Credit Risk Assessment in Digital Lending: Evaluating Alternative Data vs. Traditional Credit Scoring

Abstract:

Abstract

The rapid growth of digital lending platforms and financial technology firms has transformed retail credit delivery across India. Traditional commercial banks and non banking financial companies (NBFCs) have historically depended on centralized bureau scores, such as TransUnion CIBIL, along with formal salary slips and collateral security to evaluate borrower credit risk. While this conventional framework works well for salaried professionals in formal sectors, it leaves out millions of self-employed individuals, gig workers, small merchants, and first-time borrowers who lack formal credit records. In response, digital lending entities now evaluate alternative data sources, including Unified Payments Interface (UPI) transaction flows, utility bill settlements, goods and services tax (GST) filings, and Account Aggregator statements, using machine learning models to assess repayment capacity. This study compares traditional credit scoring against alternative data driven scoring models within the Indian banking ecosystem. Analyzing loan application and repayment data from 12,500 retail borrowers, the study evaluates predictive performance, credit approval rates for thin-file applicants, and twelve-month default outcomes across traditional, alternative, and hybrid underwriting approaches. The results show that relying only on bureau scores yields an AUC-ROC of 0.734 with a thin-file approval rate of just 19.2%. When consent-based alternative cash flow data is combined with bureau records in a hybrid model, the AUC-ROC improves to 0.882, thin-file approvals reach 56.4%, and portfolio defaults drop to 3.2%. The paper concludes with recommendations for Indian lending institutions to adopt compliant hybrid underwriting aligned with Reserve Bank of India (RBI) guidelines.