groups. The firs group is related to independent acquisition (the so-called organic traffic) and the second group consists of the partners' resources usage (partner or affiliate marketing).

- Organic traffic is one of the most common and easily implemented methods, when a company uses various marketing instruments in order to attract traffic to its web sites, lending pages as well as mobile applications, for example, using seo tools, search advertising etc. Potential client directly links to company's web site and gets the access to either a product or a service. The advantages of such approach is transparency as well as direct access to the potential audience and the opportunity to implant all the tools available for analysis and decision making. The main difficulties are rather high cost and high standarts set on team's technological expertise - a company constantly needs to work on the targeting and advertising mechanisms improvement, test various scenarios, copywriting and ad creative. Basically we can say that such way should be described as "business into business" model with it's own P&L and KPIs.
- Partner or affiliate marketing is a way to acquire new customers by means of partner's resources, who get access to the traffic using a model when an affiliate earns a commission for marketing company's product, for example, issued loan. The main advantage of such model is gaining rather quick and relatively budget-friendly access to the large volume of targeted traffic in comparison with the various tools used for organic traffic growth on the web site of a company, also scaling and automatization become much more easier. Besides there is no need in far-reaching expertise from online business behalf, in general, all the acquisition comes down to the necessary APIs connection of those lead generators or brokers, which are going to provide a company with the relevant traffic.

In this article we are going to consider in details all the advantages and risks of such customer acquisition approach.

**What approaches can be pointed out within the frames of affiliate marketing?**

- Redirect - when a broker, lead generator or online seller within the frames of loan selection services or product/service realization redirects the client to financial institute's online platform (for example, via iFrame). Often it happens seamlessly for a client and financial institute acquires the direct access to potential client and can use all the tools for underwriting, risk assessment and verification.
- Closed API - is this case the source of an external partner is used also, for example, aggregators, financial platforms or loan selection services, but, contrary to the previous redirect approach, the mechanism of work with financial institutes is different. A client fills in a form or an application on a broker's or aggregator's platform and this application is sent to potential creditors via closed API (back-end systems of a broker or an aggregator is integrated directly to back-end systems of financial institutions), the number of potential creditors can be significant. Every creditor gets rather limited data set on the applicant.

This method is widely used in many countries. For example, it is rather illustrative example in EU countries, where using such method a company can generate from 50 to 90 % of traffic (before the pandemic this number was closer to the upper limit, however during the pandemic and due to all the restrictions related to it, in the midst of economic hard times, this rate was slightly declining but after that increased back again).

**What are the main risks of this approach?**

1. Applications flow quality and risk level - in comparison to the "organic" method, risk level in this situation is much higher and it also affects greatly portfolio P&L in general and also within the terms of attracting of new clients. Creditors often find it difficult to assess the loan applications due to the limitations implied by a narrow personal data vector fields on a volume of data transferred. Also any alternative ways of data assessment are usually out of reach. This situation leads to a higher risk level on such traffic or lower conversion level, or, sometimes, both.
2. Commercial fraud is the most widespread and complicated problem, as within the frames of detection, but also within the frames of financial effect and, at the same time, due to great amount of typologies and patterns, the share of such fraud is increasing. In commercial fraud several patterns can be distinguished.
- Repeat leads

In a model, when an online company pays commission for an application, it can get a lead which has already been sent to the other companies with a higher rate and also, most likely, with higher expected risk level (since this application has not been approved several times, obviously, there were some reasons for that).

In a model when an online company pays for a loan issue, a broker or an aggregator get the commission as a percent of the sum of a loan or a credit, however the situation has not changed much either, - the higher application flow risk leads to a reduced conversion, increased operational costs, losses and decreased LTV, as selling of repeating loans and other cross sales usually get complicated in this case.

- Early repayment

Another risk is related to the situations when a loan is repaid in advance, for example, the very next day it was issued. From the perspective of credit risk such borrower is clearly reliable, mostly because the creditor does not sustain losses, but speaking about P&L such situation is not preferable for any financial institute, since the creditor bears expenses for lead acquisition, financial product issue and won't get any profit currently or in the future, as such loan or a credit can not be used for cross sales or repeat loans. If a share of such applications in broker's or lead generator's channel is too high, there is a chance that such applications are generated artificially in order to gain the commission for clients acquisition and its reselling to some other creditor.

- Multiple leads sent to an unlimited number of users

It seems that such model must increase the conversion by means of larger offer, however, in fact, it can lead only to customer frustration and dissatisfaction, since there is a false evidence of credit availability for high risk clients, however their application will most probably be rejected, or it may also imply a lot of unnecessary communication for low risk clients, while they only wanted to get the best proposal rather than interact with different financial institutions.

**Advertising fraud - main types and trends**

Now we would like to discuss the most common types of advertising fraud aimed at organic traffic manipulation;ations as well as related to affiliate marketing. Also we are gong to talk about the ways to detect and prevent them. Many of this types are connected with the artificial generation of unproductive traffic, which actually is no surprise, because, [according to the latest researches](https://nymag.com/intelligencer/2018/12/how-much-of-the-internet-is-fake.html?ref=juicyscore.com), less that 60% of all the internet traffic is human. Depending on a budget for internet promotion of a company, ad fraud depends on the campaigns results and outcomes and also on ROI.

![--------------2021-12-20---11.52.54](/blog/content/images/2021/12/--------------2021-12-20---11.52.54.png#img-float-left) One of the most primitive strategies of unfair competition, which usually prevents the company from estimating the real volume of organic traffic, coming to company's website, - is related to context advertising invalid clicks. Competing company employees or some special stuff which were hired by a rival competitor, click on a context advertising link, video, advertising in social media or browser search result. Sometimes those who click the links use anonymous VPN in order to bypass the filters of advertising systems. Of course, such source of traffic does not lead to income in any way, a lot of companies even refuse to use such promotions channels and come to a conclusion that it just does not work. Click bots are also often used for it - specific programs or scripts configured to click on a context advertisements. Speaking of the percentage ratio, the number of such clicks may be from 20 to 60%, which has a significant impact on ROI - every click either does harm to a company's budget, or takes away the potential client.

![--------------2021-12-20---11.53.28](/blog/content/images/2021/12/--------------2021-12-20---11.53.28.png#img-float-left) SDK-spoofing or preset is relatively new type of mobile fraud, which implies the situation as if mobile application is been downloaded by means of malicious code in SDK-file.

![--------------2021-12-20---11.53.58](/blog/content/images/2021/12/--------------2021-12-20---11.53.58.png#img-float-left) As for the main types of fraud related to affiliate marketing, click injection slightly resembles click-bots fraud, however during the installation performed by an application user to Android operating system, malicious program installed on a device initiates a click on an application attributing this action to it's advertising account. Thus, for every such click a company pays commission to an affiliate, which practically did not anything in order to attract a client.

![--------------2021-12-20---11.54.30](/blog/content/images/2021/12/--------------2021-12-20---11.54.30.png#img-float-left) Cookie stuffing is also one of the most popular types of fraud related to affiliate marketing. Using the cookie files stuffing in a browser, a fraudster can easily add up to several dozens of partners resources undistinguishable. By clicking the link a user practically does a targeted action and a company Pais the commission for a lead which was attracted independently from any partners. Such fraud scheme also leads to the situation when a traffic has initially no chance to be converted into a sale of a product or service.

One of the main evidences of this type of fraud is a sudden increase of partner marketing costs without the relevant or noticeable ROI.

![--------------2021-12-20---11.55.05](/blog/content/images/2021/12/--------------2021-12-20---11.55.05.png#img-float-left) Ad stacking - rather popular type of mobile fraud which can usually be found in payment-for-clicks advertisement. A few advertisement are stacked one upon another and an end user only can see the last one. So an advertiser pays for the clicks and impressions, though a user does not see his advertisement.

![--------------2021-12-20---11.55.47](/blog/content/images/2021/12/--------------2021-12-20---11.55.47.png#img-float-left) Pixel stuffing is advertising sized 1X1 pixel. A user obviously can not see it and, at the same time, a company has to pay for advertisement shows and impressions.

**Which channels are exposed to ad fraud most?**

In the first six months of 2021 because of fraud related to mobile applications a lot on online companies lost about 1,6 billion dollars. Most part of it accuses to APAC. At the same time financial sector is the most vulnerable, it accounts for more that 40% of total losses from fraud risk. Bots are the most common threat within the frames of application downloading globally.

According to analytical report of Advertising Fraud statistics, by 2022 the total damage from thee most common types of fraud in online business mobile channels is going to reach 65 billion USD.

![](/uploads/2021_12_15_16_22_12_e3fbaa9370.png)

https://www.businessofapps.com/ads/ad-fraud/research/ad-fraud-statistics/#1

To sum up all the above mentioned, affiliate marketing model has an upward tend and prevails in many markets all over the globe, since it lets to boost the volumes of traffic steam. On the other hand, such approach leads to the maximum risk level as regarding the fraud-patterns variety and the difficulty of detection and prevention of those patterns, but, which is more important, regarding the level of negative impact on financial institute P&L breakdown, and the excessive enthusiasm in using closed APIs makes financial institute even more vulnerable. Moreover, the risks are not unique for one particular country, and in some proportions exist in many regions, including Russian Federation.

**What methods of risk reduction do exist?**

The core and basically the only way to decrease such risk is to increase the transparency dealing with aggregators, brokers and financial platforms. In real case scenario it may be achieved by the increase of initial flow accessibility for the alternative instruments of risk assessment applications, received from lead generators. It may be done in different ways - for example, via providing the access for the alternative risk assessment technologies directly to the brokers or aggregators platforms, in credit organizations' widgets - either on a web site or SDK in mobile apps (a method, which is constantly used in installment credit selling) with the following transfer of a safe token to the financial institution's side, which, by means of such token, gets the access to alternative risk assessment instruments online, for example, with JuicyScore solution.

In such case JuicyScore lets to achieve a number of goals:

- Device authentication when a user enters aggregator's web site, which means, in particular moment when an intention of obtaining financial product is being converted to an action;
- User behaviour patterns identification, including various anomalies, risk-markers, stop-markers related to the device and internet connection in the very moment of filling the application form for credit product obtaining;
- The increase of information value due to the data different from traditional sources. In a wide range of markets such solutions may be unique and more effective in terms of decision making ability.

JuicyScore lets to filter out applications with the evidence of high fraud risks as well as commercial fraud risk, the number of which has increased recently.

Apart from JuicyScore solution we also propose our clients to use Juicy ID, which has the same device authentication technological stack as JuicyScore, but also allows to strengthen the exterior client's personal account perimeter on a web portal or mobile application of financial institute, to verify the great number of dangerous anomalies on the device, to highlight all the connected devices of the same user, and, as a result, to decrease the number of losses connecter with repeat loans.

The combination of various alternative risk assessment methods will allow to preserve all the advantages and technological effectiveness of integration via API, and also will increase the transparency and effectiveness of this channel, this also lets to decrease risk level and increase the conversion rate, as well as to provide the additional information value in those segments, where the level of loan rejections is related to the lack of sufficient data volume in order to access the application.

Juicyscore experts believe that all the online-based companies need to adjust their technologies of fraud prevention and protection in accordance with the constantly developing online fraud technics as well as to prevent new types of fraud. The access provision of alternative assessment instruments on online companies' resources will give those brokers and aggregators competitive advantage and also will create a high level of confidence to financial institutes by means of the transparency increasing in all affiliate marketing in general and closed APIs in particular, and, as a result, will lead to better financial results for all the perticipant of this process - the increase of convertion into sales, risk level decrease and customer loyalty increase.

Online companies are constantly searching for the new methods of customer acquisition as well as new ways of services provision on the financial market. However, speaking about online business in

News_Nov.png

Risks of commercial fraud and closed APIs

Information value

With the development of online services and the emerge of numerous financial products on the market, the number of companies ready to render these services also increases dramatically every year. Industry leaders give high priority to product parameters and characteristics, customer experience, clients' base expanding and keeping the loyal customers. The reverse side of this process is increasing complexity of clients risk assessment via online channels as well as rather large proportion of high risk users, who sometimes may come to the website of a financial institution with malevolent intentions. Thus, companies have to divert a significant share of resources (data collection, technology, staff) to reduce risks and to detect fraudulent users.

Along with the development of the companies doing business online, a large number of different vendor companies have entered the market. They provide solutions designed for reduction of the risk of fraud. Many of these solutions provide a set of risk markers that can be used to block some types of applications in application flow. However, this approach has several flaws.

JuicyScore team believes that apart from providing the markers themselves, it is also necessary to pay great attention to the information value and methodology of data usage regarding the risk assessment process. We have developed a number of methodologies and approaches that allow us not only to identify high-risk segments, but also to perform segmentation and risk assessment throughout the entire application flow. Today we will provide you detailed information about one of these approaches.

**The essence of JuicyScore fraud prevention approach basing on devices detailed analysis**

In our case, we examine approach to fraud prevention based on devices characteristics.

This methodology is based on a great focus on the device - accurate and precise authentication, identifying the various parameters that characterize the device, its environment and how it is used. But why do we pay so much attention to the device itself?

1. The most obvious reason is related to the fact that often such assessment may be enough for making a decision about financial product application. While there is no need to store or process users' personal data (natural persons direct identifiers, such as full card numbers, photos, emails, passport data etc.) Despite of huge investment of the companies to the IT-infrastructure as well as organizational measures, cases when direct identifiers, contact information and sensitive data fall into fraudsters' hands are, unfortunately, quite common. This may lead to the loss of information, cause damage to reputation and result in financial losses to all parties. The second advantage of this methodology is predictive power - device analytics allows to assess the risk of all Internet connections with a level of information availability of 99.9% + and identify the virtual user intentions in most cases at the first attempt.
2. The core of our approach within this methodology is careful and accurate device authentication taking into account device randomization and virtualization. This methodology is based on the developed set of instruments/ technologies for device metrics - a stable, probabilistic approach to device authentication.
3. A support element of this technology is a set of instruments which allows to identify different variations in the randomization and virtualization of devices (as aspects of distortion in the process of device authentication). Virtualization issues have recently been in the spotlight of many experts' attention and today we would like to give you detailed information about JuicyScore approach to assess risk on device virtualization and randomization as well as discuss some aspects that online business owners need to pay attention to.

**Which device categories we distinguish?**

If we represent the entire probability space as a devices projection, we can distinguish the following categories, which cover 4 types of devices. There are physical devices, virtual devices (virtual machines), physical devices with randomization and virtual devices with randomization. Let's dive deeper into these categories that you can see on the picture below.

![](/uploads/image_3_d65369b838.png)

1. _Physical device_ in our paradigm is used when a person applies for obtaining a financial product or user's personal account login. Most common laptops, smartphones with more or less typical software without any serious anomalies in software or program components may be regarded as such type of device. Within the category of physical devices we also distinguish new and repeated devices - the risk level of repeated ones may, in some cases, differ significantly from the level of risk of the new ones.  
    
Repeated devices are distinctively marked devices, for example, when the device was used in order to apply for a loan and some indicators of fraud were identified or payday loan has been in delinquency for more than 90 days. In such case it may be a second attempt to apply for a loan.
2. _Virtual device_ is a virtual machine deployed on a physical device, that is used for nontypical activities of a user with unknown intentions. Usually such virtual devices may be either not designed for making operations connected with financial activities (for example, to apply for a loan or a credit), or may relate to grey legal area.  
    
What is a virtual device or machine? It is any type of device (PC, tablet PC, smartphone etc.) created with special software or program code. In fact, such device doesn't differ from any physical computer/laptop/smartphone or even a server. It also has a processor unit, memory module, data and file storages, and it also can connect to the Internet if needed. However while real computers have physical storage system, memory modules and microprocessor chipsets, virtual machines or software-defined computers exist only as a code. Virtual machine may be a great instrument for dealing with issues related to data protection and safe program delivery, code testing, software performance research.  
    
Virtual device make company's IT-infrastructure operations much easier and also increases productivity due to resource optimization. The use of such technologies in financial products and services obtaining may indicate user's malicious intentions and, therefore, may lead to high risk for the business.
3. Speaking of _randomization_ we imply the use of any software or code used for adding various anomalies for device fingerprint obfuscation or user activities concealment. One can randomize a physical device or add some interference to virtual device (in terms of our categories description).
4. _Randomized virtual devices_ is the most sophisticated type of fraud, which requires high technology skills. In this study we don't consider this type of devices in detail, as well as the aspect of network connection randomization. From the point of view of practical fraud prevention, it is more important to identify the randomization technologies on the device, which, in most cases, is enough for making a decision. Using randomizers in order to simulate the operation of network connection as well as code testing is absolutely normal, however using this software to apply for financial products and visiting the websites of online lenders is hard to consider as normal human activities.

The presented approach allows to cover the entire device probability space completely and also very thoroughly specify segment and assess risk. As we can see from the picture above, physical devices have the lowest relative risk of fraud, while repeated devices among physical ones may have a higher risk. Such devices can be assessed in terms of credit risks and, in some cases, with facilitated verification. Virtual devices are highly risky and, therefore, additional verification/validation should be carried out. Devices with randomization are the most dangerous in terms of risk and we recommend to deny such applications.

The appeared simplicity on the surface will perhaps be less so, if we dive deeper into sophisticated technologies. First of all, there is not a single device beyond the above mentioned segments. Therefore, this approach lets us to capture the whole picture. Secondly, this approach may be characterized as really sustainable and high effective, validated by our work experience in 20+ countries all over the globe.

**How our technologies help to assess risks?**

Index variables is an essential part of JuicyScore data vector attributes. We payed great attention to them in one of our previous articles - [Deep Machine learning: on the path to the truth](/blog/en/deep-machine-learning-on-the-path-to-the-truth/). But how exactly we can assess risk using JuicyScore data? In order to understand technological essence of our approach better, we need to look at a few variables (or variables of the IDX type in our standard data vector), which we create using the algorithms of deep machine learning.

IDX1 is a combination of 50+ rare events, which show high probability of fraud through technical manipulation of the device. This variable includes the whole variety of device randomization tools, techniques of interference into "digital fingerprint" as well as determines the most dangerous markers of user high risk behaviour and network connection markers. The variable can be used both in rules and also as a component of a fraud prevention model to identify the most dangerous customer segments. The risk level raises along with the parameter value, high values may be used as filters for automatic denial.

![](/uploads/image_4_4bac5c3fcc.png)

As we can see from the graph, variability indicator 0 means low risk, 1 - middle risk, when additional verification/validation should be carried out, indicator 2+ means high risk level and in such cases we recommend to deny such applications.

Apart from the use of randomizers, which are identified with IDX1 as well as with the separate stop factors - for example, copying of another device's session (vector's variable session clone), identification of anomalies in the header of the web session (vector's variable UserAgent Issue), indication of manipulation with the color palette (variable Canvas blocker) - also great attention should be paid to browser or operating system anomalies.

IDX3 is a combination of secondary risk markers and device anomalies, where every single anomaly may reflect possible risk, which should be taken into account during the borrower's verification. The combination of such markers highlights a high risk segment. Similarly to IDX1, the risk level is raising along with the variable value, and high values may be used for automatic denial.

![](/uploads/image_5_410d5ca644.png)

As we can see from the graph, variability indicator 0 means low risk, 1 - middle risk, when additional verification/validation should be carried out, indicator 2+ means high risk level and in such cases we recommend to deny such applications.

Cross-tabulation of various indexes values may also provide some important information on risk level. For example, if IDX1 and IDX 3 are equal to 0, thus there is a high probability that no randomisation and virtualization are identified, and we are dealing with a physical device.

This is just a small illustration of the practical approach toы randomization and virtualization detection technologies and their use in risk assessment and fraud prevention.

Currently the most informative and successful antifraud solutions have to meet the requirements commonly accepted by the industry:

1. Real time fraud risk detection: hundreds of fraudsters can attack a financial institution's resource in a short period of time;
2. High data information value - in order to improve the quality of decision making systems;
3. Analysis of user behaviour and verification of hidden correlations.

However, as we all know, in antifraud and risk management there is no single and universal approach that would solve any problem and give 100 percent result. JuicyScore team believes that every new self-sustainable approach will find it's place.

With the development of online services and the emerge of numerous financial products on the market, the number of companies ready to render these services also increases dramatically every year.

News_oct_pic-1.png

Device analysis true value - JuicyScore secret methodology

Company

Dear Friends!

JuicyConf 2021 was held 8-10 September, it was devoted to the new trends and risks in digital risk management. All the guests who attended the event — risk managers from Banks and Microfinance organizations — presented reports and discussed the topics related to the most relevant and burning problems regarding to fraud prevention market, current and expected trends of online business and the best ways of preventing the new types of online fraud. Today we would like to tell you about the most relevant topics that we managed to discuss, share with you the key results, some interesting conclusions and photos of our event.

We are really proud to announce that, according to the results of a survey we conducted right after JuicyConf 2021, all the respondents pointed out that the topics discussed during the conference were extremely useful and rather relevant. Companies' case studies, virtual machines and device randomization, a demonstration of fake digital fingerprint creating turned out to be the most interesting. A lot of the attendees noted that it was rather useful to communicate with the colleagues in an informal setting within the framework of the further business development.

![](/uploads/Pavel_MNLT_1_7665f2b9e3.png)

Pavel Bolshakov, JuicyScore CEO and Founder initiated the discussion under agenda about randomizers issues. In recent years many financial institutions have witnessed the rapid growth of online fraud along with the damage caused by it. The increase in risks occurs due to several factors:

- manipulations with a digital fingerprint;
- the rate of darknet expansion, including fraudsters' technological skills - the process of randomizers creating has been simplified greatly.

Many fraudsters know how to bypass banking authentication systems, so there is a growing need for new fraud prevention solutions that would allow business owners to fight online fraud rather than its consequences. However, the emergence of new risk management technologies inevitably leads to the new fraudulent technologies and randomizers development.

In fact, one of the main problems is that online business losses caused by fraud amount to a significant proportion of companies' revenues and the numbers tend to grow (according acfe.com, financial institutions and other online companies lose from 5 to 15% of their income annually due to damage caused by online fraud. The growth of losses may increase subsequent to the growth rates in online banking and fintech markets ([CAGR is approximately 14% per year](https://www.alliedmarketresearch.com/online-banking-market?ref=juicyscore.com)).

According to JuicyScore, the growth of losses caused by online fraud is rather disturbing. Speaking about the new risks, business owners should pay great attention to the problem of randomizers and anti-detection tools, since they are one of the reasons for the sharp increase in the level of risk. According to our data, over 3 years the proportion of randomizers using cases has increased from 0.5-1% to 1.5-2% (in addition, the traffic itself has grown, and, consequently, the number of cases; the volume of using randomization tools can reach 5-10% on peak days). Moreover, about 60-70% of social fraud schemes occur with the use of various anti-detection/randomization tools during the money withdrawal/transfer. In the Russian Federation we observe 170+ thousand incidents per year, which involve randomizers in order to obtain loans. It is 20 times more than 5 years ago. The total amount of prevented damage over the past 12 months amounted to 3+ billion rubles.

![](/blog/content/images/2021/09/Timofey_MNLT-1.png)

The head of risks in ID Finance in Spain and Latin American countries Timofey Kostin presented the report related to the criteria of choosing a provider. Among the most important criteria he named reliability, efficiency, interaction and functionality. With JuicyScore data attributes, the company was able to identify 0.5-3% of high risk population in the issued loans portfolio, to subset the low risk segment 1.5-3 times better than the average. In addition to that JuicyScore data vector attributes allow enriching internal risk score models and define new types of fraud patterns.

Among the main components of the fraud assessment process, Timofey noted the device, network and fraudster's behavior. Great attention should be given to devices overlap or duplication, rare combinations with high risk level, combinations of connection parameters with high risk level, type of use (basing on IP address).

![](/blog/content/images/2021/09/Marina_MNLT-1.png)

Marina Avramenko, Head of the Anti-Fraud Projects Department at Home Credit Bank in Russia, told to the audience about the big surge in social engineeringfraud in banks online channels in 2020. The fraud scheme is arranged in the following way: the client receives a call from an unknown person (fraudster often introduces himself as a bank employee or an employee of the Ministry of Internal Affairs/Central Bank etc., he also often uses number spoofing, imitation of call center and IVR). Fraudsters mislead the client, informing him about a suspicious transaction or application for a cash loan, after that they persistently offer to cancel the operation, saying that they act solely in the interests of the client, trying to imitate the bank's support service ("transferring the conversation" to another employee). Next, the fraudster asks the client to send his personal data - card details and a code from SMS. The client often does not read the text of the SMS messages, where the purpose of the codes and passwords is indicated, and gives out all the personal data to the scammers. According to Home Credit, in case when a fraudster constitutes a loan agreement with a withdrawal of funds to a debit card, on average, the client gives away about 5 codes.

Analyzing JuicyScore data and creating derived attributes, Home Credit Russia was able to expand the data that could identify anomalies in online customer behavior. In addition, significant results were achieved due to the configuration of the architecture and integration, in particular, making improvements in the processes and the big data platform. Among the most effective data and effective rules Marina noted digital fingerprint, device models, operating systems, tracking user activity by session.

![](/blog/content/images/2021/09/Alexei_MNLT-1.png)

A lot of guests noted the demonstration of creating a fake digital fingerprint, which was presented by our partners from NN2 - Alexey Orekhov and Maxim Klimashevich. In addition, they touched such issues as multi-accounting, compromised credentials and spoofing of a legal user. The table below shows the main stages of attacks and protective measures for online business:

![](/blog/content/images/2021/09/Table_MNLT_eng.png)

In addition, a detailed analysis and examples of the most common inventory tools of online fraudsters were presented, as well as the degree of its hazard to financial institutions.

JuicyConf 2021 is a way to understand our customers' needs better and make our technology even more effective for your business. We were very glad to see all of you and talk about the topics and current issues you are interested in. It was nice to hear that our solution is one of the best on the risk-management market and one of the most efficient ones.

In preparation for the next event we are planning to focus on the following steps:

- Further development of randomizer fingerprints identification and solutions on system virtualization;
- Improvement of the efficiency of remote access detection via user behavior analyzing technologies;
- Creation of a platform/solution for joint traffic filtering for financial institutions and marketplaces;
- Development of automatic decision making component by adding a third method of self-learning of the system;
- Improving of the device matching technologies;
- Dynamic authentication.

Thank you for helping us to improve our solutions. See you on our next conference!

Dear Friends!
JuicyConf 2021 was held 8-10 September, it was devoted to the new trends and risks in digital risk management. All the guests who attended the event — risk managers

MNLT---------1.png

JuicyConf 2021: digital risk management outcomes, ideas, breakthroughs

Most companies that create antifraud solutions for online business do not disclose their methods and technologies of high risk patterns and behaviour markers identification. We at JuicyScore have different point of view and reveal some interesting details about the antifraud technologies to our partners and clients apart from the tools for fighting fraud. This helps us to add some transparency to the ways we identify high-risk segments in application flow, shreds light on the operation principles of the markers and attributes presented in our new API 13 and also increases customers' confidence. New API 13 was enriched with the wide range of new risk-markers and today we are going to give you some interesting details about them.

One of the main challenges when making a new product, especially when we speak about such a significant thing as development of new API generation, is verification of the most dangerous and risky markers and patterns that incur direct losses for the financial institutions. Search and implementation of such markers lets to decrease financial losses significantly: for example, it will protect the company from issuing a loan to an unreliable client and also will allow to reduce operational costs during the verification of presumably fraudulent application.

One of the most dangerous patterns which, actually, can be commonly found in online lending is applying from the device with randomizing effects. So, first of all, the main difficulty related to such devices is that it is impossible to verify whether this device has already been used for submitting an application (and if this client has already applied for a loan). Therefore, it is really difficult to assess the history of his behavior in the past. Secondly, the fact that there are a lot of randomization technologies on the device gives us the reason to presume that we are dealing with an unusual client who has rather high technical qualification and for some reason does not want to disclose information about himself, and, therefore, carries an increased risk for the company.

One of the new markers that has been added to the vector of JuicyScore API 13 is the operating system font randomization marker, which provides information about how many manipulations a borrower performed with the operating system parameters before applying for a financial service. In this case we are not talking about basic settings and customization for a specific device user, we mean much more complicated technical settings that require special skills and qualification - outrageous cases when settings and parameters are very different from the standard ones, and the "customization" of such settings extremely exceeds the limits conceivable for an ordinary user who intends to apply for a payday loan. According to our statistics, the share of such devices in the flow usually does not exceed 1% with potential peaks at the moment of activities of potential high-risk borrowers.

![](/uploads/2021_08_12_18_17_56_f8e305bf38.png)

In addition to the settings of the device itself, which can be manipulated by the defaulters, the analysis of applications installed on the applicant's device is also an important part of the risk assessment. The set of applications, their categories and number is an important behavioral marker and such parameters have already been presented in our API earlier. However, in the new version of API we added a technology stack to detect manipulations with installed applications - it is one thing when the official application was installed from an authorized marketplace, bun an entirely different thing when an installed application undergoes changes and modifications and becomes a completely different application. In this case, there is a risk that the borrower applies for a credit with malign intent.

Another important parameter for risk assessment is different frequency responses, for example, the number of risky applications from the same device or IP address, for a limited period of time or simply for the entire observation period. Usually the risk increases with the frequency increase, since this may indicate both - the risk of credit shopping or risk of fraud, especially if such activity is supplemented by manipulation with other data. In JuicyScore APIv13 we have expanded this group of parameters - now, in addition to the fact of application, we will also be able to provide our clients with the information about the various types of web resources the borrower visited in the last 24 hours before applying for a loan. Such metric will allow us to make applicants' risk assessment more accurate as well as indirectly determine his credit profile.

![](/uploads/2021_08_12_18_17_21_6139ac09f5.png)

![](/uploads/2021_08_12_18_18_25_2f019fb06d.png)

The markers we discussed in this article are just a small part of the expansion of information value that we proposed in JuicyScore APIv13. We understand how important it is to expand the set of markers and stop factors that help to save the expenses and reduce losses from the risk of fraud - our team is constantly working on the technology development in order to provide our partners and customers with the tools to outplay suspicious and high-risk clients and to resist the fraudulent activities.

Most companies that create antifraud solutions for online business do not disclose their methods and technologies of high risk patterns and behaviour markers identification. We at JuicyScore have different point

API-13_selected-ideas (1).png

API 13 selected ideas - discovering a new layer of anti-fraud technologies

Vacancy

We are looking for business development manager, based in Russia to join our international team. The candidate must have high level of self-discipline and capable to effectively collaborate with distributed Team. At least 2 years experience in b2b sales/business development is the must. There is an opportunity for promotion in the next 12-24 months to lead business development team in the region.

**Responsibilities**

Business development manager is going to be responsible for international business origination:

- Search for potential clients in digital lending, banking, insurance, payment, fintech industries to build robust sales pipeline;
- Be comfortable to cold call, email, communicate in social media with prospects;
- Perform product presentations and evaluate needs of the client;
- Take part in local offline and online conferences and events, be active member of local fintech community;
- Work in our internal CRM system and keep all activities and deals up-to-date.

**Requirements**

- Have 2+ years of proven b2b business development experience in fintech;
- Have a basic understanding of risk management and fraud prevention in financial industry;
- Familiar with business development automation tools;
- High level of self discipline and organization skills;
- Excellent English.

We are looking for business development manager, based in Russia to join our international team. The candidate must have high level of self-discipline and capable to effectively collaborate with distributed Team. At least 2 years experience in b2b sales/business development is the must. There is an opportunity for promotion in the next 12-24 months to lead business development team in the region.

business-development.jpg

International business development manager

Dear friends and partners!

We are proud to inform you about a remarkable event of this year: a few days ago we launched a new generation of JuicyScore API. We faced quite a challenging task: we had to find and develop fundamentally new technologies. However, this time we did our best to outplay ourselves and launched a new generation of products providing significant additional effect in comparison to the previous versions.

The new product generation has a range of significant changes from the perspective of information value and efficiency for our partners. Here are the key changes and **additional features coming up in JuicyScore API v.13:**

- Information value: we added **28 new attributes** to the output data vector, some of them are unique. Many attributes were created as a result of our researches and some of them have already been patented under applicable law. Our output data vector is the most comprehensive among all the existing solutions for application fraud prevention and account protection at the moment;
- Device authentication: we expanded the **stack of device authentication technologies**, increasing their number up to 31 in the new JuicyScore API generation, which makes our device fingerprint even more stable over time and more resistant to various anomalies. It turns out to be really important in terms of highly competitive and high risk level markets;
- Malware: we ultimately redeveloped and enhanced technological **stack of malware detection algorithms**. The stack now includes 350+ technologies;
- **Infrastructure:**

- **Reliability.** In a new version we increased the reliability of our systems globally as well as the reservation level. We are expecting that the global level of accessibility of our services will exceed 99.9%+. And we are expecting that the loss of availability this year will be 2.5 times lower in comparison with 2020.

- **Monitoring.** We implemented the monitoring of the key nodes and operating indicators across the globe and now we are planning to develop it further and to conduct the monitoring of the key nodes and messages for our clients.

## In addition:

We increased stability and information value of almost all aggregated Index variables IDX;

We significantly extended the list of stop factors;

Data vector information value was increased by more than 30%+ in comparison with the previous API generation.

Juicy Team never stops and always moves forward: this year we are planning to launch a range of mid-release product updates that will allow to reduce risks as well as to get a better understanding of users online behavior without the use of personal or sensitive data.

Despite the difficult situation all over the world as well as the essential changes in the work of all the financial market players, Juicy Team keeps up high development speed in order to extend the functionality of our products and generate more value to our clients and partners.

We permanently conduct researches to identify new attributes, work on strengthening of our expert valuation in digital risk management to help our clients and partners in their day-to-day work, fraud prevention and risk assessment. Thanks to all of you for your support and sharing your ideas!

New version of API is already available to all of our clients!

We would be glad to receive your feedback. Please feel free to contact us by email help@juicyscore.com

Sincerely yours, JuicyTeam

Dear friends and partners!
We are proud to inform you about a remarkable event of this year: a few days ago we launched a new generation of JuicyScore API. We

API-13-2.png

Higher information value, lower risks - JuicyScore API 13 increases IV by 30%+

Amid rapid economic growth of fintech industry which we have witnessed for the last few years, strong and reliable partners make this growth even more sustainable and significant. JuicyScore started a very successful collaboration with ATM online - one of the leaders of online lending industry. The group companies conduct very remarkable and successful activities in Vietnam and in a number of other countries in South and Southeast Asia.

# Cooperation with JuicyScore

The companies started cooperation in the second half of 2020 by launching a pilot project in Vietnam on a new credit applications flow. Sustainable and stable device identifier as well as the informativeness of data vector of JuicyScore solution helped to create stronger risk technology stack and to deal with several challenges related to fraud risk prevention as well as enhanced company's scoring models.

The key result of the pilot project is implementation of score and a range of other vector parameters in antifraud model and borrowers' credit risk assessment; since February 2021 ATM Online launched JuicyScore product commercial use on their platforms in Vietnam.

As a part of fraud prevention JuicyScore made it possible to filter out the applications with evidence of high risk and commercial fraud.

According to Denis Aleksandrov, Risk Director of ATM Online Group in Asia, “Data vector parameters, such as combination of user online behaviour markers, combination of markers of internet connection quality, device technical parameters have been included in credit risk assessment models, which allows to add 5 Gini points to the model resolution. Among with implementation of antifraud rules based on JuicyScore device identifier it also reduced the level of credit risk as well as yielded in quick ROI during the first months of cooperation (expected 5 times ROI with strategy further adjustment on 6-month horizon)."

JuicyScore also helped to reduce the level of commercial fraud significantly.

# JuicyScore

JuicyScore company was established in 2016. It provides technological solutions for fraud prevention in online-business and financial products risks reduction, it also provides customized score, analytics and consult on score modeling and decision making system in 17 countries all over the globe. All the solutions are customized due to the clients need providing analysis of data flow and score. The main feature of JuicyScore is that personal or sensitive data as well as direct user identifiers are not used, which has certain positive impact on the clients' reputation. The company is compliant to EU legislation on personal data (GDPR) all the device authentication technologies and data algorithms are proprietary, the company has 4 patents.

# ATM online

ATM online is an international online lending company that offers customized consumer loans with the single-platform solution. The company was established in 2015. The service is based on the latest technologies such as holistic scoring and highly predictive models. The service has over 500,000 registered users and is growing immensely - today's credit portfolio reaches 7 mln USD. Retention rate of the company is extremely high: 9 out of 10 customers come back to take another loan. ATM online integrated all major payment channels, such as Payoo, Viettel and bank transfers.

Amid rapid economic growth of fintech industry which we have witnessed for the last few years, strong and reliable partners make this growth even more sustainable and significant. 

image--3--1.png

Two times 5 in Vietnam: +5 Gini to the models, 5x data ROI. ATM Online case study

Despite a slight slowdown in microfinance sector during the pandemic period some experts prognosticate the rapid growth of microfinancial industry [in the next 5 years](https://www.reportlinker.com/p05799111/Global-Microfinance-Industry.html?utm_source=GNW) due to the rise and implementation of advanced technologies. In our new article we are going to tell you about the latest business model approaches that can be powerful instruments in the way of evolutionary growth of microfinance organizations and neobanks in a constantly changing and challenging environment.

International microfinance sector development experience shows that setting interest rate limits on loan products is an inevitable process and financial institutions that can successfully adapt to such conditions have an opportunity for sustainable growth. In addition, these financial institutions can rely on strong customer support and loyalty, which improve the image of the companies in the opinion of regulators as well. The experience of such countries as China, Great Britain, Spain, Poland and Latvia that managed to implement the principle of "create first, regulate later" shows that this is the only possible way, while the restrictive government policy regarding fintech industry leads to significant growth of the gray market and to a lack of transparent regulation and consumer protection.

Basing on our experience and work in 17+ countries JuicyScore experts believe that the following business model development scenarios are worth to be taken into consideration by fintech companies:

- Changing user acquisition model and reducing lead generation cost;
- Gradual transition from PDL to Installment (IL) model by increasing the average duration and average size of the loan from 15-30 to 45-60 + days;
- Reducing the cost of risk through investments in risk assessment technology, new data and increasing the proportion of loans provided with mobile applications;
- Development of online POS lending based on the installment plan / online loan model;
- Applying of microfinance organizations to a pawn shop transformation model, development of a pawn mechanism in the lending system based on the pawn of a personal mobile device, vehicle and/or other available assets of the borrower;
- Development of partnership with banks and large financial institutions in order to work with the “rejected” flow of applications;
- Optimization of OPEX to reduce the cost of new customer acquisition and customer service.

In this article we are going to take a closer look to the above mentioned scenarios which may help to adapt to the decreasing rate conditions and find your way to a sustainable business model.

## Changing customer acquisition model and reducing lead generation cost.

The structure of the loan cost can be represented as a combination of five components: cost of lead, cost of data, cost of risk, cost of operational support and cost of funds (Chart 1).

![](/uploads/eng_1_9413bfa84f.png)

As you can see on the chart consistent work with each cost will lead to an overall decrease in the cost of a loan by 1.5-2 times.

The left column (Max) shows the peak values. As you can see, a significant share is the cost of lead (there are some cases when the cost of lead is 60-70% of the cost of a loan). The middle column (Average) is an intermediate stage between the current and potential target. Reducing the cost of these components (all of the them or at least some significant part) will allow to shift to a model with a target cost level (Min) gradually.

The main part of the cost of a loan is the cost of lead, so the most tangible and significant result may be achieved by reducing this particular component. The emerging trend of regulation (and, as a matter of fact, limitation) of lending rates will bring pressure to bear on lenders as well as on their efforts to reduce the cost of customer acquisition. Reducing the cost of acquisition by increasing the share of natural traffic, extensive use of low-frequency contextual keys, media partnerships, etc. is an important step towards the target cost model.

## Transition from PDL to Installment (IL) model by increasing the average duration and average size of the loan.

Rate regulation and reduction may be rather painful for any microfinance organization. On the one hand, rate reduction increases the availability of borrowed funds, on the other hand, it reduces the target audience in those segments where such rates cannot cover the level of risk. The main challenge of the market is finding a balance between the rate and the level of availability of borrowed funds.

Chart 2 shows an example of yield enhancement scenario by increasing the loan duration.

![](/uploads/eng_2_37510d3801.png)

# Reducing the cost of risk through investments in risk assessment technology, new data and increasing the proportion of loans provided with mobile applications.

Last November we conducted [a survey](/en/blog/keep-the-balance-information-value-and-compliance-en/) among our clients, who are the top performers in microfinance market all over the globe. According to the results of our research, the main categories for data market turned to be the balance of information content and payback as well as compliance with current and prospective regulation rules, especially taking into consideration the trends and changes we observed within the last 2-3 years (i.e. requirement strengthening in the Russian Federation regulations, a project regulations similar to GDPR in China etc.). Therefore, data data payback and maintaining the balance between the cost and data informativeness will play an increasingly important role and become an important step in reducing of the cost of a loan.

Risk reduction can be achieved by improving the methods for risk assessment, including the development of a technology stack, dynamic authentication, processing of all data sources available on the market in order to increase information value, quality and speed of decision-making. Increasing the proportion of loans provided with mobile applications will also reduce the level of risk for it will also increase the amount of data available for risk assessment.

![](/uploads/eng_3_6da9549507.png)

Chart 3 shows a general approach to the cost reduction of the components described above and to the loan cost optimization. As you can see the transition from one product model to another will allow to stabilize the yield of the loan in absolute terms in the context of gradual decrease of the maximum interest rate.

## Development of online POS lending based on the installment plan / online loan model.

POS lending is one of the fastest growing segments of the microfinance market. Online trading turnover has grown significantly: according to the Retail Companies Association (ACORT), the impact of the pandemic will accelerate the growth of e-commerce in Russia by 6% per year until 2024. Thus, "going online" spurred the growth, which was restrained by the decline in real income during the quarantine and the subsequent restrictive measures. Development of online-POS lending with a relatively low-risk audience and cooperation with retail will become one of the growth drivers in microfinance and neobanking sectors.

## Applying of microfinance organizations to a pawn shop transformation model, development of a pawn mechanism in the lending system.

One of the effective tools that will reduce risks by means of the assessment of the collateral as well as maintain the interest income at the same level while reducing the burden on the borrower, is applying of microfinance organizations to a pawn shop transformation model. The collateral is a certain guarantee of debt repayment, depending on the category of collateral (its value and liquidity), the risk assessment may change, which makes it possible to implement a flexible approach to the range of credit products offered.

## Development of partnership with banks in order to work with the “rejected” flow of applications.

Speaking about borrower selection, banks have more strict and rigorous approach and this trend has increased lately. In 2018 about 41% of loans in the Russian Federation were approved. In the end of the last year the share of approved applications dropped below minimum, reaching 32.2%. Experts are attributing this this to the measures taken by the Bank of Russia against the growth of household debt load as well as decline in borrowers' income. This situation offers great promise for cooperation between banks and microfinance organizations in order to work with the “rejected” flow of bank clients' applications. Development of this cooperation will allow microfinance organizations to get access to better traffic, because the segment of potential bank borrowers usually has better payment discipline and borrowing capacity. Access to the rejected flow of large financial institutions will reduce the cost of acquisition and lead generation.

## Optimization of OPEX to reduce the cost of new customer acquisition and customer service.

Operational expenses that are not related to the cost of the data still remain an important field for the loan cost optimization. By such expenses we mean the entire cycle of operations occurring within the organization for getting a client, maintaining all the documentation, accounts, payments, as well as a call center and collection of debts. Reducing the OPEX for case management often requires many changes in the way organization operates as well as an increase in the internal efficiency. Among such measures can be a transfer of a significant part of transactions from offline to online, reduction of the number of required documents for obtaining a loan, usage of various technologies to reduce the burden on call centers, optimization the debt collection costs etc.

The year 2020 brought many challenges, but we can't but agree that even in this difficult period microfinance organizations shoulder the burden of a very important function of developing a new model of online lending stepping aside of the classical banking model. Microloans have their own position in the financial services market and microfinance organizations have become a significant institution that is capable of solving many social and economic problems. The microfinance sector deals with an important problem of financial inclusion that has become a genuine concern in the recent years. Microfinance organizations are sometimes the only alternative to traditional lending. According to [creditsummit.com](https://www.mycreditsummit.com/microfinance-statistics/?ref=juicyscore.com), about 65% of the world's microcredit consumers live in rural areas where there are no banks or ATMs, while less than 20% of borrowers from developing countries received loans using traditional financial instruments. High competition among microfinance organizations, openness to further innovations will allow them to increase their market share and to defy competition with traditional banks.

Despite a slight slowdown in microfinance sector during the pandemic period some experts prognosticate the rapid growth of microfinancial industry in the next 5 years due to the rise and

runnung-2.jpg

Microfinance evolution: developing an effective business model

The past decade has seen a series of technological breakthroughs which opened many opportunities for the fintech industry and took it to new heights. The Covid-19 pandemic has had a significant impact on the financial sector and fintech is no exception. The crisis forced players to improve efficiency rapidly, optimize costs and implement new solutions. Based on the latest research and expert observations, we are going to tell you how the fintech market reacted to the pandemic in different regions, how it affected the risks and what trends will determine its development in 2021.

Challenges 2020
---------------

The year 2020 gone by and speeded up the processes that many companies were trying to implement before and also shed light on some new problems.

# Go online.
The process of active digital transformation has become the main growth driver for many companies. For the financial sector it has become obvious that moving online a significant part of business processes is inevitable and now is an essential condition for the development and maintenance of any company's position in the market, as well as for obtaining new business in the future. Fintech industry leaders now are and have always been online early on and have all the necessary infrastructure and technology stack, so today they have an additional advantage over traditional financial institutions, which had to restructure their work root and branch. For example, JPMorgan Chase, the largest bank holding company in the United States ranked by total assets decided to expand its presence in the UK, but instead of opening branches the bank offers services only via a mobile application.

[According to the McKinsey study](https://www.mckinsey.com/industries/financial-services/our-insights/a-global-view-of-financial-life-during-covid-19?ref=juicyscore.com), which covers 30 countries and 86% of the population, about 40% of respondents choose online channels much more often when using financial services, about the same percentage of respondents said that they would not return to their previous habits and would use financial services online in the future, despite the end of strict measures connected with the lockdown.

# Global investment flows dip.
Amount of investment in developing markets has decreased compared to previous years, however it should be noted that the main fall occured at the peak of the pandemic - the end of Q1 2020, when [the volume of investments fell to the level of 2017](https://www.cbinsights.com/research/fintech-funding-q1-2020/?ref=juicyscore.com) . Total volume of merges fell 24% in Q3 compared to the same quarter in 2019. However, amount of investment in Q3 rose to 10.631 billion USD. It has become the most significant result in 2020 and the second-best quarterly indicator since mid-2018. [According to CB Insights](https://www.cbinsights.com/research/report/ai-in-numbers-q2-2020/?ref=juicyscore.com), the problem of reducing number of transactions mainly affected startups, since traditional financial institutions that used to work with startups earlier are now creating internal development departments and working on their own financial solutions. Experts suggest that under the new conditions startups and small tech companies, which have been the growth engine of fintech development in recent years, nowadays may have some difficulties in attracting new investments.

# New regulatory requirements.
Fintech uses a wide stack of technologies to support and develop online business, designed to speed up the interaction with a client, making payments and conducting other financial transactions. At the same time the importance of privacy and data protection is growing immensely. As a way to compete successfully with traditional financial institutions it is necessary to take into account all changes in legal regulation, particularly regarding the requirements for the processing of personal data. Increasing regulation in 2020 in various regions is one of the factors that experts note among a number of reasons that negatively affect market growth. For example, regulations similar to the General Data Protection Regulation (GDPR). This problem also affected Asia, at the end of 2020, an analogue of the GDPR appeared in China - Personal Information Protection Law, applicable to foreign legal entities and individuals who process personal data and provide services to entities located in China.

In addition to government regulation, it is necessary to take into account the policies of technology companies and bigtechs regarding the software they produce: browsers and operating systems in respect of collecting and processing user data. Failure to comply with these requirements can lead to a situation when certain mobile applications and websites of fintech companies may be blocked.

# Data storage and processing security.
Due to the pandemic a large number of users switched to the online channel as a main source of obtaining financial services. In fact, online channels allow companies to reduce operating costs significantly, as well as to gain access to a wider audience on the one hand, and not to risk with one's health speaking about the clients on the other hand. However, today we are witnessing a really big leakage of personal data day by day and when it comes to the Internet it leads to increasing number of cyber attacks globally. That is why the secure storage of user data is becoming one of the most important issues in today's online business.

[According to Security Intelligence](https://securityintelligence.com/posts/whats-new-in-the-2019-cost-of-a-data-breach-report/?ref=juicyscore.com), the average cost of data losses in 2019 amounted to 3.92 million USD. Leakage of customer-sensitive data such as payment card numbers, passwords, financial transaction data, passport data and other personal documents can lead to a complete loss of customer's trust. Users want to see the transparency of their operations as one of the criteria of high-quality services provided. The year 2020 has shown that fintech companies need solutions that combine both factors - transparency and security - more than ever. The best solution is a reasonable reduction in the turnover of personal data and direct user identifiers, combined with the use of alternative sources of information, such as device data, Internet connection data and user behavior.

# Credit risks & business pivot.
A smooth sea never made a skillful sailor, so in 2020 those players who understood the importance of maintaining balance between reduction of credit risks and the speed of adjustment and creation of new business models turned out to be the most successful. Financial institutions that concentrated all their efforts on a quick solution of the problem, in particular, companies that focused on changing risk-based pricing, setting new limits etc., as well as those who continued to work hard and did not suspend most of their work processes during quarantine, managed to successfully return to normal rhythm in Fall 2020 and essentially won the market. The time has shown that companies that did not cope with quick adaptation reduced traffic volume, lost citation in context and faced other problems that affected the rather painful process of returning to the usual schedule.

Trends 2021
-----------

# Formation of new ecosystems.
The new formats of fintech market eco-systems turned out to be more competitive in a number of countries in comparison to the captive/internal eco-systems of corporate giants due to a higher development speed, better coverage and greater flexibility/adjusting to the specific needs of the client, as well as the development of partnerships with other companies. Moreover, in 2020 we noticed that small companies left larger companies behind speaking about the speed of development and implementation of new technologies. Forming a proper ecosystem around them they will be able to compete with such giants as Apple or Google. One of the most successful fintech startup examples [according to Forbes](https://www.forbes.com/fintech/2020/?ref=juicyscore.com#11622e3a4acd) is the MoneyLion neobank. Its application lets not only to open a new account or check it up, receive services related to cards, the company also offers an innovative automated solution for managing ETF investments.

# Big Data.
[According to statista.com](https://www.statista.com/statistics/254266/global-big-data-market-forecast/?ref=juicyscore.com), the global Big Data market will reach 103 billion USD by 2027 and will double compared to 2018. Financial companies try to collect as much information about the client as possible in order to meet the demands of the customer and to personalize offers and increase customer loyalty by analyzing the data. Big Data allows you to make your business more client-focused and achieve better results. Fintech companies can focus on customers' segmentation, receive real-time information about users, optimize pricing policy and predict which services may interest the customers in the future.

However, the more data you get, the more difficult it is to manage it effectively ensuring the security of customer information from third parties at the same time. Over the past ten years financial industry has invested heavily in data collection and processing technologies: data warehouses, analysis tools, data visualization tools, forecasts based on current information. According to experts, fintech companies will actively continue working on this in 2021, since user expectations have changed quite dramatically. More difficult conditions have led to a greater competition and therefore the companies must process all the data quickly and efficiently keeping in mind that today security really matters.

[According to Oracle](http://www.oracle.com/us/industries/oracle-era-ready-anything-2969053.pdf?ref=juicyscore.com), financial institutions that offer more personalized financial services to their clients raise their income 18% per year. According to use case of one of the branches of American Express Bank, it created a complex forecasting models using Big data technologies that helped to prevent customer outflow: based on the analysis of previous client transactions, the bank was able to determine which accounts were likely to be closed over the next few months. This may help to take necessary measures and reduce customer churn.

# Cybersecurity.
Cyberattacks lead to rather harsh consequences and loss of reputation of any financial company. Problems arise due to many reasons, the main of them is cooperation with third-party organizations, extensive use of mobile technologies and online data transfer, the natural growth of hacker attacks. [Cybersecurity Ventures estimates](https://cybersecurityventures.com/annual-cybercrime-report-2017/?ref=juicyscore.com) that cybercrime-related damage reached 6 trillion USD a year by 2021.

![](/uploads/image_1_3daaf1b2c1.png)

Size of the cybersecurity market worldwide, from 2017 to 2023 (in billion U.S. dollars). Source: Statista.com

Among the focus areas of cybersecurity work for fintech companies experts pointed out the following ones:

- Security of transactions, risk management technologies.
- Technologies of secure storage and exchange of data, including personal data protection.
- Remote authentication technologies and biometric solutions.

New conditions - new opportunities
----------------------------------

The main thing the last year showed in the context of the fintech industry - the challenges of 2020 increased the speed of change in a number of areas and set a new bar for business efficiency as a result of investment decline.

The second thing that is also really important to note: Fintech leaders showed really strong competitive advantages compared to traditional financial institutions. For example, the revenue of the fintech sector fell by 10-20% at the peak of the crisis, while the bank losses were several times higher - 40-70%. Large banks that received funding in 2020 could not oppose anything to fintech, except for the growth of mortgage.

Third, along with the difficulties the pandemic has brought new opportunities. The pandemic marked the final step for many companies to go online as well as brought the era of full digitalization, which gives the chance to fintech to significantly increase market share.

The fourth important conclusion is that the speed of new technologies implementation is becoming one of the main factors of market success. Organizational set-up of fintech companies and their data providing and processing policy have to meet these requirements.

This year Juicy Team will continue to maintain a high pace of development and delight our customers and partners with new product functionality, for you can ensure high competitiveness and business safety only through efficient and hard work in a dynamic digital world.

The past decade has seen a series of technological breakthroughs which opened
many opportunities for the fintech industry and took it to new heights. The
Covid-19 pandemic has had a

fintech-3-1.jpg

Fintech 2021

Data

We always pay attention to the details related to the business of our clients and also try to keep abreast of the latest trends in the data market to make our products even more valuable. In addition, we always keep in mind the important conditions that must be taken into consideration working with the big variety of data. In this November we conducted a customer survey and asked our clients to express their opinion on various data sources they use in decision-making system, which particular sources they choose and also asked them to share their business experience and thoughts about data regulation. A few interesting conclusions emerged during analysis of our survey's summary results: in what way lenders around the world approach the choice of data sources for making their credit decisions for new applications? Now we are going to tell you about the prospects of the development of the data market in the nearest future.

# Most usable data sources: know your trend

It is essential to mention that our clients’ business is located in 14 countries, including Russia, Kazakhstan, a number of European countries, different countries in Middle East, Southern Asia and South-East Asia, so the information that we collected is unique and rather valuable and reflects the real picture of the present markets. We were aimed to verify most common approaches to data sources valuation as well as to develop important guidelines which are true to all the markets.

Among the main data sources that can be used by loan organizations for making a decision on new applications approximately 80% of our respondents pointed out 4 main sources: credit application data, credit reporting agencies (or credit bureau), device data and telco data. It seems fair to say that the above mentioned sources are the touchstones for credit organizations of the countries of our presence. Such choice is fully justified by the number of reasons. Any credit process is initiated by a credit application and all the data indicated in such application may be regarded as a starting point of any decision-making system: application data assessment (mostly identification or socio-demographic data) determine the degree of reliability of information and, as a result, the level of financial institute credibility towards an applicant. Credit reporting agencies data directly affects the probability of loan approval or and defines loan parameters, as such data describes most relevant applicant's behaviour pattern and level of responsibility, as well as intend to fulfil financial liabilities. Along with that credit reporting agencies are used to check the validity of initial identification data or, often, contact data. Telco data is the important supplement to the first two categories, which usually describe behavioural or lifestyle component of an applicant, may be used to estimate the probability of fraud risk or disposable income level. Device data allow getting additional information value via alternative view on the applicant through the non-personal data analysis and to complete borrower's portrait especially in this cases, when classic direct identifiers based data sources are lacking some information. Combination of all data categories allows getting holistic applicant view and to make the efficient decision on financial product and its parameters.

Despite the availability of credit reporting agencies data in some Asian countries, the results of our survey showed that there is a great lack of strong data institutions or quality of the data provided by them is rather low. Meanwhile our data is in the first or second place in decision-making system in these countries, which led us to a conclusion that our data is even more valuable for our clients than we could ever thought. Data services scalability as well as high hit rate are becoming the key factors for all the marketeers.

![](/uploads/_2fe7ed5573.jpg)

As you can see on this graph, device data is behind free application data as well as credit reporting agencies and telco. Therefore data provided by JuicyScore may be classified as one of the most demanded and frequently used in decision-making systems of online business and intensely influence on quality of decision-making process, overall risk level and portfolio profitability.

Despite the availability of high-quality data of credit reporting agencies in Russia and some other countries as well as the the other sources, JuicyScore rank second to none in decision-making systems in more than 35% of cases (needless to mention free sources such as application score, internal credit history, negative lists, etc.) In 65% of cases JuicyScore data is ranked in top 3 data sources. Only a few years ago device data were coming off as an optional solution that had partial influence on decision-making system for the main part of players, however, as the time goes by our data have proven its efficiency and usefulness and now it becomes more clear that device data have become one of the main fundamental parts of decision-making systems of loan organizations all over the world.

Speaking about the most important parameter for choosing a data source for a decision-making system, which was highlighted by most of our clients, it turned out that the balance of information value and payback takes the center stage and has the highest priority.

![](/uploads/_83b9c7a8c8.jpg)

It stands to mention that combined with the functionality of our solution we also pay great attention to the balance of information value and data payback and we are really proud of our strong performance. This parameter, to our mind, plays the most important role in efficiency calculations of every component of decision-making system and serves as the reminder of the necessity to observe the information value balance and and data source legitimacy.

# Data legal regulation and market long-term outlook

Analyzing the results of our survey we noticed that our respondents do not consider compliance with current and prospective regulating rules as a top-3 criteria. However, basing on the trends we have witnessed over the last 3 years, we can confirm that the above mentioned factor may play crucial role considering selection of data source. Vice versa the source which is compliant to all the data regulations may be the eligibility criteria required for further successful cooperation with official financial institutions. Herewith by regulation we imply not only compliance with the current regulation, but also with the rules set by Central Banks in different countries and any other regulators, browsers’ security policies etc.

Moreover, many companies leading online-business have already been affected by some difficulties related to inconsistency with law regulations (or being not entirely compliant to them). This refers to the categories of collected and processed data regulated by law, including financial transactions data, calendar data, correspondence data, medical history, passport and other documents details, email, address, education and work information etc.For example, if a mobile application does not inform a user about the categories of data collected, the application may be removed or suspended by the marketplace operator (for example, by Google). All these circumstances may pose reputational or financial risks to any company, particularly in case if a high percentage of applications comes through a mobile app. Sometimes companies have to deal with more complicated cases when they collect and process some kinds of data which relate to sensitive or private categories or if collected data processing is prohibited or limited by the law. Such categories of data may be highly informative, however the price is even higher - a revocation of license and regulator's permit to run a business on the territory of the country.

Among the underlying trends which, in our view, are gaining steam may be noted the expansion of the list of data regulated under the country legislation system, expansion of the list of personal and private data, trend to online user privacy increase as well as the settings of operating systems and browsers recent and upcoming versions, closer attention to mobile apps of financial institutions and data content, which are processed within these mobile apps and some other.

While developing of functionality and technologies of our solution we are compliant to all the strict regulation requirements in all the countries of our presence in order to provide stability of decision-making systems of our clients.

As for the development of our solution we deliberately stay away from the data that may be viewed as direct identifiers or may be considered to be a part of a bank secrecy or related to the privacy of correspondence.

In early November this year we stopped using document.cookie for it is not compliant anymore to the security policies of browsers (Chrome, Safari and some others). We replaced it with method of event subscription. Moreover, all our products are compliant not only to the current regulation rules, but also to some of expected changes in the legislation of Russia and a number of other countries.

We express our gratitude to those partners who took part in our survey and also mentioned growth area which they consider to be important in the future. Our clients pointed out the following promising areas:

- Developing of new technologies of fraud risk markers and high risk markers detection
- Increasing of data information value
- Score models customisation by products types and geographies

Among the most valuable features of our products our clients emphasized reasonable data speed and data payback, quality of the data presented, high speed response, high quality of our support service and ease of integration of our solutions.

We are encouraged to strengthen the quality of our solutions, keeping the balance of information value and being compliant with all the data collection and processing requirements.

We always pay attention to the details related to the business of our clients and also try to keep abreast of the latest trends in the data market to make

Libra-5.jpg

Keep the balance: information value and compliance

Technological change, revealing and adding new features to our products is the very nature of the JuicyScore team. No false modesty, we help our clients to achieve tangible and quantifiable results. By polishing up our technologies we find solutions that lead to a balance between conversion rate and usability of internet resources vs risk management for online companies.

We are really proud of launching two new products this year – JuicyID and JuicyBio – both of them are solving the abovementioned problem and today we would like to tell you more about all the features and advantages as well as the opportunities of joint use of our products.

JuicyID and JuicyBio are based on JuicyScore API 12 launched in March 2020. Device authentication system is based on 29 technologies that identify sustainable and resistant device fingerprint.

# What is JuicyID?

JuicyID is the light version of JuicyScore. This product is intended to protect online accounts of various online-services. JuicyID not only took all the best features of JuicyScore, we also added a range of new utilities and higher-speed response.

Returned Device ID which is the most stable and steady digital fingerprint that exists on the market at the moment and vector of 81 attributes allow designing various rules to prevent risk of fraud and to balance payback and costs, taking into consideration the fact that online fraud has become one of the major issues of online business nowadays.

# What is JuicyBio?

JuicyBio is an expansion for JuicyID and JuicyScore enriched by descriptors of the photo of prospective applicant for a loan or insurance. The applicant’s passport or other ID document’s photo is compared to the photo taken at the moment of application. Combining Device ID and descriptor comparison results, the authentication of the virtual applicant can be enhanced significantly. In response you can see whether applicant’s photo matches the passport or other ID document’s photo and the risk level is specified. The main advantage of JuicyBio is high level of accuracy and security: JuicyBio does not collect personal data or direct identifiers and does not store or process the original photos or their parts or any other personalized tags of these photos (for example, titles).

We have already revealed all the secrets of seamless authentication in our previous newsletters, today we would like to tell you about our practical experience and tools as well as the extended spectrum of online industries, which associate with high risk clients, personal accounts compromising and many other cases that lead to higher risk and can do harm to financial or quasi-financial information.

# Dynamic authentication

Taking into consideration the fact that all the products of JuicyScore do not process personal data and direct user identifiers, we will tell you about the part of the process related to device seamless authentication and virtual user.

Seamless authentication consists of the following stages:

![]({{UPLOADS_HOST}}/uploads/Flow_en_189fa22502.png)

In order to design a decision support system it is essential to find a balance between risk reduction, conversion management and costs.

![]({{UPLOADS_HOST}}/uploads/1_eng_1_ffc9cc4f54.jpg)

Stages 1 and 2 of dynamic authentication allow to filter out up to 80% of high risk traffic without considerable reduction of conversion rate as authentication process is seamless for virtual users and rejection is driven by negative factors that barely have anything to do with the good clients.

It has to be said that the third stage of seamless authentication should be placed to the system really carefully. Here are the key issues of biometry:

- direct data forgery (paste-in slip, panoramic photo etc.), false and fake profiles in aggregated biometric data systems (via direct insertion of forged data or personal accounts hacking);
- incomplete database of biometrical information and low image quality;
- normal increase of irrelevance of biometrical data impairs the accuracy of matching;
- reduction of risk and reduction of conversion balance;
- high cost of biometry solutions including both direct operational expenses and the expenses for technological integration and system ownership.

# What are the advantages of JuicyBio?

- combination of algorithms for digital profile design and search for matches in images of the client and client’s ID document’s photo;
- high response speed;
- safety – all the photos are stored in the client’s device or at the online business framework and are not copied to JuicyScore infrastructure;
- economically justified expenses which are much lower than the expenses the company could possibly bear for any other solution. It is also rather convenient that it does not require any costs to support the product by online business.

For that reason collection of data and analysis of biometry by the “safety biometry” technology should be made before the stage of client’s personal data collection and scoring, which leads to reduction of risks related to this category of data and costs for external data sources calling.

According to the results of dynamic authentication the application may be denied or client may be requested some additional standard personal data sources which is necessary for making a decision (stop-lists, verification of invalid passports, credit bureau, mobile network operators etc.) and the decision concerning the application for getting the financial product itself or, for example, access to personal account.

As an example of final metric of any system of seamless authentication the following metrics may be used:

- the total cost of verification of an applicant or a client – it is usually measured in absolute numbers and is considered to be the calculation basis for payback of every source of data used while seamless authentication is processed;
- end-to-end conversion rate - this metric is used to balance marketing and risks;
- risk level - this metric is one of the main tools to evaluate profitability and losses.

All Juicy product line is available for tests and installation and has proven its effectiveness in Russia and 13 other countries.

Effectiveness of use of our products is easy to count as it is based on service and data ROI principles.

All our products don't use or operate with personal or biometrical data or direct user identifiers.

Which seamless authentication solutions do you use? We will be happy to receive any comments or feedback from you!

Technological change, revealing and adding new features to our products is the very nature of the JuicyScore team. No false modesty, we help our clients to achieve tangible and quantifiable

Apples-1.jpg

Awesome Juicy trio: JuicyScore, JuicyID and JuicyBio on guard for online business safety

Fraud prevention, as management of all other types of risk, is an endless process of improvement and technological development. This allows financial institutions to invent additional tools for decision-making system efficiency improvement. Such tools help to manage new types of risks and define better ways to cut already known risks. The endless variety of the online world, on the one hand, makes it possible to create a convenient environment for receiving financial products and services, and on the other hand, it brings risks and leads to difficulties, for both financial institutions and their clients.

Everyone loves beautiful and functional web resources with various graphic solutions, fonts and other design and visual elements. Many of these elements are implemented by loading various components from external resources (domains). A significant number of such domains are reliable and relate to trustworthy organizations; various elements downloaded from such domains are used by many financial institutions, retailers and other service providers operating online.

However, among all this diversity, there is a proportion of cases (according to our estimates, from 0.5% to 5%) when the downloaded data and components do not benefit the financial institution, but are malicious and can be used for dangerous purposes. Often we are talking about various web portals that are not directly related to the bank or microlender, since the market leaders usually closely monitor what is posted on their sites. We are talking about those portals where f the user, or example, can fulfil an application for a loan or insurance product without direct reference to a specific financial institution.

# Where do dangerous domains come from?

There are two principles which may be distinguished in terms of how such resources operate:

- deliberate actions of the virtual user, the owner of the device used for application fulfilment and sending; in this case, the application inherits links to all domains which will be automatically attached to the application for a financial product, including those domains which may bring a threat to both the applicant and the financial institution;
- actions of third parties who secretly or explicitly install scripts without the applicant’s or Internet resource owner’s consent.

# What is the danger of such domains?

In some cases, there can be manipulations with browser settings, Internet connection settings, various scripts for automated or automatic loan application fulfilment. Even the presence of such markers during an application fulfilment process brings an additional risk for a financial institution. In the most difficult cases, fraudsters have the opportunity to gain remote access to the applicant's device, steal logins, passwords and other personal and sensitive information of the user through such dangerous domains. All this leads to risk both for the user, whose data has been compromised, and for the financial institution receiving a financial product application with sometimes knowingly false data.

# How do we identify such cases?

We have developed a methodology to proactively identify such kinds of domains. This means that there is no need to wait for the application to mature and to confirm the applicant's bad faith.

Dangerous domains are defined based on the following algorithm:

- Determining the level of danger of a domain - the analysis is based on a number of metrics and parameters, such as, for example, the use of JuicyScore values ​​to identify the concentration of risk for such domains, statistics on the claims occurrence frequency with such domains, cross-validation with other attributes processed by the JuicyScore, domain registration and owner information, behaviour patterns, domain characteristics and functionality verification, and much more;
- Identification of domains which scripts were uploaded to a financial institution web resource and which were additionally revealed during application fulfilment - the most interesting cases are related to those domains that appeared in loan applications for the first time;
- Identification of patterns associated with frequent changes of low-level domains (for example, n01.vwxyz.com, n02.vwxyz.com), when the same script is downloaded from different domains associated with the same group.

In addition to that the cases related to the absence of any single domain associated with the application fulfilment and typical for a given segment of applications are also suspicious and may be treated as a negative anomaly.

Such domains analysis results should be used to identify groups of related applications with a high level of fraud risk, therefore, this functionality is already available within the most recent version of JuicyScore API v12, and the fact of such dangerous domains presence in a loan or insurance application is returned within the response. This information will allow to identify the most dangerous cases and will enable our partners to respond promptly to them before the actual risk matures and mass loan disbursement occurs for such applications.

Over the past 4-5 years, we have been developing a number of ideas, such as, for example, reducing personal and sensitive data turnover or balancing privacy and anonymity. All these aspects are deeply utilitarian, and applying them in practice makes it possible to reduce risks and obtain an economic effect. We are constantly developing products and technologies so that our partners and customers have the opportunity to stay one step ahead of attackers, to obtain a high level of data ROI and an efficient balance of risk reduction and the cost of this reduction.

Fraud prevention, as management of all other types of risk, is an endless process of improvement and technological development. This allows financial institutions to invent additional tools for decision-making system

Thunder-1.jpeg

Not All Domains Are Created Equally Safe And Useful

In times of crisis, when the markets drop, one of the key goals of any company is to maintain long-term and time-tested partnerships. This is especially important when the crisis affects a huge number of areas, industries, countries and macroregions, when the usual models and rules suddenly stop working, and there is a minimum of time to build new ones.

JuicyScore has been operating for 4+ years, providing services to financial institutions in 13 countries and a technological solution for assessing various types of risk in online lending based on proprietary technologies and without using borrowers personal data and direct identifiers. The digital fingerprint of the device (JuicyDeviceID) provided as part of the service allows to successfully prevent technical fraud, device manipulation and randomising. The vector of informative data, which consists of almost 200 attributes, including anti-fraud scoring, a set of combined aggregates by device, virtual user and network connection, strengthens internal credit risk models and allows setting-up a set of rules for filtering high-risk applications with fraud markers. The depth and constant development of technologies allows moving forward and provide additional means to reduce losses from operational risk.

We are happy that there are companies among our partners whom we cooperate with from the very beginning of our work and continue to fruitfully cooperate today, helping to mutual growth and synergy. We are inspired that our products and technologies help companies to become the leaders in the industry.

One of these partners is the [Robocash Group holding](https://robocash.group/) (represented in Russia as MFC Zaymer), which has rightfully become the leader of the microlending market in the Russian Federation and occupies a strong position in the markets of 6 more countries. Through this partnership, JuicyScore provides a fraud and credit risk assessment service to significantly reduce losses from high-risk applications. In addition, JuicyScore, within its competence, assists in the investigation of credit applications related to high risk cases and fraud schemes.

In February-March 2020, as a result of the joint work of experts from both companies, MFC Zaymer portfolio segments analysis was carried out in order to identify the most difficult cases of fraud and their further prevention. As a result of this project, a number of anti-fraud rules were developed based on the data vector, which is provided as part of the JuicyScore solution. For example, rules were built on the basis of anti-fraud scoring, a combination of high-risk markers, a set of Internet connection anomalies, as well as a number of individual parameters for high-risk versions of browsers and devices.

According to Arvis Strods, Robocash Group CRO: “The JuicyScore service allows detecting at least 75% of applications with a high level of operational risk in the online stream. The data vector provided within the service allowed us to strengthen our decision engine both in terms of reducing bad loans and identifying potentially “good” customers, in cases where traditional data sources do not provide sufficient information. The data ROI is at least 15 times on the application inflow in the Russian Federation and a number of other markets”.

## [Robocash Group](https://robocash.group/)

An international financial group, founded in 2013, operating in the segment of alternative consumer lending and marketplace funding in Europe and Asia. The company is developing robotic financial services, providing consumer microlending to clients in Russia, Kazakhstan, Spain, Philippines, Indonesia, Vietnam and India, and operates its own P2P investment platform operating in the European Union and Switzerland. As of August 2020, it has a client base of 11 million borrowers, the total amount of loans issued for the entire period of operation is about $ 743 million. The holding independently develops products using Artificial Intelligence, Machine Learning and DATA-technologies to ensure accurate and comprehensive risk management, comfort and speed for clients and efficiency for business.

## [MFC “Zaymer”](https://www.zaymer.ru/)

One of the largest micro lending companies in the Russian Federation, which main business focus in online loan disbursement for PDL and Installment segments. Owns two fully automated lending services “Robot Zaymer” and “Robocredit”. According to the Expert-RA, [at the end of 2019](https://www.raexpert.ru/rankings/mfi/2019) Zaymer is the leader in the microlending market in terms of the of micro-loan volumes issued and the micro-loans portfolio volumes in the PDL segment. According to the National Rating Agency, the company's leadership in terms of the size of its loan portfolio has been confirmed as of [Q1-2020](http://www.ra-national.ru/ru/taxonomy/term/3296?type=rating). It is part of the international holding company Robocash Group.

## [JuicyScore](https://juicyscore.com/en/)

The company was founded in 2015, operates in Russia and 13 other countries around the world, provides a solution for assessing various types of risk for online business in the financial services industry. It has the widest stack of device authentication technologies on the market (now there are 29) - all technologies are proprietary and protected by patent laws in a number of countries. The technology stack based on proprietary authentication and device assessment technologies is one of the strongest in the markets where the company is operating (Our patents are available at links <https://new.fips.ru/registers-doc-view/fips%5Fservlet?DB=RUPAT&DocNumber=2691830&TypeFile=html>, <https://new.fips.ru/registers-doc-view/fips%5Fservlet?DB=RUPAT&DocNumber=2723679&TypeFile=html>).

Within the entire operations, the company has processed profiles of more than 3 billion devices around the world.

In times of crisis, when the markets drop, one of the key goals of any company is to maintain long-term and time-tested partnerships. This is especially important when the crisis affects a huge number of areas, industries, countries and macroregions, when the usual models and rules suddenly stop working, and there is a minimum of time to build new ones.

robots-min-1.png

Robocash: JuicyScore Service Reduces Operational Risks By 75%

We are pleased to announce the launch of the next generation of our JuicyScore product - APIv12\. The new generation is not just a serial number increased by one, it is a number of important changes to the product in terms of its information value and effect for our partners and customers. Our approach remains unchanged - we do not use any personal data and direct customer identifiers and continue working on Internet security for financial market.

Today we will tell you about the most significant updates to our product.

**Key changes and adds-on which will be introduced to our clients and partners within the API v 12:**

- An extended stack of **device authentication technologies**, we have added 5 new technologies recently and now there are 28 of them, which makes our device fingerprint even more stable in time and much more resistant to various anomalies. This is especially important for highly competitive and high-risk markets.
- We have significantly extended the range of detected **randomizers** \- more than 270 algorithms and techniques aimed at identifying software of this type are reflected in the new attributes, as well as the supplemented Index variables (IDX) of our data vector and score model.
- **17 new attributes** in the data vector - an extended set of **behavioural markers** on the applicant fulfilling an application for a financial product will help to identify high-risk segments more effectively, as well as a number of new **parameters for a network connection** for a better assessment of the Internet infrastructure and the risks associated with it.

**In additional to that:**

- **Improved stability and information value** of the most of the Index variables (IDX);
- **Extending the list of IDX1 stop factors** \- supplemented by a number of new markers for device randomization;
- **The increased information value** of the whole data vector by 25% compared to the previous version of the API.

This time we paid special attention to the score model. We managed to achieve a significant increase in predictive power by 25% compared with the previous generation model. The model is still calibrated for several reference values ​​(recommended cut-off levels depending on risk appetite), and a number of new behavioural factors made it possible to significantly increase the detected high-risk segment even after all stop-factors have been triggered.

We will not stop our product evolvement and also plan additional updates later this year, which will reduce risks and provide better understanding of online users behavior without using personal data and direct identifiers.

Despite the crisis and instability in the world that all players of the financial market faced in the first half of 2020, we keep investing resources and time into developing technologies and new data attributes, strengthening our expertise in digital risk management in order to help our customers and partners in their day-to-day activities in fraud prevention and risk assessment. Thank you all for your support and new ideas!

Do not hesitate to contact us via [info@juicyscore.com](mailto:info@juicyscore.com) and provide any feedback

Sincerely yours, JT

We are pleased to announce the launch of the next generation of our JuicyScore product - APIv12. The new generation is not just a serial number increased by one, it is a number of important changes to the product in terms of its information value and effect for our partners and customers. Our approach remains unchanged - we do not use any personal data and direct customer identifiers and continue working on Internet security for financial market.

12-3.png

Remote work should be productive - we launched a new version of API v.12 working in a home office format

Today, when technology is developing rapidly, and people are more and more immersed in the world of mobile devices, there is more and more talk that traditional service providers, especially financial ones, will give way to young and more mobile fintech companies (online lenders and neo-banks). In addition to fintech companies that have already become part of the financial services world, we should not forget about large Internet companies, both global (Google, Amazon, Facebook and Apple, so-called GAFA), and local, which have a large share in each separately selected country or group of neighbouring countries, for which financial services are not the main and significant source of revenue. These companies have accumulated a huge client base and an array of information for each client, and they already offer various financial services that were previously considered traditionally banking.

For example, less than a year ago Apple announced the launch of banking card (<https://www.apple.com/newsroom/2019/08/apple-card-launches-today-for-all-us-customers/>) based on MasterCard payment system and in partnership with the bank Goldman Sachs. Virtual card may be used in the ecosystem of Apple devices and the Apple Pay service, the plastic card issued and linked to the same account may be used at offline points of sale. Similar steps have already been taken by another Internet giant - Amazon, which launched a line of financial products. (<https://www.cnbc.com/2019/06/10/amazon-launches-a-credit-card-for-the-underbanked-with-bad-credit.html>). Not far behind, Google, announced a partnership with CitiGroup (<https://www.coinspeaker.com/google-citi-bank-big-techs/>).

In the Russian market, banks and Internet companies also have a rather long history of partnerships, including at the level of the chartered capital.

Earlier in our articles, we have already mentioned what are the key differences of banks and fintech companies in different aspects of their activities. Today we will take a closer look at what the future of online lending may look like in the competition and partnership of fintech companies and banks, especially during crises and other force majeure. We will also look at the role of large Internet players in this process and the place they can take in the financial system.

**What are the basic differences between banks, online lenders and large Internet companies, what are the advantages and disadvantages of each of these categories? Let's consider some of them.**

**Technology**

In this case, technologies are understood as the possibility of using, adapting, and introducing a modern technological stack in the operations and productive systems, i.e. real use of these technologies for a wide client audience, and not just in the framework of tests and pilots.

Since fintech companies usually do not have a layer of legacy systems, heavy and expensive to maintain and upgrade, their technology stack in productive systems is more modern and flexible, especially in terms of adding new functionality.

Banks, many of which have been operating on the market for dozens, and some even hundreds of years, operate on older platforms and infrastructure, which are subject to increased regulatory and information security requirements. These constraints reduce the ability to quickly introduce and adapt new technologies for key processes, for example, the process of making a credit decision.

The technological stack of large Internet companies allows, on the one hand, to quickly implement various client services and products, and on the other hand, it is not intended to provide traditional financial and credit services, since the architecture was originally designed for the provision of services of a different type.

The technological flexibility of fintech companies makes it possible to better adapt the product line and the business itself to various constraints and innovations, which can not only remain competitive in a crisis, but also increase the client portfolio.

**Audience for lending and other financial services**

Fintech companies allow getting access to funds for a wider audience of borrowers. They can lend to borrowers with a higher credit risk in terms of traditional banking risk policies, as well as an audience with moderate or low credit risk, which is difficult to prove their reliability and creditworthiness through traditional data sources. Here we can also talk about the segment of individual entrepreneurs and small and medium-sized businesses, which may require borrowed funds to close the cash gap. Having no large operating expenses and staff, they have the opportunity to offer financial products on better terms.

Banks traditionally lend to borrowers from more premium segments, the risk of which can be assessed using traditional data sources and verification procedures. In addition to that, banks are limited in their ability to take a higher credit risk due to strict requirements on interest rates and reserves.

Internet companies rarely lend to their customers, financial or quasi-financial services are limited mainly by the infrastructure for making payments, in some cases by issuing various payment cards (pre-paid, debit, credit) to pay for purchases, receive cashback, withdraw cash and money transfers.

In times of crisis, when banks usually seriously tighten their policies and requirements for new customers, fintech companies that are not spoiled by premium customers gain access to a whole segment of borrowers with good payment discipline, but no longer meet the more conservative requirements of banks. By providing loans, a better level of service and more individual conditions for the products such customers can stay as fintech company client for a long time.

**Channels**

Fintech companies are, first of all, a website or a mobile application, which are usually quite simple and functional, providing the simple process of financial service (for example, a loan) application fulfillment. The online office of a fintech company or neo-bank is open 24 hours a day and 7 days a week and is accessible from anywhere in the world if Internet is available. The customer journey, which begins via a tablet or mobile phone right from the couch, is usually more pleasant and easy. In many cases client can do everything on his/her own without getting assistance from the company's staff. In addition, a number of online lenders and all neo-banks issue payment cards, which allow using their services in the offline world.

For a bank, a web-site is a door to a catalog of services offered by the bank and its partners and a list of addresses of the nearest bank offices; you rarely see the opportunity to fulfil an application for a service for a non-client. The mobile application is also focused at the existing client, it can provide a great functionality and a set of services and payments, but for non-customers the application’s functionality is very limited or even absent.

Internet companies provide services through an online channel based on personal accounts in their services and / or social networks, in some cases through the use of native software and / or devices, they do not have offices or branches in the traditional meaning, which makes it easier to obtain services for customers. On the other hand, there may be restrictions on receiving services (for example, Apply Pay works only on native Apple devices).

The ability to receive services online, especially in the current situation that we have seen in the world since the end of 2019, can seriously change the balance of power in the market. In such extreme conditions, online becomes not just a convenient way to attract customers and provide services, but a condition for the survival of the company and the ability to lay the ground for further growth.

**Credit application decision process**

The speed of decision-making on new customers for fintech companies is much higher than for banks - due to the large amount of data collected, internal and external, due to the real use of powerful technologies and analytical algorithms, it is possible to quickly provide a personalized product and provide a high level of service. This factor is especially important for short loans (for example, the so-called payday loans or PDL), where a borrower needs a decision on a loan and money almost instantly. This also includes the online POS-lending segment for the purchase of various goods, equal payment plans, online instalment loans and online insurance of various categories.

Many banks today offer loans online with a minimum application review time, including pre-approved offers, but this is usually available to existing customers with a client history. For banks, an online channel is the way to access their own client, a shorter way to the services offered, and much less often is a channel for attracting new borrowers. Even if the borrower contacted the bank via the Internet, he would still be sent to the traditional offline decision-making channel, which lengthens the customer journey. In addition to that, banks are required to make a personal identification of the borrower.

Internet companies are not widely presented on the lending market, offering a more convenient infrastructure for making payments and purchases on partner web resources, so in this case there is no need to talk about some kind of decision-making process. Nevertheless, Internet companies have enough capabilities to design a quick decision-making process, since they possess the necessary technologies and a large set of data and knowledge about the client.

**So is everything so bad with banks and will they really give way to fintech companies and Internet giants and what can help the banks to compete with both types of companies?**

Regulatory requirements and restrictions imposed on banks are also their advantage - they are more resistant to various crises and periods of uncertainty, since each of the banks is under the close supervision of the regulator, and requirements on the capital size and sufficiency therefore they have better chances to reserve the client base, especially during the crises.

Considering the product line, fintech companies mainly specialize in one product where they can provide close to ideal customer journey. Internet companies can offer a wider range of financial services, but have a number of regulatory and technological restrictions on basic financial services. The bank offers a wide range of products and services in one place; traditional banking products remain and will remain the prerogative of banks only - they have a license and the necessary and expensive infrastructure to provide such services. In this case, banks will have access to “long” and cheap deposit money, which means that they will have the opportunity to issue the “long” loan products, which allows them keeping their clients for a very long time.

The third important advantage of banks is the presence of a long history and reputation, which many fintech companies have yet to develop. Reputation allows attracting a new audience in good times and survive in difficult times. There are some cases and a number of countries where large Internet companies also have a high level of reputation and customer confidence due to greater flexibility and a higher level of service, which in combination with the product line and services development may allow them to compete successfully with banks.

 
There is also a number of other criteria in the table below:

![](/uploads/image_6bbdfb8587.png)

Possible scenarios for the development and interaction of all three types of companies might look like:

- Fintech companies will be able to compete due to further development and implementation of new technologies and data sources, improvement of customer service, as well as attention to operational metrics and regulatory requirements - all this will increase their stability, improve reputation and will be a driver of customer base growth. A higher speed for ecosystem development will be one of the key advantages for fintech companies in the nearest future.
- Banks will actively develop their own technological base and build partnerships both with fintech, and even more so with large Internet companies. Partnership with fintech will allow increasing the flexibility of the decision-making system, faster launch of personalized products and providing more convenient interface to the services; Partnership with Internet companies will allow access to a new audience and higher penetration rate of banking products. At the same time, there will always remain a segment of traditional banking, which is less dependent on technological innovation and prefers personal communication in the office of a financial organization, especially in the segment of premium banking services and a number of other client segments.
- Internet companies already have enormous technological and financial potential, they have sufficient capabilities to compete in the financial services market; their further growth in this market may be associated with increased availability of their services, regulation, especially regarding the use of accumulated data arrays which is not always transparent to clients. Partnership with banks will be one of the ways to resolve issues of regulation and licensing of their operations in local markets. Increasing customer trust will allow to consider such companies not only as an ecosystem for convenient payments and purchases, but also as a substitute for traditional banking services.

Crisis situations that occur with a certain frequency in the world affect the players of the financial world in different ways, and the crisis is an excellent indicator of how the companies operate and develop, the quality and contents of services and the way how they are provided. The fair balance of high-quality and convenient services, technologies, and reputation will allow any company to withstand crisis situations and develop at a double pace in calm times, and attention to the needs of customers during a crisis will strengthen their loyalty and trust.

The future promises to be extremely exciting and interesting ...

Today, when technology is developing rapidly, and people are more and more immersed in the world of mobile devices, there is more and more talk that traditional service providers, especially financial ones, will give way to young and more mobile fintech companies (online lenders and neo-banks). In addition to fintech companies that have already become part of the financial services world, we should not forget about large Internet companies, both global (Google, Amazon, Facebook and Apple, so-called GAFA), and local, which have a large share in each separately selected country or group of neighbouring countries, for which financial services are not the main and significant source of revenue. These companies have accumulated a huge client base and an array of information for each client, and they already offer various financial services that were previously considered traditionally banking.

bike.jpg

The future of online lending in the era of global digital ecosystems and crises - Banks & Fintech part 3

Privacy

In our publications, we have raised the issue of privacy and anonymity when working on the Internet multiple times, as well as the issue of access to various types of services. In one of our previous articles ([https://juicyscore.com/blog/en/security-vs-privacy-in-search-of-a-fair-balance-en/](https://juicyscore.ai/blog/en/security-vs-privacy-in-search-of-a-fair-balance-en/)) we wrote about why users tend to limit monitoring of their online activity. Today we will continue the conversation about anonymity and privacy, in particular, we will talk about cases where the anonymity of the user directly or indirectly does not mean additional protection, but becomes a tool for unfair behaviour and obtaining services illegally.

There is no secret that the development of technology gives companies and financial institutions a lot of tools, including user data collection. Big business traditionally tries to collect as much information as possible about visitors of their web resources (such as web-sites or mobile apps), information about their devices and various behaviour stats in order to improve their services, to provide security / availability and to provide the maximum number of services.

**Why do companies need this?**

Collecting a virtual user digital “trace” or digital profile allows to solve a number of problems: first of all, companies can differentiate the normal traffic created by the real client and the synthetic one; secondly, it reduces the risk of fraud and other operational risks, which are associated with the theft of personal data and digital identity; thirdly, it makes it possible to personalize settings, tips, offer details or to configure advertising in a proper way.

**What are the difficulties in building a digital user profile?**

In this thorough work of building the right mechanism for data collection, there are various difficulties that can be considered in two major aspects, which are essentially two sides of the same coin.

Since the parameters of the software pre-installed by the manufacturer on the device can be virtually identical for one device family, the set of parameters used to build the digital fingerprint should be as large as possible. This allows to secure differences between devices and some limited stability over time.

But if we look even more broadly, the biggest difficulty was born on the intersection of these concepts - a whole layer of software appeared, called device randomizers, the purpose of which was to impede the existing technologies for constructing the device digital profile and to show the device as new one every time. The more users notice an upward trend in the data collection, the more popular are all kinds of tools that give the user the opportunity to keep their behaviour partly private and completely anonymous. This also applies to actions on a particular resource, and even the fact of the user's presence on the site.

**Excessive anonymity: where is the red line?**

First of all, it is quite difficult to be sure in the privacy of the Internet in the age of information technology, although anonymity is achievable. Secondly, due to the use of certain technologies the experiences a lower speed of information downloading from various web resources and low online content rendering quality. Thirdly, such mechanisms of “anonymity” can be often in the “grey” legal zone.

The randomizers themselves can be conditionally classified as follows:

- according to the Internet connection parameters, which are randomized;
- by the method of robotization or copying the virtual users behaviour on the site and / or in mobile applications;
- by the method of randomization, when various anomalies are embedded into the digital profile of the device.

At first glance, the user's desire for excessive anonymity seems harmless, but in practice this opens up a whole layer of problems that lead to fraudulent schemes that are difficult to identify. In pursuit of unattainable privacy, which should not be confused with achievable anonymity, some unscrupulous users cross the line and, unfortunately, try to use other people's digital profiles to disguise themselves. As a result, this can lead to large losses incurred by a financial institution, which will suffer from the wave of a technically well-trained fraudulent attack.

From a technical point of view, a randomized device looks like new, as the digital profile will be tracked as different with each new entry to the site. This case, such device may be used for unlimited number of requests (for example loan applications), from some anonymous user, who in many cases pretends to be another user.

According to statistics, the share of devices with installed randomisers in the application stream is relatively small, it usually ranges from 0.1 to 1%, with possible dangerous bursts of up to 5% - 10% over a short time range.

The current statistics for different countries according to the share of devices with randomizers in the loan applications flow for Q1’20 is presented below.

![](/uploads/image_1_2234631830.png)

Such situation leads to two very serious consequences:

- the share of confirmed fraud risk (for example, loan default without payments and contact) is extremely high and ranges from 70 to 95%;
- at the moments of prepared attacks, the percentage of applications with randomized devices jumps from less than 1% to several percent, and tens of percent in some cases.

Thus, even the fact of using a randomizer on devices can already indicate a risk of fraud or a risk of loan default, which means it brings a danger both to a financial institution that does not want to provide a service to an unknown anonymous user with potentially unfair intentions, and to the users whose excessive suspicion and an attempt to anonymize their actions will become an obstacle to obtaining a financial service, a much needed loan or an insurance policy.

**What should financial institutions do in this situation?**

At the beginning of the article we briefly covered typical technological scenarios that are used to build a digital fingerprint. In order to build a digital fingerprint that is robust and stable over time and resistant to various anomalies, it is necessary to use a wide stack of device and Internet connection authentication technologies.

JuicyScore constantly monitors new randomization technologies and software in order to quickly develop and to implement respective techniques into our service and make it possible for our clients to determine these new randomisers and the level of their risk. For example, over the past month, about 10 new techniques for determining such anomalies have been into production mode.

We recommend financial institutions to conduct users digital profile multi-factor research through deep authentication methods which should not be limited to traditional and long-existing profile-building technologies and this will help to reduce the number of fraudulent activities. You should watch for reasonable anonymity, limit the use and processing of direct identifiers and real user personal data and constantly update your technological stack - this will place you one step ahead of the most sophisticated fraudulent schemes and attacks.

In our publications, we have raised the issue of privacy and anonymity when working on the Internet multiple times, as well as the issue of access to various types of services. In one of our previous articles (https://juicyscore.com/blog/en/security-vs-privacy-in-search-of-a-fair-balance-en/) we wrote about why users tend to limit monitoring of their online activity. Today we will continue the conversation about anonymity and privacy, in particular, we will talk about cases where the anonymity of the user directly or indirectly does not mean additional protection, but becomes a tool for unfair behaviour and obtaining services illegally.

monster--1-.jpg

Pursuit for unlimited user anonymity vs risks for the online world

The information is an extremely valuable resource in the 21st century. The amount of information on the Internet increases each year and various methods are used to process these data arrays. Today we will focus on three concepts: artificial intelligence (AI), machine learning (ML) and deep machine learning (DML). There are still a huge number of methods that are used by data analysis specialists and experts, and the purpose of this article is not only to explain the features of these concepts, but also to show successful examples how some methods can be applied to solve practical problems to prevent fraud risk.

Machine learning (ML) is one of fundamental areas of artificial intelligence, the idea of which is to find a pattern in the available data, and then to spread it to new objects. In other words, it is a certain set or sample of values, which used to “train” the algorithm, in order to further apply for solving various types of problems, for example, forecasting and classification.

Deep Machine Learning (DML) is a subtype of machine learning. Its main feature is the use of machine learning methods and neural networks to solve real problems similar to human ones. DML searches for deep intermediate relationships between factors. Each element of the found dependence should be checked for stability and could be used to solve the problem of the next level - a particular hierarchy of attributes that were obtained by one or another statistical algorithm is built in the system, and each new layer has data about the previous one. From a practical point of view, to solve the top-level problem, an ensemble of models is used, each of which solves one of the problems below in the hierarchy. For example, the task of recognizing a human face might be a combination of several tasks: determining the points of the face contour, determining the individual parts of the face, as well as the layout of the found face elements within a previously defined contour. Other related business tasks may include fraud / spam detection; speech / handwriting recognition; translation and imitation of many other human cognitive functions.

The third concept that needs to be clarified before is rare event or the so-called “anomaly”. The basic character of a rare event is not only that it has a low frequency of occurrence, which follows from the name, but also the fact that the occurrence of such an event is usually accompanied by significant consequences, both are positive and sometimes negative. The example of this event is a natural disaster of great destructive power. In the field of finance industry, this could be a case, when an event leads to a high risk of loan default or loss on insurance contract.

![](/uploads/scheme_1_1_9115abd234.png)

On the one hand, it is important for companies to be able to predict the occurrence of such events and use them in their models in order to avoid the subsequent risk. On the other hand, since rare events do not follow the normal distribution law (which by definition requires a representative sample of target events), modelling of such events is seriously complicated.

For example, if we design a linear regression on a sample of 500 observations, where only 5 target events are represented (i.e. 1% of the sample), the result is an equation where the coefficient for an independent variable will receive the first significant digit only in the fourth or fifth digit after decimal point, which makes such a model completely inapplicable in practice.

What could be done in this case?

Being one of risk assessment and fraud prevention solutions, JuicyScore uses deep machine learning algorithms to develop variables. Examples of such variables are Index variables (or variables of the IDX type in our standard data vector), which, on the one hand, extract useful information value from the factors underlying these Indexes, and, on the other hand, make it possible to level out data collection related issues and the insufficiency of useful values ​​of each of these factors. Indexes allow using the synergy of many such factors that can be used as separate variables, reflecting the anomalies of one aspect of the Internet connection, for further research.

How we achieve high information value of the Indexes:

- Building the certain combination of rare events, each of which occurs a small number of times and is insignificant from the point of view of classical Gaussian statistics, we manage to achieve the significance and statistical strength of the combination of such events;
- The variables included in the indices necessarily have a similar nature and transparent mechanics of formation, which allows, firstly, to achieve their stability, and, secondly, to reduce the cross-correlation of different Indexes.

**Index variables as an example of the practical application of machine learning algorithms**

We will clarify how we use machine learning methods using one of these variables from our classic set of output attributes, namely IDX4 - Connection details. This index describes anomalies associated with an Internet connection, attempts to manipulate its various features and parameters. The risk inside this index grows with the value of the index.

What is the nature of this index, and how exactly did machine learning methods help build dependent factors and prepare this variable? Let's start with a description of the subject area, which is schematically represented in the picture below:

_There is a device with which the user accesses the Internet. The device has an IP address, date and time of access to the network, geographical coordinates and a mailing address, a data set describing user behavior, and a large amount of other information. The device is accessing a certain web resource, for example, the site of a financial institution, which means that you need to obtain the IP address of this site, for example, find the nearest DNS server. The signal to the DNS server may not go directly, but for example, through one or more intermediate and / or proxy servers. Both the DNS server and proxy servers also have their own attributes, such as a time zone, geographical coordinates and another data set._

Each component of the subject area shown in the picture can be detailed and described with its own set of attributes, parameters, and the main task is not only to identify the mechanics of occurrence and features, but also to define the predictive power and degree of influence on the simulated event. For example, the probability of write-offs on a loan. For a deeper understanding, we will consider several parameters from the described subject area.

**IP address** is a unique identifier for a node located on the network. In this article we will keep in mind classical IPv4 - we have already considered the risks and possibilities of IPv6 [earlier](https://juicyscore.ai/blog/ru/ipv6%5Frisks/) and we still have to study the possibilities and anomalies of this type of IP-addresses.

One of the most common anomalies associated with an IP address is the mismatch between device and IP-address time zones. Why we call it common? Because, firstly, in a number of regions of our planet this situation is usual and may br treated as normal. Secondly, time zones may not coincide for objective reasons, for example, when synchronizing the device time zones and IP address time zone after a flight on a plane as this process can take up to a day. Thirdly, such discrepancy is a risk factor.

Time zone mismatch can occur from 0.1% of cases when this is an anomaly, and up to 15-20%, in those regions where this may be common. In any case, the presence of this factor almost always indicates a risk that needs to be taken into account and to be added as one of the factors to the risk engine, depending on the risk appetite.

The other common feature of the Internet connection is the presence of a **proxy-connection**, when for one or another reason, the virtual user hides his location or uses a non-transparent Internet infrastructure. Such events occurs from 1% in the Russian Federation to tens of percent in the countries of South-East Asia. In cases where there is not too many of such events, this most often indicates the user's deliberate actions to hide their activities from the owner of the web resource either lender or other financial institution. Otherwise, when a number of such cases is significant, the user often does not even know how exactly he goes to the Internet and how the network connection is built. From a practical point of view, the first group of cases is the most interesting, since it indicates the presence of a small or medium risk, if this is the only anomaly, and the presence of a high risk, if this anomaly is amplified by a number of other factors.

The third example, which in some cases might be considered a connection anomaly, are the **DNS server** settings. DNS is used to resolve all common web resource names like www.juicyscore.com into IP addresses. Most often, a host accesses the nearest DNS server or servers located to it, until a server is found that has information about which IP address corresponds to the entered web resource name. But it often happens that the closest DNS service for some reason is not quite the closest and can be separated from the actual IP address of the device by state borders and a certain number of time zones and hundreds of kilometers. Depending on the country, the proportion of such events is also quite small, usually no more than 3-5%, but in cases where this anomaly appears, it indicates moderate or even high level of risk, especially when accompanied by other anomalies.

What do the three examples have in common?

- all of them relate to different characteristics of an Internet connection and they are united by common mechanics and the same subject area.
- all of them are rare events - in most of the cases, the share of each of them in the applications inflow for the financial institution’s web resource is lower or significantly lower than 5%; this usually means that such events are difficult in use or modelling and Classical Gaussian statistics (e.g. linear or logistic regression) based methods are not applicable to them;
- all of them significantly affect the risk of fraud, as they are often associated with forced and deliberate actions of the device user, and do not happen by accident.

On the one hand, these factors are extremely important due to their nature, and on the other hand, it is difficult to use such factors for credit engine rules and settings automation. Now imagine that there are not three such factors, but five, ten, or even several dozen. And each of them is a rare event with a large risk component inside. What to do in this case?

In this case machine learning methods such as both classical interpreted methods (for example, decision trees or logistic regression), and uninterpreted methods can be used. We are at JuicyScore prefer using interpreted methods, because, firstly, in our opinion, they are more applicable to forecasting financial risks, and secondly, they are more resistant to the peculiarities of the data themselves that are used for modelling.

And applying deep machine learning approaches

- We have obtained a number of stable factors that describe individual parameters of the Internet connection.
- Each of these factors is independent and can be used separately.
- The combination of factors allows us to solve the problem of a higher level of the hierarchy - determining the risk of fraud in the flow of credit applications.

As a result, when using combinations of the methods stated above, a variable is obtained that has a distribution of this kind.

![](/uploads/IDX_4_distr_1_5c0d2ab128.png)

The blue histogram in the graph shows the distribution of the proportion of observations by the values ​​of the IDX variable, and the red line shows the relative risk level (BR, or bad rate) in each cohort.

As can be seen from the graph, using machine learning methods, it is possible

- to make significant cohorts in terms of the number of observations in each and a share of the total number;
- to achieve monotonous distribution of the target variable level - in this case, the FPD level;
- to use the variable for rare events risk modelling, and for credit risks modelling, which are continuous, rather than discrete.

And here is how the distribution of the index values ​​looks like depending on the applicant country.

![](/uploads/IDX_4_country_split_1_1b60f2d441.png)

The graph shows that the proportion of observations where there are minor (cohort 1) and significant (cohort 2+) Internet connection anomalies is highly dependent on the region. It is not surprising that in Vietnam, where due to Internet infrastructure features and the lack of IPv4 addresses, the share of these cohorts is higher than in Spain, where there are much fewer such events, and the lack of IPv4 addresses is not felt on such a scale. What these two countries have in common: the presence of anomalies complicates the analysis of the device and, subsequently the analysis of the applicant, and therefore it increases the probability of risk.

Machine learning methods are a powerful tool. When the nature of the data is clear the correct modelling methods may be applied to solve problems, such as various kinds of risk forecasting:

- to identify dependencies which are not clear and obvious;
- to allow them to be brought to a form that makes them suitable for use in various kinds of models embedded in the credit engine or in actuarial calculations;
- to allow achieving stability of distribution over time and interpretability of the results.

Since any subject area is a complicated system that can be described by an infinite number of factors, parameters, metrics and attributes, the process of new useful factors and features extracting is a non-stopping exercise, and the number of indexes obtained in practice is infinitely large, which means that there is the opportunity to keep adding the information value to of our service.

The information is an extremely valuable resource in the 21st century. The amount of information on the Internet increases each year and various methods are used to process these data arrays. Today we will focus on three concepts: artificial intelligence (AI), machine learning (ML) and deep machine learning (DML). There are still a huge number of methods that are used by data analysis specialists and experts, and the purpose of this article is not only to explain the features of these concepts, but also to show successful examples how some methods can be applied to solve practical problems to prevent fraud risk.

Cover-1.png

Deep Machine Learning: on the path to the truth

_With the development of technologies, human life is accelerating in all areas: we have to process a lot of information on our way to the office, while waiting in the queue for a cup of coffee and even while reading a book to your child. We are precipitately switching to mobile devices to keep up with our working processes and we turn on our PCs less and less often. Read in our new article how to maintain sales on the same level or even increase them and reduce risks while living at this tempo._

## Trends in numbers

  
Over the last 5 years, the number of active Internet users has grown from 3 to 4 billion people; half of them are from Asia. At the end of 2019, the International Telecommunication Union published its annual “Measuring digital development. Fact and figures” report which says that there are 4.1 billion Internet users (53.6% of the world’s population) at the moment. The annual growth is estimated at 5.4%. The number of active mobile Internet users rapidly increases, with an 18.4 % year-on-year growth per 100 inhabitants. An important characteristic of Internet development in many countries is the systematic conversion from fixed connection to mobile subscriptions. It has been noted that before the appearance of smartphones, there were not so many households with Internet access in the world. “In recent years, however, more households in many countries have had Internet access than computers. This is because computers are no longer necessary to connect to the Internet, and many people connect using devices such as smartphones”, says the report. The Russian Ministry of Communications confirms this statistics: Internet traffic in Russia was 7% bigger in January—September 2019 than over the whole year of 2018.

Mobile device manufacturers fit up their gadgets with ever dishier stuff, and the users are steadily migrating towards mobile use only. This is why stable sales of many services can be maintained only supporting the dual system, when the web resource retains its broad functions and a handy mobile app is being developed at the same time.

### Web vs. Mobile: Who Wins?

  
 On our example we can only say that web and mobile channels are perfectly complementary, which means we should develop both directions at the same time. Using both options we help financial institutions and business community develop more rapidly. We could find the golden mean even for those clients who only use our mobile app and have come up with a unique solution: we recommend using our mobile libraries and install them in their native applications. This will allow processing a lot of data on the user’s device in a safe way to seamlessly evaluate the risks and gloss over better information value. All these functions are also available for web users, since we are present in both channels. This allows us to take care of our clients whatever platform they use.

### Where to focus?

  
 At the moment, the analytics do not expect complete rejection of fixed traffic. First, web resources have broader functionality; they can reach a wider audience, do not depend on the device, let easily update the information and manage how it is displayed without requiring any action from the website visitor. Apart from that, web resource is a cheaper solution in the terms of creation, development and support and tends to be less spending.

The mobile application allows identifying a potentially more loyal audience with lower risks, actively managing customer retention campaigns, but at the same time it can demand a sufficiently large number of actions from the potential user of the service and reduce conversion into sales. Moreover, the application has a much better multiproduct support. In addition to that, mobile application provides higher level of web-resource security as it may partially work autonomously from the main servers.

### How we can help in this changing world?

  
 Here we could cite a lot of theoretical examples but we would better talk about practice, that is, our experience. JuicyScore helps financial institutions to evaluate risks using both web and mobile channels. Let’s take a closer look at the working mechanisms. Our API is unified and identical for both web and mobile channels in regard of data collection and sharing, which simplifies integration and analysis of the data received, and its implementation into the decision-making system. We support cross-platform: we have created libraries for native applications of our client financial institutions for the two main platforms, Android and iOS, and we keep them up to date. Data collection via the mobile application is safe and corresponds with Apple and Google policies. While processing user device data, we do not analyse the content of private messages, SMS, or contact list. Data are aggregated across the entire stream: we identify the device and the user and provide a stable fingerprint regardless of the channel the applicant used to request the service. All risk factors, frequency characteristics and metrics are calculated across the entire stream rather than single channels.

### Our SDKs give a number of opportunities for native applications of financial institutions:

  
1\. Use of a broader stack of device and user authentication technologies, which allows to authenticate the device and its reference with one or more virtual users more accurately and thus reveal high-risk fraudulent activity (e.g. on connected devices, repeated devices, mass fraud cases);

2\. Analysis of applications installed on the user’s device: information about particular applications installed on the mobile device without registration might indicate higher risk of the applicant, while registration in one or several apps usually gives positive impact on the risk level;

3\. Possibility to identify and analyse more specifically technical characteristics of the device and the user’s behaviour related to it (for example, accelerometer and screen usage patterns may speak for some tinkering with the device and technical fraud or on the contrary, for some positive traits of the applicant and lower risks);

4\. Use of geolocation information within the limits of accepted precision (100-1000 meters and with higher error): the collected data is enough to build up a set of risk indexes related to the location of the device itself and also the parameters of its network environment. The information does not become personal as it is not enough for unambiguous identification of the personal data subject.

In the modern world, the diversification approach has been applied for a long time already in many areas. In fact, we observe the same process in the digital world. Companies that offer their services only via mobile applications are acquiring skills of web management, expanding their functionality, and those working in the web area are trying to compress the range of their functionality to the quick and handy mobile interface. All this is done for the sole benefit of reaching ever larger audience, retaining the existing customers, and increasing the number of clients.

With the development of technologies, human life is accelerating in  all areas: we have to process a lot of information on our way to the  office, while waiting in the queue for a cup of coffee and even while  reading a book to your child. We are precipitately switching to mobile  devices to keep up with our working processes and we turn on our PCs  less and less often. Read in our new article how to maintain sales on  the same level or even increase them and reduce risks while living at  this tempo.

VS_2-1.jpg

Web&Mobile: How to Use The Synergy to Boost Sales and Reduce Risks

Finovate

In November 2019 we were demoing at Finovate for the third time – it was Finovate Middle East in Dubai. Similar to previous Finovate conferences which we had honour to be invited in 2018 we keep showing non-personal data efficiency and secure use of biometry for authentication and risk evaluation.

Within last 3 years we have been evolving the ideas of end-user security and using non-personal data and discovered that in many cases this data is sufficient not only to make decision on financial product application but also to build the elements of online application risk monitoring system for both web and mobile channels.

Launching our biometry solution last year we were able to enhance device authentication algorithms and to reduce the “grey” zone in decision making process. This time we decided to focus on those who are on the front line of fight with fraud and high risks and to furnish out with usable interface to conduct online risk evaluation and monitoring.

Today, we are pleased to introduce Risk Management as a Service, which is a quintessence of technology for extracting and processing non-personal data on device, user and user behaviour, secure biometry and multifunctional user interface.

We invite you to the first demonstration of our Risk Management as a Service platform to share this important moment with us!

The product prototype we presented represents a practical implementation of our principles and technologies, combines the three most important components:

Already familiar to many of you, the necessary set of libraries and SDKs for collecting data about the device, Internet connection and user behaviour in the web and mobile channels;  
Open API with a high-informative data vector for model design and risk evaluation and  
The client’s personal account, where there is the necessary functionality for the practical application of the obtained data and the design of the risk management process and solving the problems associated with it, which the risk manager or underwriter faces in everyday work.

Three and a half years ago, we opened a new chapter in digital risk management and during that time showed that the online financial world can be made safer by refusing personal data and significantly reducing risks for all its participants; and many thanks to our partners and clients who share our principles and succeed, including with our help and support

In November 2019 we were demoing at Finovate for the third time – it was Finovate Middle East in Dubai. Similar to previous Finovate conferences which we had honour to be invited in 2018 we keep showing non-personal data efficiency and secure use of biometry for authentication and risk evaluation.

finovatmiddleeast-copy-2.png

Finovate Middle East 2019. Risk Management as a Service prototype demo

The modern world has moved online since years, having extended and in some spheres even completely taken the place of the offline reality. We can now see an interesting line of thinking in some contemporaries’ discourse — if you cannot find any information about something on the Web, for most people it will mean this thing does not exist at all. 
For many of us, the search and easy finding of a product or a financial service on the Web starts by touching the browser icon on our smartphone, yet few people ever think about the way the request goes, which server logs the information about it remains in, which DNS server would open the way to the required web resource, and all that happening behind the scenes. Even less people think of the risks that may lurk in wait during Internet surfing.

The evolvement of the modern Internet goes hand in hand with the TCP/IP protocol implementation, where IP address\* is an obligatory parameter. Let’s have a dip into the history. The good old IP address or IPv4\*\* in the 256.256.256.256 format appeared in 1983, and back then everyone thought that 4.5 billion of unique addresses would be entirely sufficient for all and for any purpose. That’s why in the 80-s many organisations got more addresses than they needed. 
Unfortunately, system miscalculations by evaluating the number of hosts on the Web and ineffective assignment of the addresses led to IPv4 address exhaustion. In the table below, we give a list of countries with population number to the share of dedicated IP addresses.

![](/uploads/IPV_6_eng_ea3851893f.png)

As you can see from the table, only the UK has enough IPv4 addresses per population; the other countries are rated with < 1\. Yet all the EU countries have a fair rate of 0.5 to 1\. Abnormally low coefficients mark the countries of Southeast Asia, which is though due to the fact that the drastic economic growth, population increase and a dashing digitalisation have begun there but 35 years ago. 
Already in the first decade of the 21st century, the IPv4 address exhaustion narrowed down the access to the Internet, which caused stagnation in improvement and digitalisation of services, including financial ones. That made the developers look for solutions. For example, big companies in the countries with the most severe IPv4 address exhaustion have found a practical solution in supporting the IPv6 deployment.

On 6 June 2011, the World IPv6 Launch and a successful IPV6 testing took place. The IPv6 addresses are 128 bits and contain a colon as separator (e.g. 2001:0db8:85a3:0000:0000:8a2e:0370:7334). The leading zeros are usually omitted. A string of zeros can be omitted too and replaced with a double colon (e.g. fe80::1). There can be only one such omission in an address.

The main advantage of IPv6 is that this system creates a sufficient amount of numbers not only to resolve the crisis with the deferred IPv4 addresses, but also to prevent any other possible crisis in the future. That’s why the professional community believes the future belongs to the IPv6 and not the existing NAT and CIDR technologies which are currently used to overcome the shortage of IP addresses.

Should clients, Internet service and content providers completely switch to IPv6, there will be no need in NAT since all devices can have globally routable addresses. This will allow applications intercommunicating directly (there will be no need in solving the problem of NAT bypass for collaboration apps and other systems). IPv6 usage lets partly reduce the network infrastructure.

The IPv6 addresses are now used much more often: only according to our information, in 2019, the traffic share with IPv6 addresses was at least 2–3% in the EU and Russia, on the Philippines it was 15%, in Indonesia – 20%, in Vietnam – up to 40%. About 15% of the global web resources have switched to the IPv6 system. 
https://w3techs.com/technologies/details/ce-ipv6/all/all

The problem with the IP-shortage now seems to be solved, but is it indeed so?

Speaking about the advantages of IPv6 addresses, we cannot leave aside the difficulties coming along with the implementation and usage of this standard, first of all, the high costs of IPv6 transition and impossibility of an instant switching to the new format which demands support of tech stacks and infrastructure supporting both formats.

Furthermore, IPv6 and IPv4 networks are not directly intercompatible; there are no URL mapping rules either. Although the problem is partly solved by tunnelling IPv6 packages within the IPv4 packages, it is still a long way from a complete solution.

Another obstacle to a complete IPv6 transition is the lack of auxiliary dictionaries and lists of high-risk IP addresses and geolocation data; there are no sufficient technical protection tools for Internet connection and network infrastructure objects, ranging from simple DDos filters to more complex tools. As a result, it is much more difficult to protect yourself from various threats in IPv6 networks, as it is often impossible to recover or replicate the risk management and protection technologies which became common and are widely used for the IPv4 format.

This problem is particularly compounded by the fact that almost the entire volume of IPv6 addresses has not yet been distributed among Internet providers. While every IPv4-address has its owner, which facilitates administration and risk response, as well as the possibility of investigation in case of any wilful actions, IPv6-addresses are almost in free access, which means they can be used for dishonest purposes and scot-free.

So, what do we have in practical terms? The new tool solving the problem of resource shortage and boosting the digitalisation of all and everything has turned out to be not so simple and neat, as it opens the floodgates to new risk categories and a widest range of both unscrupulous actions and potential fraudulent attacks that are yet to be learnt to detect and counteract.

We in JuicyScore, for our part, invest in the studying of this problem and have already included the minimum essential functionality for working with IPv6-addresses in the current APIv11: a technology stack for determining the real IP address in the IPv6 format and responding to other risks connected with its implementation.

If you wish to know more about it, please do not hesitate to contact us!

The modern world has moved online since years, having extended and in some spheres even completely taken the place of the offline reality. We can now see an interesting line of thinking in some contemporaries’ discourse — if you cannot find any information about something on the Web, for most people it will mean this thing does not exist at all.For many of us, the search and easy finding of a product or a financial service on the Web starts by touching the browser icon on our smartphone, yet few people ever think about the way the request goes, which server logs the information about it remains in, which DNS server would open the way to the required web resource, and all that happening behind the scenes. Even less people think of the risks that may lurk in wait during Internet surfing.

Ipv_1-3-1.jpg

IPv6: New Opportunities, New Risks

> “_Dear kid, this is Santa Claus writing. This year I cannot send you the presents because you haven’t signed the consent to personal data processing!_”

In July 2019, we had the honour to speak at the International Financial Congress at the consumer fraud and anti-fraud techniques section. The high level of the event and its guests highlighted that the topic is exceptionally relevant and important for the market. It looks like consumer fraud mechanisms in retail lending and microlending are a serious concern for both financial institutions and their clients.

While analysing typical [fraud](https://juicyscore.com/dangerous-trio/) patterns banks and microfinancial institutions usually face, we noticed that the most dangerous schemes with the most drastic consequences involve personal data compromise. Imagine if personally identifiable information comes along with sensitive data and other information: such a happy kit becomes a real Pandora’s box once a fraudster gets his hands on it, in no time escalating the negative effect and multiplying the loss for financial institutions up to dozens and even hundreds of millions of rubles in particular cases, and all of this in the microlending sector with an average loan sum of 10,000 rubles.

_What can be considered personal data?_

In the modern world, which includes financial services, there is a dazzling array of various data generated and processed with every operation we make. For example, services that are free of charge for users can sell their data across other sectors. Client data become assets, and an access to a huge pool of personal information boosts economic cost of such platforms.

By personal data is usually ment some information that can be used to identify a person and has a direct correlation to the person. Personal data refers above all to the full name, date of birth and passport details. Yet in the online world, other kinds of data can be used in order to get services too, financial ones as well – instead of the conventional information. This kind of data has not been seen as personally identifiable before.

As we see it, all contact information, such as phone number, e-mail, even when detached from the full name and / or passport details, should be qualified as personal data and require regulation of processing and access.

Why is that so? The reason is, this information is quite enough for a financial operation such as applying for a loan, and it can also be used for a two-factor authentication, for example, when we register on public services websites and others. Finally, it can be used by intruders to gain an access to a wider range of personal and sensitive data.

Do you remember in details what your e-mail box contains or how many contracts, bills, copies of financial documents, passwords have been sent there? Meanwhile, pass of such information into the public domain or the Deepnet can be downright unpleasant.

One more unpleasant consequence of uncontrolled personal data processing is their getting to unprofessional data operators. Everyone has at least once filled in application forms for a discount or loyalty card or registered at incalculable websites. But if we think about it, do the companies operating the information ever have the intention and ability to protect our data from getting to criminals?

_A spammer has been attacked by unknown. Several millions of e-mail addresses on the list of suspects._

Sensitive data does not always have an unambiguous correlation with the person, yet their request and processing can be hurtful. One of popular categories is disposable income: you can find a detailed analysis of this topic in our previous [article.](https://juicyscore.com/en/pti/) Such data should also be reconsidered and put on the list because it is often requested and used by financial institutions for different purposes.

Another issue has become burning for the last 5-10 years due to growing popularity of online financial services, which is data emerging at the moment of Internet connection and their transferring to the Web. This information can be of big value for purposes of evaluation of the borrower or financial service receiver.

Information about the device technical parameters and other characteristics, the so-called device fingerprint, is also ranked with the same category. Yet such information cannot be called personal because it doesn’t actually define the natural person. Hence its regulation, if required, should follow some other standards. Although it might be valuable, unlike personal data it is not enough to identify the person and perform financial operations on his/her behalf in case of data leakage.

Already now we should develop a wider view on the data which at the moment includes not only personal data, but also other categories such as sensitive and non-personal data that evolve in the online world and don’t identify the natural person directly. When the volume of information is booming and digital technologies are bouncing along, we should turn from generalizations to practice and start categorizing, classifying, clustering and regulating data by common sense, not pro forma.

If a too mild and insufficient regulation of personal data results in problems and loss for institutions and their clients as well, shattering confidence in the digital environment, an excessive regulation of all the information, even not personal, on the other hand, can push to the Deepnet and significantly hold up buildout of digital technology and online services. We witnessed that when the General Data Protection Regulation came into force, dramatically changing the terms of data collection and processing for data operators in Europe and for Russian companies operating with the EU citizens within their area of activities. Let alone that strict penalties affect economic growth of the businesses and digital economy in general not in the best way.

Summarizing the above, we suppose that it is necessary to strike a good balance between a safe, authorised access to the data and regulation and guidance to enable effective work of data consumers.

Is it possible to make this seeming no-win situation a win-win for all? The solution might be the differentiated attitude to the regulation of data processing, corresponding with the risk level within specific data categories and potential severity of loss which can emerge in case of leakage.

Ultimately, we believe that future looks bright ahead with the right vision and practice: new business models should remain customer-focused and back on both legal and ethical norms while operating data and analyzing digital portraits of the clients, and we are here to keep you on the safe side with right solutions.

binary_2-1-1-1.jpg

Is all data personal?

Legal

Previously, we have been covering mostly the banking peculiarities of a financial business and its regulations. Since beginning of July such changes have come into force directly concerning microfinance market and changing the operating conditions of the microlenders. As a partner of several market players on this field, we wish to share our vision and tell how we could be useful in response to the new upcoming challenges.

OOn July, 1 regulation changes will come into effect. The main part of them is about limiting the Pay-Day Loan interest rate to 1% per day. The rate has already been limited to 1.5% before, but the [the new regulations](https://www.rbc.ru/finances/24/10/2018/5bcf076b9a79477ca46014ca) will significantly reduce the previous limit.

The rate limiting will have an impact on the profits of the MFO business, which raises concern among established market majors. Keeping focus on the heterogeneous and complicated market since 2011 and being a mediator between borrower and microfinance companies, the regulator is seeking to establish the basis for a healthy, competitive and compliant with the modern market needs.

This will affect above all the market structure.

Rate reduction and regulation changes are always a challenge. Everyone will have to face the new difficulties. And we cannot know for now who will be affected more – small and smart young companies or solid established institutions. Yet the basic concern of the regulator is to repress and put out of business disreputable players, shadow lending market actors and market players with a poor business model, which will be the first step towards market stabilization.

Secondly, the interest rate limitation is about changing the financial business model.

Borrowers of higher risks on the one hand and disregard of control on the other have resulted in raising the rate in order to solve the problems. Putting aside some cases of excessed and unmotivated rates which were possible before the regulations came along, the interest rate is a combination of cost of funds, risks and operation costs. Now, when the rate is fixed and notably lower than usual (even compared to the 1.5% in the beginning of the year), the institutions will have to reconsider their operating activities, take serious their risk politics and reorganize their business to a great extent.

Thirdly, facing the market changes and changing its own philosophy, the business will have to confront with some changes in its brand and audience as well.

Microfinance companies are sought not only by those whose loan requests were declined by banks due to payment delays, but also by clients who cannot confirm their income in a traditional way, by self-employed and by young people, by residents of remote towns and settlements , by people with “thin” credit history, etc. This alone speaks for audience improvement, for these are no “fraudulent” clients but people who cannot correspond with the typical bank requirements or a standard customer portrait. Yet the general attitude to short-term loans and microlenders remains negative. Environment changes, new business models and evidential changes in positioning can help to improve public image of microlenders and attract wider groups of clients, also those of higher welfare level.

Gearing up for the new competitive environment with a new and hopefully active demand, in the context of the changing business model, we would like to suggest a decision aimed to lower the risks and operational costs.

How we can help to reduce credit risk:

We filter out credit shopping and search out factors affecting social default. We trace repeated and multiple applications sent to different market players within a short period of time, repeated use of the same device and real IP address, manipulations with the application data, etc.  
We identify good rules for low risk segments which are difficult to evaluate using traditional data sources. Instead of them, we rely on device, software, real IP and network analysis.  
We assess and verify disposable income using alternative data, such as evaluation of device quality, the Internet infrastructure and economic factors.

How we can help you cut down on operational costs:

We help to set auto-reject rules using filtration of 10–30% high risk inflow and help with manual verification allowing to carry it out more quickly. This is possible due to an accurate device authentication, randomizers detection and other factors of technical fraud, data manipulation, digital theft, synthetic applications inflow, etc.  
We reduce marketing costs and time on search for “good” borrowers, helping to find the best possible sources of potential clients who meet the business risk policy.

Changing of rules is always a difficult period of time. However, on the other side, it gives a chance to take a fresh look at the business, find new solutions and start using new tools. We hope to improve risk management of financial institutions and help to find new clients who will be good to work with during this new stage of market development.

Your Juicy Team

Previously, we have been covering mostly the banking peculiarities of a financial business and its regulations. Since beginning of July such changes have come into force directly concerning microfinance market and changing the operating conditions of the microlenders. As a partner of several market players on this field, we wish to share our vision and tell how we could be useful in response to the new upcoming challenges.

0619-Jump.jpg

MFO Regulations: New Market Challenges And The Ways of Working with Them

We are glad to introduce our new big API update launch. We have focused as usual on raising impersonal data informational value for better decision-making. We have significantly raised the service value using new data, new device authentication technologies and more powerful fraud prevention score.

Today we are going to tell you about the most important upgrades we’ve developed for our product.

What’s new?

API v.11 implements over 20 new predictors, among them:

- New IDX6: **evaluates the Internet infrastructure quality** using network connection quality and development rate of an applicant’s region analysis combinations;
- New IDX7: **evaluates risks** by analysing quantity, character and categories of the applications installed on the mobile device;
- New IDX8: **evaluates application data risks** and how frequently they are changed.

Moreover, we have broadened **the attributes list in the data vector,** such as apps structure on the applicant’s device, several behavioural markers of the virtual user, randomizer detector and many more.

Updates:

- **More stable and informative** IDX variables;
- **Extended list of IDX1 stop markers**;
- **70% more informative IDX5 device quality index** due to adding new parameters of hardware evaluation and evaluation standardisation for new regions, etc.;
- Several new techniques of **device authentication**;
- **40% higher prediction ability** of the score model.

Geographic growth:

- From now on we support some new partners in India;
- During the last year, we also gained some new partners in our current regions of presence;
- Data from all our current parents has been taken into account while developing new attributes and model upgrades.

Thank you for inspiring us to take a new look at data, improve our expertise in the field of the Internet and devices and refine on our product. We believe that in response to the growing competition and regulations getting more and more stringent, you will find our updates useful and helpful in gaining a stronger foothold in the market.

We look forward to your feedback. Please, do not hesitate to contact us at [info@juicyscore.com](mailto:info@juicyscore.com)

We are glad to introduce our new big API update launch. We have focused as usual on raising impersonal data informational value for better decision-making. We have significantly raised the service value using new data, new device authentication technologies and more powerful fraud prevention score.

0519.jpg

Some Good News to Keep You Warm This May: We Have Launched API v.11

_Last year Russia showed unprecedented speed of credit obligations growth. The total outstanding balance for all types of loans reached [14.9 trillion rubles](https://www.rbc.ru/finances/29/03/2019/5c9de8039a7947a07148827d). The speed of lending growth is still high which leads to new challenges for risk management. Debt refinance and consolidation programs offered by financial institutions and regulator’s actions aimed to control the debt burden are the evidence of general concern._

_Our today’s article will be about the debt burden, its estimation difficulties and how alternative data can help in the calculations._

**Debt burden** is the outstanding balance on all credit accounts. For a borrower, it is essential that this amount corresponds with his or her income. To measure this correspondence, debt-to-income ratio is used, commonly defined as DTI or PTI (payment to income)\*. Basically, it reflects how much debt the borrower can handle.

As experience shows, the conventional “normal” PTI ratio would be between 0.30 and 0.50\. At this range, the debtor would be able to cover a loan evenly, without cutting too much on the everyday spending. At the point of 0.50 and over, most conservative financial institutions would offer the loan at less desirable conditions or even reject it. The closer to 1, the higher the risks of both giving an irrecoverable loan and also pushing the borrower to bankruptcy. The whole burden of responsibility may then lie on the lender that has incorrectly calculated or evaluated the PTI ratio.

## Some recent problems about PTI calculation

Let’s say, there is a client who has applied for a loan and needs to provide information on his income.

The primal thing to do would be providing an income tax certificate, also known as 2-NDFL (or another standard document on the list). We have discussed though some peculiarities of the information indicated [there](https://juicyscore.com/loan-for-temper/): it can be insufficient or inaccurate. For a client who has obtained a loan before, it may be possible to calculate the new level of income taking into account some public data (e.g. adjusted to the official inflation rate). A payroll customer’s income is the most transparent, yet by far not every loan provider boasts a big portfolio of payroll customers.

If it is not possible to estimate the disposable income (for example, if the borrower is self-employed), it will be more difficult to make a decision which may even result in rejection. This would be an undesirable outcome for business yet this is what we currently see in practice.

Let’s consider the second element, which is the debt burden. There are some exceptional cases here, too.

The main source of such information on a borrower for now is the credit history that shows all current and closed credit obligations, as well as payment discipline. Credit history is the most reliable argument in decision making; the problem is that the information is updated within 5 working days. With his back to the wall, a client may overgrow with quick loans and debts, and the corresponding information may not only hold off, but it may stay behind at all. The consequences of such a burden will not be long in coming. The same effect may have social ways of raising one’s loan burden, such as borrowing from neighbours or friends. Information about such loans will never reach any informing or regulating institutions, yet the total loan burden will rise dramatically.

If we look at the credit bureau statistics, debt burden rate is one of the key factors for credit delinquency, even if the client has never violated his payment discipline before.

## What do we have on the whole?

There we have a client whose income may consist of declared salary, which is easily verified, and of undeclared and unstable payments as well. It is partially spent to cover regular or unforeseen daily costs. The debt burden may include both transparent legal loans and money borrowed on trust.

As a result, PTI calculation looks like an equation with two unpredictable unknowns, whereby calculating PTI is not only a need, but also a must for a financial organisation, and the regulator is getting more and more [concerned](http://expert.ru/2019/02/27/kredit/) about this factor

This is what we have as a problem when we come to estimate and calculate a client’s PTI:

- There is still no reliable information on the borrower’s income;
- The information on the loan burden may be irrelevant or incomplete;
- Due to the lack of 100% reliable data, the tool is far from perfect, while all the risks related to the decision will lie with the lender.

## What are the possible solutions?

We suppose that alternative data may be used in addition to the conventional information sources. It will help to estimate the client’s income and debt burden indirectly, and reconfirm the estimate given by traditional sources such as credit bureau. Working with alternative data, we offer a solution that allows making up for the lacked information and raise the value of the debt burden estimation.

We can tell more about the client’s income upon:

- Internet infrastructure quality evaluation that was used to send the loan application. The higher the quality, the higher the income, and vice versa. Therefore, if it is an infrastructure of a higher quality, the risk that the credit may turn out irrecoverable or paid with delinquency will be lower.

![](/uploads/PTI_1_eng_1_241890c6bc.png)

  
Graph 1\. Relative quality evaluation of an Internet infrastructure (from G0 to G4) compared to a relative delinquency

- Device quality evaluation that was used to send the application. Hardware quality, device year of issue and other parameters can be put together, which will give a rather high correlation with the income. If you add the period of time since the device has been associated with the user and his behavioural data, the reliability of this evaluation will be significantly higher (R2-level over 75%).

![](/uploads/PTI_2_eng_f55edd24fe.png)

  
Graph 2\. Relative quality evaluation of the device (from 0 to 5) compared to a relative delinquency

We can also fill gaps in the information distributed to the lender due to:

- Analysing loan applications frequency from the device and virtual ID that indicates the applicant’s need in a loan and recently opened credit obligations. The higher the applicant’s short-period demand for credit or loans, the higher the risk.

![](/uploads/PTI_3_eng_1_c5ced66fff.png)

  
Graph 3\. Relative delinquency depending on the number of loan applications during one week for a random user

- Cooperation with 80% of microlenders, credit unions and pawnshops on the market, which adds to the data on loan applications and improves the accuracy of the borrower’s risk estimate.

In combining this information with conventional evidences of income and outcome, it is possible to improve the accuracy of PTI estimate, making financial organisations safe from potentially irrecoverable debts and risk regulators, also preventing borrowers from excessive debts.

Your Juicy Team

---

[\*](https://juicyscore.com/en/pti/#%5Fftnref1) We prefer the international term “PTI” counting the borrower’s monthly loan payment in regard to the monthly income (a more conservative estimate is often used, when the income is calculated net of indispensable and essential costs). These terms are frequently used as synonyms, in reference to the PTI’s meaning, because debt-to-income ratio takes the total amount of the borrower’s loan obligations into account and is not so demonstrative, especially when the borrower has an open long-term loan of a large amount of money (such as car loan or mortgage loan).

Last year Russia showed unprecedented speed of credit obligations growth. The total outstanding balance for all types of loans reached 14.9 trillion rubles. The speed of lending growth is still high which leads to new challenges for risk management. Debt refinance and consolidation programs offered by financial institutions and regulator’s actions aimed to control the debt burden are the evidence of general concern.

0319.png

PTI: Practice and Alternative Data

Fraud

_We’ve been telling enough about the advantages of alternative data and risks associated with collecting conventional data on your borrowers. Today we will demonstrate how personal data usage can not only turn into risk and damage for the financial institution’s working process but also spell disaster for the whole business. We will draw on several cases from our experience, which are: Seasonal Aggravations, Fraudulent Brute Force Attacks, and the Black Swan. What are these risks and how to get protected?_

## Seasonal attacks

New Year’s Eve is a magical time when all our wishes come true. But to really make them true, a grown-up dreamer has to be a good employee for a whole year. If you want something more, you can always call on a lender, and your credit will be surely approved – after all it’s December, the most wonderful time of the year! We expect everyone to be a little kinder, and a credit to be given as easy as a present to be found under the Christmas tree.

To be serious, there is a strong consumers’ belief that big holiday period is the time when credit institutions are happy to approve a loan everyone who needs some spare money and asks for it. Spoiler alert: that is not true. Neither the seasons, nor the weekends or holidays have any effect on the approval range or the risk policy extent.

Yet the demand grows and the number of loan applications grows dramatically as well. As a result, during this “high season” you will receive 100 applications instead of the average 10, for example. Supposing that each tenth application would be fraudulent, we should say there will be ten times more of them than normally. For credit institutions, such a flow is extraordinary yet natural and casual. And as it is casual, it is possible that you will not focus enough on the growing risks.

Nevertheless, these seasonal attacks bring a lot of troubles due to the growing percentage of the “black sheep”: it’s been several years that we register an increase in the number of fraudulent applications at the end of December. The number of such fraudulent requests frequently grows by several times.

![](/uploads/1_2_580eb98141.png)

![](/uploads/2_2_44e7b30b8b.png)

  
Incoming applications risk level at a credit institution in December 2017 and 2018\. A significant growth of fraudulent applications of a high risk can be seen in the twentieth of December. At this moment, the institution is exposed to a higher share of fraudulent applications.

It can be seen on the graphs that the late December was loaded with credit applications on general and risky applications as well, both in 2017 and 2018\. What should a credit institution do in this case?

In order to neutralize such seasonal attacks, we recommend taking a more prudent look on the risk policy. The “seasonal” fraud is usually not of a profound background and most of the risky applications can be easily cut off, saving the business from collateral damage.

Basically, the most efficient way to protect yourself against this type of fraud is setting more rigorous cut-off levels after existing patterns. Of course, this will lead to rejecting some reliable clients too, but if we have to choose between safety and greed, it’s better to choose the first one. The missed profits can be compensated by offering credits with extended limits to a well-known and reliable part of your portfolio.

## Fraudulent brute force attacks

Let’s think of the DDoS attacks, familiar to anyone related to online lending. A cybercriminal overloads the incoming channel with an enormous number of applications in a short period of time until the system breaks down or the security tools kick in, cutting off the source of the threat on one or several IP addresses. In such a case we imagine one or several malicious computers which are the source of the attack and an army of zombie devices willing to destroy their aim.

When we talk about fraudulent brute force attacks, it becomes more complicated. Such an attack is launched from a big number of devices which are no “zombies” but have an individual IP and some supporting data preventing them from being cut off by conventional security tools in one operation.

They also aim at finding weak points in the applications screening system. Such weak points are requests to the data sources which make the basement for the whole system. Credit bureaus, telecommunications operators and other data suppliers give information valuable for the workflow and used in the credit decisioning.

Each supplier has its own limit on requests from a client (credit institution) to be processed. When the limit (usually the lowest) is overcome during an attack, the source is being backed up or falls out of the processing. A security leakage appears where the whole flow of risky applications swoops down to. The decision is taken at random then and the whole flow is being identified either as reliable which brings direct damage or as neutral and is being transferred to manual processing, which slows down the whole thing and draws in the number of credits approved, that is, some profits missed.

![](/uploads/3_37c2059702.png)

![](/uploads/4_2_56fa1f9859.png)

  
An average incoming flow of applications on clients (blue) and two financial institutions from Southeast Asia (orange), affected by a Fraudulent brute force attack in February 2019\. The number of risky applications grew by 3-4 times, which looks abnormal compared to the average level and out of proportion to the normal incoming flow at the beginning of the month.

How could we deal with such evil? The most efficient solution would be monitoring the risk level using fraud score detecting and blocking the most suspicious and active devices, analyzing application frequency from the IP address and subnet. Yet such a method implicates a constant control over the incoming flow by risk managers.

## The “Black Swan”

“Storing and using personal data is dangerous”, we keep telling this over and over again. Sticking to our today topic, we will demonstrate how dangerous it can be. Let’s think for a minute about the worst: another data leakage has occurred and it has end up in the black market. But this time it’s not only about ID details of a client, but all personal data, all the sensitive information, all we had about residence addresses, IDs, insurance numbers, tax reference numbers and any changes made to this data for several years. And it’s not about one unlucky client, but one hundred employees of an unfortunate company in a nameless town.

The situation is fictional of course, yet the threat is more than real. At fraudsters’ disposal, such a data set would be an absolute Black Swan, and the conventional data processing system has no solution against it.

![](/uploads/5_bc52b90341.png)

![](/uploads/6_30f242f7b2.png)

![](/uploads/7_1_e80ae84916.png)

![](/uploads/8_f8acf5ecd3.png)

  
Several “black swan” cases. Average clients’ figures (blue) against clients exposed to the attack (orange). We can see an abnormal grow of risk rating by 4 times.

Although consequences of such attacks are severe, they are not commonplace. It is possible to get prepared, bringing in extra tools of data and loan applications analysis and using poorly reproducible alternative data in the loan workflow.

Thus, detecting repeated devices, combined with cross-checking the data used in different loan applications sent from a same device along with analysis of large deviation markers could be useful.

We hope that the cases demonstrated in our article will foster clients’ determination to use alternative data and extra tools for loans approving and security. It’s useful to remember though that fraudsters do not stay behind in their methods as well, which means that we should always be on our guard.

Our service is a relevant tool to fight against fraud, in line with the today’s realities. It will help you to reduce significantly the consequences from the attacks described above. Yet how useful this tool can be depends on the man, that is, the risk manager who is familiar with modern security tools, literate in data use and understands the incoming signals, has a fast track to handling the threats and helps the business to develop in a most successful way.

**Take care of yourself!**

**Your Juicy Team**

We’ve been telling enough about the advantages of alternative data and risks associated with collecting conventional data on your borrowers. Today we will demonstrate how personal data usage can not only turn into risk and damage for the financial institution’s working process but also spell disaster for the whole business. We will draw on several cases from our experience, which are: Seasonal Aggravations, Fraudulent Brute Force Attacks, and the Black Swan. What are these risks and how to get protected?

0219-1.png

The Dangerous Three: the Most Recent Examples of Fraudulent Attacks

_How long have you not been online today?_

_Since 2007, humanity stood on the path of Internet addiction.The famous “We have reinvented the phone” sounded 12 years ago on the presentation of the first iPhone. That phrase marked the beginning of a new era. The attitude towards information as well as the entire service industry have been changed and reinvented alongside with the possibility of unlimited and permanent Internet connection. Everything happens online today as every even a little self-respecting company should have a solid and functional website, complemented by a mobile application. Wi-Fi-networks development, the connection quality, and speed improvement provide the continual contact with customers._

_But the wonderful new world brings up new problems, which started bothering both users and business every day. For example, few minutes of being online lead to user data availability in the web for a much longer time. As the result, users became focused on their Privacy. At the same time, financial institutions and the online business in general began paying more attention to their Security._

_These two things seemed to have simply improved the lives of both business and users. But pursuing their own needs, the financial institutions security and end-user privacy do not always correspond, and often contradict each other._

_Why is it happening?_

Let’s try to figure out what we mean under“Security”.

Financial services are closer than ever today. In the very beginning, financial institutions expanded their office network to almost every location. Within a number of years this approach gradually converted into expanding the business via online through developing fully operative personal online accounts. Any financial transaction can be made today via constantly improving mobile applications. Common day-to-day operations have become more intuitive and effortless with every newly appeared features of smartphones. We are assured that banking is easy. Those who do not not have at least one bank-app on their phones are left far behind the progress or at least complicate their lives.

As the result of this we have got the illusion that loans (as one of the main financial services provided) are easily accessible. You can get “vacation”, “car”, “smartphone” or some other advertised goods by just one finger snap. Active and attractive marketing campaigns demand attention from even an apathetic client who more and more often replies to such demand. In addition to that, the true hunt for user’s information is always hidden behind these bright images or clips. In order to successfully “offer” a loan, lenders are collecting any possible data regardless how much they really need it; their systems receive all necessary information with every credit application. Analytics and risk policies are based on every new personal and sensitive data requested via questionnaires and other similar stuff. In some cases personal data becomes a separate source of revenue and always becomes a source of additional risk for both the organization and the data owner.

Gigantic questionnaires and forms are compiled and used for data collection; all this information needs to be stored. As the result additional data increases the infrastructure overload, complicates internal and external processes and operating costs. Similar to a demanding child who needs thorough attention and care data security requires even more costs and restrictions. The attitude towards data becomes either pragmatic or extremely cautious; data storage becomes the most protected and inaccessible part of the business, hidden behind many locks.

The process is also complicated due to methods of data storage and security. Data hashing is thought to be one of the most common ways to provide data security. In this case data is depersonalized, converted into an alphanumeric code and stored in a hashed form. Seemed to be convenient and reliable this method may lead to data and business damage. At first, depersonalization algorithms (no matter how sophisticated they are) have a chance of a collision. For example, different data lines may receive the same combination of letters and numbers which will lead to an innacuracy of carefully protected personal data. Secondly, there is always a possibility of a reverse conversion even for hashed data. There is no need to be Turing to break this code. With basics in mathematics and server capacity is is possible[ to convert hashed data to personal again](https://juicyscore.com/en/security-vs-privacy-in-search-of-a-fair-balance/#).

Despite all efforts, the most successful and protected companies’ databases suffer from hacks and leaks. Unfortunately, recent scandals with [Marriott](https://juicyscore.com/en/security-vs-privacy-in-search-of-a-fair-balance/#), [Facebook](https://juicyscore.com/en/security-vs-privacy-in-search-of-a-fair-balance/#) and [Google+](https://juicyscore.com/en/security-vs-privacy-in-search-of-a-fair-balance/#) will be replaced by new ones very soon. Despite all protective measures and efforts taken to secure the data it remains a valuable commodity on the [shadow market](https://juicyscore.com/en/security-vs-privacy-in-search-of-a-fair-balance/#). Publicity of such scandals brings enormous company brand risks and makes people scared every time they hear about such cases. It becomes clear to users that they have to take Privacy in their own hands as personal data cannot be always entrusted to businesses and IT companies.

Human is being an owner by his nature. Being protective of their personal and sensitive data becomes just one of the aggravated behaviors, especially in the 21st century. Dislike for unnecessary questionings, suspicion of the Internet and great efforts to make the stay on the network as invisible to third parties as possible are the results of protective reaction and privacy thirst. People’s mood is also worsened by stories about the “dark side” of the online existence. A simple distrust of the open Internet transforms into excessive protective actions. My home is my privacy. Hundreds of private browsers, firewalls, anonymizers, VPNs and other services aimed to make online presence private (or create such impression) grew and disappeared within last 10 years. The TOR-browser, which theoretically allows hiding actions on the network, in fact, can lead to getting busted and feeling all the consequences of this “privacy”.

As a result, privacy subsequently became a barrier: turning location phone function off while calling an Uber is minor issue. Financial organizations begin to consider the clients at the stage of their visit on the web-site and everything they have to consider are the “guests” without any information about themselves.

Due to excessive privacy a potential client is considered risky by a financial machine, striving for “cleanliness” and openness. Organizations are not ready to accept additional risk acquiring users who hide themselves without any “understandable” reasons or some reassuring markers; company’s immune system triggers and declines a client at the application stage, baring a direct loss to the business.

Here is a paradox: end-users seek for Privacy in order to save their personal data and companies controversially have to increase the level of Security to protect the business. Everyone would like to receive better conditions for themselves. Both sides are not ready to meet each other’s needs in the current circumstances. Mainly because of the reason that data has to be collected at certain point of time by the financial institutions.

\* \* \*

Within the last several years Juicy Score has been successfully used a solution based on alternative data. This solution allows avoiding personal and sensitive data collection. We process the alternative non-personal data that is many cases sufficient to make a decision on applications for a loan or insurance. So there is no need to start talk to a customer via a questionnaire.

This is only one of many steps. In any new world a code of laws or a code of conduct (even Jack Sparrow … captain Jack Sparrow had a code) is sooner or later arisen. Therefore, we suggest looking at our points, which, in our opinion, will help to establish a mutual understanding between the applicant and the lender. This should help to reduce useless accumulation of personal and sensitive data, which in many cases can be considered excessive in the 21st century.

### 1. Personal Data

- To reduce the use of personal data to the necessary minimum and to enhance alternative solutions which may simplify the decision-making system and risk policies;
- To stop collecting, storing and transmitting personal data without meaningful need and without user notification, freeing up resources and unloading business processes;
- To develop, maintain and switch to using alternative data that is not valuable for fraudsters and less risky for users.

Sensitive information (for example, disposable income, religion or other socio-demographic data — data that does not carry a material threat to a person, but can cause social damage)

- To form a commonly accepted definition and make a list of such data, recognizing its status;
- To regulate the use (at least at the level of a gentlemen’s agreement or code of ethical online-business conduct) of the data in order to exclude uncomfortable situations for customers and reduce risk.

### 3. Security & Privacy Balance

- To manage finding a balance between the necessary security of businesses and comfortable end-user privacy;
- To maintain and regulate balance in a dynamically changing online world, responding to changes in the activities of both groups;
- To reach a new level of friendly service, where security does not contradict to respect for privacy.

Having made these small but very important steps, it is possible to transform the businesses and to create a new approach to security: the personal data will no longer be the weak link. The main thing is to start and to remember that this is a goal, that can be achieved only by joint efforts.

We are ready to assist and to support on the way to this transformation, and we invite you to try our service, and will be happy to disclose our capabilities and advantages.

_How long have you not been online today?_

_Since 2007, humanity stood on the path of Internet addiction.The famous “We have reinvented the phone” sounded 12 years ago on the presentation of the first iPhone. That phrase marked the beginning of a new era. The attitude towards information as well as the entire service industry have been changed and reinvented alongside with the possibility of unlimited and permanent Internet connection. Everything happens online today as every even a little self-respecting company should have a solid and functional website, complemented by a mobile application. Wi-Fi-networks development, the connection quality, and speed improvement provide the continual contact with customers._

js_safe-2-1.jpg

Security VS Privacy: in search of a fair balance

_At the end of October, we had the honor to speak at the Finovate Asia 2018 conference! Like a few months earlier, we continue to demonstrate the efficiency of non-personal data, telling about the successful resolving of client authentication and risk reduction tasks through device and user network behaviour analysis._

_However, this time we tried to prepare a special presentation and for the first time demonstrated a prototype of our new product that implements the principles of dynamic authentication and increases the end-user security!_

[Earlier we shared ](https://juicyscore.ai/blog/en/loan-for-mona-lisa-en/)our thoughts regarding the difficulties of applying classical biometrics and the importance of dynamic authentication. Today we are happy to go from the idea to the action and to present you a service that allows you improving your credit rating via selfie!

We invite you to the first JuicyBio demonstration to share this important moment with us!

Finovate Asia 2018\. Presentation of the prototype of our dynamic authentication solution.The product we presented allows enrichment the data provided by the classic JuicyScore, increasing authentication accuracy, reducing the second type error. Moreover, it significantly improves end-user security and prevents from answering unnecessary questions and sharing a lot of sensitive information.

Meanwhile, the picture does not leave the applicant’s device and not transferred and not stored anywhere outside as only a few descriptive attributes are collected. This means that very soon the use of biometrics will be safe and simple as taking a picture as a keepsake!

![](/uploads/30_1_7f44d640a7.jpg)

Recognize us by orange ties and do not hesitate to skip personal data!

At the end of October, we had the honor to speak at the Finovate Asia 2018 conference! Like a few months earlier, we continue to demonstrate the efficiency of non-personal data, telling about the successful resolving of client authentication and risk reduction tasks through device and user network behaviour analysis.

30-JS11-1.png

Finovate Asia 2018. Dynamic authentication solution prototype demonstration

_In February, we published an article about how the psychotype affects credit service users financial behavior. In addition to that, in June, we launched a laboratory for the evaluation of non-personal data — [psychotype.online ](https://psychotype.online/). Today we turn back again to the topic of the psychotype influence on the borrower’s credit behavior and income assessment. Referring to our research, we will talk about various data types synergy to gather more information about the client._

There is no secret that “standard” documents and data (for example income tax certificate, also known as 2-NDFL), are not always efficient factors for a client’s rating, since in many cases it is difficult to confirm or deny the indicated information.

To make a decision on loan, it is extremely important to know the disposable income of the client. However, any information about the finances and income of a particular person is extremely sensitive. People do not like to discuss it. Moreover, It will also be uncomfortable to answer, in case a prospect client is asked online about his or her income rate. In addition to that, such questions may discourage the client in service use. According to our estimates: about 50% of clients do not answer question about their income and at least 15% of those who answered — manipulate with numbers.

Although the level of income is not personal data, it is extremely sensitive information for the client. However, within the existing credit decision approach, credit organizations still either use a little or even ignore alternative data. That is why focused to develop a model that does not use consumer personal or other sensitive information.

This model can be built on device quality data and Internet-connection characteristics. In addition, if we also add psychological characteristics of a person and use the results of our research on the correlation between disposable income and temper, you can get a very powerful tool with high precision.

User behavior markers may rate the device characteristics, loan applications frequency, and connection quality. Temper completes the picture, which results in an impersonal user portrait with all necessary data in it. The device quality rate (minimum value is zero, maximum is 5), together with other predictors, which have already demonstrated a high predictive power.

![](/uploads/29_1_0422cc3c44.png)

  
_Graph 1\. Device quality index and disposable income (in thousand rubles)_

As can be seen from the graph, most of the borrowers use inexpensive devices. In addition, the population total disposable income may be determined for the country or region where this data has been collected. The owners of higher-quality gadgets were distributed in the fourth and fifth groups, where the income level is pretty much big. But what if the relationship between the device quality and an average income is insufficiently indicative?

Let us look at the variable that characterizes the quality of the Internet infrastructure. From Graph 2 it follows that the infrastructure quality index can also say a lot about the client’s disposable income.

![](/uploads/29_2_1_a175fd57f2.png)

  
_Graph 2\. Categories of infrastructure quality index and disposable income (in thousand rubles)_

The higher the value of the infrastructure quality index (distributed from G1 to G6), the higher the disposable income of the applicant.

Now let us see how a person’s income depends on his psychological type. If you look at Graph 3, it immediately is clearly identified that Extroverts (the first letter E designates the type) according to statistics have much higher income level than Introverts (the first letter I designates the type).

![](/uploads/29_3_c10b24c3ba.png)

  
_Graph 3\. The distribution of disposable income (in thousand rubles) by psychological types_

This can be explained by the fact that people with a bold, strong-willed temper often become leaders and occupy executive roles, respectively, have a higher salary level. Another fundamental property of the individual is how a person prefers to perceive information — in general or detail-focused. A similar dependence is seen in connection between temper and device quality index.

![](/uploads/29_4_ba4fdf0f27.png)

  
_Graph 4\. Distribution of disposable income (in thousand rubles) among extroverts and introverts according to the device quality index_

Even if an introvert and an extrovert use a device with the same quality rate, they may have different average income.

Therefore, it can be concluded that the model, which accounts the client’s psychological type, has greater predictive power and accuracy.

Research shows that financial activity, credit services usage frequency, amount and characteristics of a client’s cash flow directly depend on temper. We have collected data available in the current version of the API, and enriched them with information on psychotypes — this is how a simple model for disposable-income determination appeared. It includes 5–10 variables, and the received R2 level is significantly higher than 0.7, which indicates its high level of stability.

The model demonstrates the effective use of alternative data to determine disposable income — the most important factor in credit decision-making. Moreover, it operates without personal data and requests from the client sensitive information.

In February, we published an article about how the psychotype affects credit service users financial behavior. In addition to that, in June, we launched a laboratory for the evaluation of non-personal data — psychotype.online . Today we turn back again to the topic of the psychotype influence on the borrower’s credit behavior and income assessment. Referring to our research, we will talk about various data types synergy to gather more information about the client.

29-JS10-fin-1.png

Temper is meaningful…and what is the value?

_[A little over a year ago, we were raising the topic of ](https://juicyscore.com/dinamicheskaya-identifikaciya/)online client identification. Today we are going to further discuss problems and present you our new technological solution, developed over the past time._

Within the last five years, the identification of client’s personality has become not only a mandatory regulatory requirement for financial organizations but also a sales driver and risk reduction tool. Compiling a real portrait and getting the right definition of a client (along with his or her needs, character and behavioural characteristics) is a golden ticket that allows offering the client what he or she needs, and receiving loyalty, maximising profits and minimising risks in return.

Is not it a dream for any bank, microlender or insurance company? But the reality turns a relatively simple and unambiguous tool into a real challenge for both online and offline channels.

Many financial operations, such as loan issuing via both offline and online, require client identification. Client ID (and a number of other documents) is usually enough for identification. Bank or microlender office employees and consultants pick up an ID, check it, compare it with an opposite sitting customer (as a reference) and complete the first stage of identification, together with a whole set of client data collected and recorded. This how it works for the offline channel.

Online is also complicated by the fact that there is no “reference” sitting opposite (as a matter of fact there is no idea of who “is opposite” to you behind the screen) and there is a digital copy instead of real ID. Then for a better “acquaintance” organizations start to collect countless personal and behavioral data, alongside with biometrics (face image and voice recording).

And that’s where the problems begin.

Picture 1: Black Sails, s1e2, 2014.- source: Youtube, channel theISDD  
Admittedly any additional action or question to a “good” and trustworthy client may lead to client’s refusal from the service. A good borrower is not at all inclined to leave too much data about himself or herself, especially when it comes to the image (especially in the online).

Then it is necessary to consider that such data bear a significant risk for the credit organization. After all, without certainty in the ID based identification (and even worse – in fraud cases), further data collection can “paint” a portrait of a completely different client. Such seemingly unique and inalienable data can still be stolen or obtained by fraudsters. As a result both the organization and the clients (especially the real owner of the stolen portrait) will suffer.  
Finally, even if we exclude the negative side, many of these data are not and may not be referent and stable in time.

On the other hand, there are developers of biometric solutions competing in comparing ID and real “portraits”. But the appearance of a person can change gradually (aging) or quite dramatically (style changes). A photo in ID changes with the document only twice (according to the law) and is not able to keep up with appearance changes. As a result, both appearance and ID can not be a stable in time reference. Slight difference between them leads to errors of the second type (incorrect or imprecise definition of the client, for example, below 95% caused by a new wrinkle or scar on the face), and a “good” client is declined for a new loan or insurance.

To change the situation additional information may be collected, but then again the dislike for additional questioning from the “good” borrowers and, as a consequence, the lost deal again.

Moreover, the fact of data storage by such suppliers bares a huge risk for the clients whose data may be compromised as well as requires precise meet of legislation requirements.

To avoid such mistakes, as well as to call additional checks only when necessary, together with partners we tried to develop our way of customer identification. The main values and principles of our approach are:

- We use stable in time identifiers;
- We do not work with the image of the client, but only collect certain and sufficient factors that describe the properties of this image;
- Information gathering takes tenths of a second, without giving sensible inconveniences to the client in the process of application fulfilment;
- We do not collect and especially do not store personal and other data that may bring a risk to the customer — data we collect can not be used for taking a loan!

![](/uploads/2_3_1_bdcb889818.jpg)

_Picture 2: the prototype of our application for dynamic identification_  
The combination of all these factors makes it possible not only to obtain high accuracy identification but also to introduce the dynamic identification when only the necessary information is requested, which helps to minimize the shades of gray in the client stream.

A little over a year ago, we were raising the topic of online client identification. Today we are going to further discuss problems and present you our new technological solution, developed over the past time.

js_mona_lisa_1-min-1.png

Loan for Mona Lisa

_[In May 18 we presented a new version of API v10, ](https://juicyscore.com/new-api-10-0/)where more than 200 predictors and improved score model are now available to our clients. Com-paring to the previous v9 release more than 30 completely new predictors were added. One of the most important add-ons was the launch of new index variables, specially developed by machine learning methods. Empowered by significant prediction ability, they show certain charac-teristics of a device, Internet connection, as well as a number of other metrics and are ready for use in the client’s credit conveyor._

_Risk management role is to collect and deeply analyze all available data about the borrower to achieve the maximum decrease in write-offs on the application inflow and in the client portfolio. If the grouping of data is right, it will allow to identify and strengthen trends and increase the predictive power of a line as a result._

_Today we are going to introduce first test results of five new API v10 predictors._

StopMarkers. IDX1

The variable that aggregates more than 20 different stop-markers (TOR, cookies copying, User Agent issues, etc.), which are usually rare (less than 1%) events, but with a high probability of write-offs.

On the charts below the histograms show the distribution of the population between IDX1 values (orange and purple for POS bank and microlenders respectively), and lines for the relative bad rate to the average.

![pastedGraphic.png](blob:https://juicyscore.com/blog/45bf4bd9-3c7c-4f9e-9c49-050fe13c18b3)

Chart 1: IDX1 distribution by population buckets and the relative bad rate indicatorIDX1 = 0 means “no stop markers” found; if IDX1 = 1, we recommend using enhanced verification for such applications. When IDX1 ≥ 2 concentration of fraud-risky applicants is high and application reject without additional checks is most appropriate decision.

User Behaviour Markers. IDX2

This variable collects behavior markers 

of the device user. The risk level depends on anomalies re-lated to such factors as the application fulfilment speed, the number of corrections, etc.

![](/uploads/1_pasted_Graphic_1_97927c3951.png)

Chart 2: IDX2 distribution by population buckets and the relative bad rate indicatorIDX2 values range from 0 to 3 can be used as an additional parameter in the lender’s integral score model. Applicants with a high risk of fraud are marked in IDX2 ≥ 4, where application reject is most appropriate decision.

Device Markers. IDX3

The variable is built on secondary risk factors associated with the device from which the application is fulfilled. Suspicious user behavior is associated with anomalies on the device hardware, as well as specific settings of its software.

![](/uploads/2_pasted_Graphic_b007359a83.png)

Chart 3: IDX3 distribution by population buckets and the relative bad rate indicatorIDX3 values equal to 0 and 1 are recommended for including into the lender’s integrated model, 2 for additional manual checks, and 3 or greater for fraud prevention rules.

Connection Markers. IDX4

The variable combines network connection of the device markers: IP address anomalies, the time zone mismatch, proxying, etc. The risk of write-off increases in proportion to the quantity and quality of revealed anomalies.

![](/uploads/3_pasted_Graphic_868a7d5979.png)

Chart 4: IDX4 distribution by population buckets and the relative bad rate indicatorThe variable shows a good monotonicity of the bad rate and may be applied to the score model for credit risk rating well. IDX4 values ≥ 4 can be used for fraud prevention rules.

Device Quality. IDX5

The variable aggregates device characteristics; it is a function of device cost and quality which cor-relate with applicant’s income.

![](/uploads/4_pasted_Graphic_f9468e38a8.png)

Chart 5: IDX5 distribution by population buckets and the relative bad rate indicatorWith the increase of the device quality index value, a significant decrease in the bad rate is ob-served. We advise using this variable for the lender’s integral score model of credit risk rate, as well as for setting a credit limit.

_As can be seen from the given examples, the index variables have a high predictive power even when used individually. So IDX1 and IDX3 are applicable to fraud prevention, and IDX2, IDX4, and IDX5 — for credit risk rate integrated scoring model. When grouped right, their combination with frequency characteristics can further increase risk manage-ment efficiency and contribute to more accurate decision-making._

_Our experience shows that only deep understanding of the physical meaning of each attribute, reasons of its value appearances, as well as the affecting factors and their correlation, can achieve a high statistical significance of the predictor. As a result, it can increase the attributes contribution to the resolving power of the conveyor and influence on business indicators of the organization._

In May 18 we presented a new version of API v10, where more than 200 predictors and improved score model are now available to our clients. Com-paring to the previous v9 release more than 30 completely new predictors were added. One of the most important add-ons was the launch of new index variables, specially developed by machine learning methods. Empowered by significant prediction ability, they show certain charac-teristics of a device, Internet connection, as well as a number of other metrics and are ready for use in the client’s credit conveyor.

JS08-fin-1.png

API 10: Magnificent Five

_[In our article](https://juicyscore.ai/blog/en/banks-fintech-companies-friends-forever-are-they/) about banks and financial companies development strategy we touched the subject of business growth speed and innovation adoption speed._

_Today we will continue this conversation and consider from a different angle: we will discuss the differences between banks and fintech companies in terms of operational management, and also how this affects product development process and effect for the customer._

To begin with, we would like to briefly outline value creation methodologies that exist on the market.

Appeared in industrial 1940s and 1950s Lean Manufacturing (or Lean) methodology stated value creation principles which are based on **cost minimization and losses elimination**. These principles are achieved through a thoroughly designed and explored process of value production.

In addition to Lean new methodology called Agile (Flexible Development Methodology) origins from 1980s. The methodology has focused on a value that should adopt to the rapidly changing **client** [needs](http://ecgroup-intl.com/agile-and-scrum-lean-and-kanban-a-brief-history/). This methodology predominantly speaks about changes in thinking and practice.

Appeared in 2000s DevOps (Development & Operations) methodology is about increasing value creation speed through implementation of **organizational changes** and building close relationship between development and operations functions.

![](/uploads/67147776_489608268469589_1827805293673709568_o_e773b558d8.jpg)

The advantages and disadvantages of these methodologies have been well studied (at least in theory) over the years and in this article we would like to compare them in terms of two critical aspects – **product value and the speed of value creation**.

The product value

Lean is about value creation once and for the long period of time and in the most cost-efficient way. Meanwhile Agile is about product value (in general or its attributes) which should change and keep in line with rapidly changing needs of the target audience. Agile together with DevOps operate with the concept of MVP **(Minimum Viable Product)**, which means the core of value, constantly evolving by adding new features and attributes.

Speed of value creation

In Lean speed is achieved by reducing redundant activities and time waste between various stages of value chain. In Agile speed increase is achieved through the distribution of competencies and skills within the matrix team. DevOps increases the speed by creating a team of diverse-skilled experts and almost complete autonomy for decision making.

It is interesting to observe how the practical implementation and use of these methodologies influenced the work of banks and fintech companies.

Bank say

- Banks began using Agile which let them following technological giants and fintech companies. As a result this allowed reducing product development and launch time up **to several weeks**  
instead of [current several months](http://www.tadviser.ru/index.php/%D0%9F%D1%80%D0%BE%D0%B5%D0%BA%D1%82:%D0%A1%D0%B1%D0%B5%D1%80%D0%B1%D0%B0%D0%BD%D0%BA%5F%D0%A0%D0%A4%5F%28Agile%5F%D1%82%D1%80%D0%B0%D0%BD%D1%81%D1%84%D0%BE%D1%80%D0%BC%D0%B0%D1%86%D0%B8%D1%8F%29).
- Banks admit that Agile teams reduce the product creation time up to 30 days, and the DevOps approach allows completing the development cycle in 5 days.>.

Fintech companies say

- With the Agile approach product changes implementation rarely exceeds 20 days (the minimum time can be only a few hours) and system changes are usually made within **[a few minutes ](http://futurebanking.ru/post/3522)**.

As we can see from the answers, leading Russian financial institutions are interested in changing approaches of product creation and change to achieve better results. Nevertheless, despite the enormous changes, the speed of development and implementation in banks is still far behind the fintech sector which includes microlenders, various IT companies, service providers, and so on.

What is the main reason?

Agile and DevOps really reduce the product development time and in addition to that, the fintech companies have also succeeded in some other aspects influencing the time of product launch.

- Time for development team coordination in fintech companies rarely exceeds 10% of the total project time which is 3-5 times less than time on the same projects in banks.
- Decisions are made by the team within constant time costs for coordination and management, rather than a set of numerous steering committees and coordination at different levels of management.
- Communication between various functional experts within the product team goes directly and not by escalating an issue to the management of one function, with subsequent redirection of this question to the management of another function.
- Specialization for efficient problem solving, significant time reduction or complete elimination of non-typical functions.

In conclusion, let’s discuss another important question about the forms of operational management that will appear in the future. What will replace DevOps? What will increase product competitiveness and quality?

Basing on the work approaches of particular fintech companies we can already identify several key trends. Occasionally ValOps (an acronym for Value & Operations) term is mentioned on various discussions and presentations and we would like to highlight some aspects that fall under this notion

- Open API – the API format which aims to increase transparency and reduce the risk to the client, increases the value of data, the product development and modification speed and should become a fundamental approach in product creation and promotion.
- The “flat” organisational structure – actual number of management levels should ensure high speed of decision-making speed, transparent communication between functional teams and keep focus on the most important items.
- Specialization and high level of skills – each team member must have an expertise in one area and general understanding in several related areas
- Further evolving of MVP idea – The”Strategic MVP” which aims to increase product competitiveness on the medium and long term perspective via changing the product core and not just the attributes.

These and many other principles are already being used in practice. The value created for the end customer should be the only measure of efficiency for both operational management and company IT systems use.

In our article about banks and financial companies development strategy we touched the subject of business growth speed and innovation adoption speed.

Today we will continue this conversation and consider from a different angle: we will discuss the differences between banks and fintech companies in terms of operational management, and also how this affects product development process and effect for the customer.

1-1.jpg

Banks & FinTech Companies, Enemies Forever – or Are They? Part 2

We are happy to introduce our anniversary release of JuicyScore – v.10.

We have significantly expanded functionality and set of predictors.

  
**15+ new significant predictors, such as:**

* Device quality and cost evaluation;
* Device active usage time performed by the current user;
* Internet connection quality index;
* Device structure anomaly index;
* User online behaviour and Internet resources active usage aggregated indexes;

  
**Virtual users seamless identification technology upgrade:**  

* Personal data and direct identifiers are not used
* Russian and EU (GDPR) laws and some other normative acts compliance

**Expanded geography**

Our experts built infrastructure and client support system for South-East Asia and Western Europe

We’ll be happy to provide you further [details ](mailto:info@info@juicyscore.com).

We are happy to introduce our anniversary release of JuicyScore – v.10.

We have significantly expanded functionality and set of predictors.

JuicyScore-Illustration-1-1.jpg

New API 10

Currently, an active reassessment is under way regarding the use of personal data in the number of regions worldwide. The new [GDPR](https://ru.wikipedia.org/wiki/%25D0%259E%25D0%25B1%25D1%2589%25D0%25B8%25D0%25B9%5F%25D1%2580%25D0%25B5%25D0%25B3%25D0%25BB%25D0%25B0%25D0%25BC%25D0%25B5%25D0%25BD%25D1%2582%5F%25D0%25BF%25D0%25BE%5F%25D0%25B7%25D0%25B0%25D1%2589%25D0%25B8%25D1%2582%25D0%25B5%5F%25D0%25B4%25D0%25B0%25D0%25BD%25D0%25BD%25D1%258B%25D1%2585) will enter into force in the EU, and this is likely to be only the beginning. We may expect new regulations on personal data – PSD3 and GDPR2 data in the next 3–5 years. Those directives are not specific to the EU. It is highly likely that we will see similar initiatives in other markets in the nearest future.

The primary reason behind those changes in the regulatory framework is regulators’ demand for enhanced security of personal data and end users.  
Today, on the one hand, personal data is the key to a wide range of online services, which is obviously a positive aspect. On the other hand, insufficient regulation leads to personal data leakages, which is the opposite side of the wide of digital services.

The existing difficulties connected with personal data are not limited to security. Another problem, which many companies are now trying to resolve, is the relatively low information value of personal data in the online area comparing with the offline world. The solution lies in finding additional information sources / data systems that are not connected to personal data or an individuals.

We are at JuicyScore – have substantially challenged the value of personal data. And we have created a full-fledged platform for risk management for financial online organisations based on non-personal data that makes it possible to significantly enrich and increase the efficiency of existing risk management solutions. This platform is grounded on the following principles:  
— no personal data is involved;  
— a wide range of online predictors is available that may be easily integrated into the risk assessment process;  
— a set of online tools is available for online monitoring and risk strategies adjustments.

Our first public presentation was held at [Finovate](http://finovate.com/), one of the most prestigious international venues for financial innovations:

![](/uploads/26037622197_99d5d22393_o_min_2_befd58794f.jpg)

[Finovate](https://finance.knect365.com/finovateeurope/) was organised at very high level. We hope that, similar Fintech international conferences will be held in Moscow in the nearest future.

Currently, an active reassessment is under way regarding the use of personal data in the number of regions worldwide. The new GDPR will enter into force in the EU, and this is likely to be only the beginning. We may expect new regulations on personal data – PSD3 and GDPR2 data in the next 3–5 years. Those directives are not specific to the EU. It is highly likely that we will see similar initiatives in other markets in the nearest future.

JS Platform: 5 in 1. Risk management automation

Risk management automatic reporting: new product announcement

Fraud traffic analysis in different regions: short study

The future of Device ID in the digital world

MoneyMan MX leverages JuicyScore to increase approval rate x 1.5 times

The company

Products

Rules