India's four credit bureaus — CIBIL, Equifax, Experian, and CRIF Highmark — together maintain files on approximately 550 million individuals. Of these, roughly 150 million have sufficient credit history for a bureau-based score to be meaningful. The remaining 400 million new-to-credit or thin-file individuals represent the largest formal credit exclusion in the world — people with the capacity to repay who cannot access formal credit because they have no credit history.
Alternative credit scoring uses data sources beyond credit bureau records to assess creditworthiness. The Account Aggregator framework is the most important enabler: it allows applicants to consent to sharing their bank transaction history — salary credits, regular bill payments, spending patterns — directly with a lender in standardised form. A borrower who has maintained consistent salary credits and avoided overdrafts for three years is demonstrably creditworthy even without a CIBIL score.
GSTN data is the alternative data gold standard for MSME lending. A small business's GST filing history — filing regularity, turnover trajectory, input tax credit patterns — provides accurate revenue and business activity data that traditional balance sheet analysis cannot capture for unaudited businesses. Lenders with access to GSTN data through the GST System's analytics portal can make loan decisions for MSMEs that banks have historically served poorly.
Telecom data — mobile recharge patterns, roaming usage, SIM tenure — has been used by some fintech lenders as a proxy for financial stability and location consistency. While predictive at population level, the regulatory permissibility of using telecom data for credit decisions without explicit consent is evolving.
The ML models that combine these alternative data sources — gradient boosting ensembles with careful feature engineering, neural networks for high-dimensional transaction data — must be built with rigorous fairness testing. Alternative data sources can encode geographic, demographic, or occupational biases that disproportionately harm already-marginalised groups. Responsible lending requires active bias monitoring and fairness constraints as part of model governance.