AntiBank ecosystem, artificial intelligence (AI), replacing people in the performance of complex tasks, is used for scoring and in case of non-repayment of loans.
The concept of artificial intelligence (AI) appeared more than 60 years ago. However, this technology did not appear in the world of financial services until the early 1980s. After a short period of “great hopes” it was at first forgotten, but now, as it seems, it is gaining popularity again.
According to CNBC, about $700 million was invested in AI only during 2014-2015.
The two most important reasons for the revival of popularity of AI are availability of powerful computer resources and availability of a huge amount of computer data.
The achievements in the field of hardware and software were simply amazing in recent years, so now computers are able to provide computing power unthinkable in old days, and computer systems with sufficiently large capacity have become much cheaper. Today, even for a small start-up, such computer capacities and volumes of information are available, which 10 years ago were available only for large banks and institutional structures.
The widespread use of social networks, mobile smartphones, tablets and so-called “wearables” (devices attached to clothing), along with development of sensors and their installation in smart cities, the emergence of the already well-known Internet of things (IoT) etc., all this generates a giant amount of data or, in other words, ‘big data’, and all this makes it possible to apply the artificial intelligence.
AI is used to calculate a rating of a borrower or, in other words, for scoring.
● Scoring begins with the collection of information from different sources.
● The extracted information is processed (structured) by special algorithms for ‘big data’ processing.
● Structured information is sent to the Information Processing Center, where this information is analyzed using the machine-lining and then serves as the basis for calculating the credit rating.
The sources for collecting information for scoring are, first of all, data from public sources. For individuals, one can use also, for example, data from utilities suppliers, electricity providers, mobile operators, data from social networks, etc.
Data on individuals are personal data (passport data, marital status, etc.), financial data (amount of income, place of work, position), etc.
Data on legal entities are data on business reputation, market situation, etc., as well as economic data (indicators of financial stability, profitability, etc.).
The database of AB-ecosystems first receives unstructured information, not tied to individuals and organizations. For further processing of information, it needs to be streamlined and structured so that the AI algorithm can correctly analyze it and draw conclusions. The algorithm of data structuring is based on the comparison of the main indicators in the databases, and on the basis of coincidences it creates a ‘AntiBank ID-passport’.
To calculate the rating of a borrower, the usual scoring card is first developed based on the usual procedures and usual data that characterize the level of the borrower’s income: education, work, neighborhood, etc. Based on this data, an initial portrait of a ‘good’ borrower is formed.
Then our system adds new parameters, which can be rarely used in practice in other organizations.
For example, we can analyze how a person behaves when filling out our questionnaire.
If he or she uses a copy-paste to insert a name and address, then one can be suspected of being dishonest. We can also, for example, determine a brand of a smartphone from which the application is filled out. It’s one thing if this is the iPhone of the latest model, another thing, if it is the cheapest and oldest smartphone, while the applicant writes that he has a high salary and position.
Over time experience is accumulated, for example: with such and such parameters, people return a loan with probability of 90% (9 out of 10 people), and with others - 10% (1 person out of 10). Of course, it is better to give out loans to people with the first set of parameters.
Which borrower parameters should be taken into account, as well as which of the parameters should be considered as very important, and which ones are not very important, it is a complex task that does not have an unambiguous theoretical solution. Sets of parameters and their breakdown into important and unimportant are selected experimentally and are constantly tested in practice.
We have several sets of parameters, and they compete with each other in the sense that the set that best predicts the results of lending is used for recommendations until another set appears that works even better.
In this way, “machine learning” occurs, i.e., improving of an algorithm on the basis of experience accumulated in the course of work of the algorithm.
Progress in the field of AI has reached such a stage now that AI is able to help not only make decisions about whether to issue a loan or not to issue, but also how to return overdue loans.
Our system analyzes data on borrowers and their friends available on the Internet, and then communicates with the borrower by phone with the help of a talking robot. Conversations are recorded and analyzed using an algorithm that then determines the wording that is most likely to have an effect on the borrower and will cause the debt to be repaid. The system also communicates with his friends and with their help asks the borrower to return the money. In practice, the share of loans returned using AI is doubled compared to conventional methods.
Thus, we will help our Investors not only correctly and impartially assess the risk when investing, but we will also help them in case of non-repayment of the loan.
In general, the use of AI allows solving the tasks of calculating a rating of a borrower (scoring) and repaying overdue loans much more efficiently than without using AI.