Case Study

Long time of waiting for a credit decision is a pain that affects a large part of bank customers. Self-learning of machines could make this process shorter basing on modules that support the generation of individual credit offers using Machine Learning.

Machine Learning in banking

Genesis

Machine Learning in banking

Filling in a bank application, credit rating, assessing collateral, risks, analysis and acceptance - these are the steps necessary to grant credit from the bank's perspective.

The decision-making process may look a bit different if the borrower is a client serviced by a financial institution for a long time. In such cases, the lender has a lot of knowledge about the history of cooperation - you can not only speed up the process of verification of the application, but even prepare a loan offer without request from the client.

Goal

Automatic analysis of creditworthiness and risk

The task that was given to Whiteaster was to prepare the automation of the client's creditworthiness assessment. This company's clients are salespeople who run theirs operations through one of the largest online sales platforms in the world.

The loan offer was to be based on an analysis of creditworthiness and risk. It was assumed that the loan should be repayable due to the sales results on the sales platform. These loans are a new product - the company did not have any historical data of granting and repayment of the loans.

Challenge

Designing seven Machine Learning models and a calculation module

In the first stage the offer is directed to current clients (using other financial products) that is white the company has a lot of data which was divided into two main groups:

  • historical data
  • current data
Historical data refers to sales, quality of customer service and statuses on the sales platform. Available information , apart from the fact that they differ substantially, cover different time of periods (from two-year to six-month periods).
The second set of information is current data that describes the activity of this platform sellers - current status information, sales information, products offered, billing, information about loans granted (currently being repaid), etc.

The analytical model was based on conclusions that were drew from data analysis and their impact on credit decisions. The final loan offer results from many factors, both historical and current. The observed increase of number of loans is a good proof that the decision system matures.  The architecture of the solution is also influenced by the method of obtaining data, the frequency of its updates and substantive meaning - the way of influencing the final opinion. Because the analytics results change - evolve during the project, the model changes as well following the data.  The architecture of the solution is also influenced by the method of obtaining data, the frequency of its updates and substantive meaning - the way of influencing the final recommendation. As a result of the project, a solution was created consisting of seven Machine Learning models and a calculation module based on designed algorithms.

Result

An application for credit decision recommendations based on machine learning

Learning of machine learning modules and improving algorithms is a time-consuming process. It requires many tests, changes, repetitions, analysis of results, diagnoses of specific cases, etc. Solution accepted for production use - basing credit decisions on its recommendations - was created in less than 1.5 months. Integration with the client's integrated system took another month.

Informujemy, iż w celu optymalizacji treści dostępnych w naszym serwisie, dostosowania ich do Państwa indywidualnych potrzeb korzystamy z informacji zapisanych za pomocą plików cookies na urządzeniach końcowych użytkowników. Pliki cookies użytkownik może kontrolować za pomocą ustawień swojej przeglądarki internetowej. Akceptuję