dr Piotr Wojewnik

Individual credit decisions with AI

BIK has been managing a unique information base on a national scale and gathering experience of experts in developing technological solutions for the financial sector for more than 20 years. BIK system solutions provide measurable business benefits – they support risk mitigation processes in financial institutions as well as ensure access and operations in real time thanks to the use of innovative solutions, e.g. scoring analyses using artificial intelligence (IA) or machine learning (ML).

BIK cooperates closely with banks and other financial sector institutions (credit and saving unions, loan companies, bank leasing and factoring companies). As a result of this cooperation, the BIK database is supplied with customer data by all banks in Poland and the vast majority of loan institutions. Banks and loan institutions provide BIK with so-called hard data, that is information on credits and loans granted, including figures, values and information on the quality of repayments made. Currently, the BIK database contains over 159 million account history entries for 24 million individual customers and 1.4 million small and medium-d enterprises.

It is difficult to disagree that this is the only set of structured data in the country on the financial situation of a large part of the population. The data held by BIK make it possible to conduct detailed analyses of the creditworthiness of financial sector customers. Given that the quality of the statistical models being built depends on the quality of the data held, thanks to the data available in the BIK database it is possible to determine credit risk precisely. 

BIK actively participates in creating new products and services, including testing new solutions on anonymised data, which result in, for instance, connecting with external sources of information (e.g. social networks). 

The use of external databases increases the degree of matching the analysis with a specific loan or credit customer. Machine learning (ML) is useful in searching for such data. This solution makes it possible to clean the available data by removing insignificant information which is not relevant for the credit scoring by finding links between individual factors of the customer’s creditworthiness assessment. Thanks to the application of data management technologies, BIK may analyse a wide database, which does not pose a challenge for ML.

ML-based credit risk model - BIK research

Due to current sector-specific regulations, provisions on personal data protection and the need to ensure transparency of the credit process, banks are obliged to document the model used and to enable tracking of the decision-making process (justification of the decision to refuse to grant a loan), which at that moment is considered to be hindered by machine learning. However, according to the results of the research conducted by BIK, it is possible to introduce changes also in banks thanks to the use of the eXplainable Artificial Intelligence (XAI) method, which makes it possible to diagnose the ML model. 

The research was conducted with the participation of experts from the University of Warsaw and Data Juice Lab Sp. z o.o., we encourage you to read the results of the research.

Even without implementing XAI methods, ML models remain functional and can be used to improve traditional models. Conclusions drawn by ML may constitute the basis for making changes in traditional models in the form of characteristics that have not been considered before. Thanks to this, it is possible to check whether there is a chance of creating a better algorithm for a traditional model.

Taking into account the potential arising from external sources of information while simultaneously using ML makes it possible to adjust the scope of the information processed to the needs of BIK and BIK partners.