BAKU, Azerbaijan, October 13. A new model designed to improve the assessment of bank customers’ creditworthiness and reduce the risk of late loan repayments has been developed in Azerbaijan in cooperation with Mastercard, Azericard Deputy CEO Zakir Ahmedov said on Tuesday at the Baku Fintech Sector Trade Mission Program, Trend’s correspondent reports from the event.
According to Ahmedov, around 70% of Azerbaijan’s population has a credit history, with the relevant information collected by the Azerbaijan Credit Bureau (ACB). However, people without a credit history also use bank cards, and their financial transactions can provide insights into their income, spending and consumer habits.
“About 70% of people in Azerbaijan have a credit history. The others do not, but they also have bank cards and conduct transactions through them. These transactions can provide information about customers’ income, expenses and the areas on which they primarily spend their money,” Ahmedov said.
He noted that these data create new opportunities to assess the creditworthiness of people with no credit history or a limited borrowing record.
According to the company representative, the model developed jointly with Mastercard analyzes various factors to estimate the likelihood that customers will make their loan payments on time. It uses the SHAP (SHapley Additive exPlanations) approach to explain how individual factors contribute to a credit-risk assessment.
“Using this model, we can identify the factors that influence whether a customer is a reliable or risky borrower. These include previous loan delinquencies, the number of loans a customer has, income, age, gender and other indicators,” he said.
Ahmedov noted that approximately 1,250 indicators were initially considered. Following the analysis, 56 key indicators were selected for the model to assess each customer’s risk of late repayment.
He added that delays on previous loans negatively affect a customer’s risk profile. By analyzing these and other factors together, the model provides banks with additional information to support lending decisions.
According to Ahmedov, the model was trained on data from the first three quarters of 2025 and subsequently tested using data from the following quarter. Its performance was compared with that of credit-scoring systems currently used by banks.
“Based on the test results, our model can perform 3%–7% better than the existing scoring models used by banks,” he said.
He added that the tests also assessed the model’s ability to identify late repayments among customers considered eligible for loans. According to the figures he presented, 0.3% of customers recommended for lending by the new model subsequently fell into arrears, compared with 16% of customers who received loans under existing bank assessments.
Ahmedov said the model is expected to be presented to banks in the near future for use in their credit-risk assessment processes.
Its adoption could help banks assess customers with limited credit histories more comprehensively, improve lending decisions and identify potential repayment risks in advance.