AI-Powered Credit Risk Assessment Using Alternative Data(Evidence from Bank Statement and GST Analytics in Indian Fintech Lending)

Traditional credit risk assessment in India relies heavily on credit bureau scores such as CIBIL and Income Tax Returns (ITR) (Reserve Bank of India, 2021). While effective for borrowers with established credit histories, these approaches are inadequate for evaluating thin-file or new-to-credit borrowers and Small and Medium Enterprises (SMEs)(International Finance Corporation, 2018). This limitation results in financial exclusion and inaccurate risk pricing for fintech lenders such as MM Fincorp. This study designs, develops, and evaluates an Artificial Intelligence (AI)-powered credit risk assessment framework using alternative data sources, specifically borrower bank statements and Goods and Services Tax (GST) filings. Historical loan data from MM Fincorp is used for predicting financial behaviour and business health. The predictive models used were Logistic Regression, Decision Tree, and Neural Network and evaluated accuracy, precision, recall, F1-score. ROC-AUC metrics were also used for model evaluation and k-fold cross-validation test was done. The study shows that alternative data can be a powerful way to predict loan defaults. The indicators such as cheques bounce, GST turnover trends, transaction stability, and filing consistency all turned out to be strongly linked with whether borrowers repay or default. Among the models tested, the Neural Network model stood out. Its ability to capture complex, non-linear patterns gave it the best predictive performance. This highlights how AI-driven underwriting can make lending smarter improving financial inclusion, boosting efficiency, and sharpening credit risk assessments.