EASE 9.0 · Socio-Economic Impact · Inclusive BankingInvisible to the Model, Not the Market
How AI agents extend credit to underbanked segments while managing risk profitably.
Financial inclusion at scale, with positive economics.
A gig-economy delivery worker earning a steady but irregular income has no salary slip, no formal employment letter, and often no bureau file at all. Traditional underwriting, built around salaried, formally employed applicants, treats an empty field as a reason to decline.
Why Manual Underwriting Cannot Serve This Segment Economically
A credit officer could, in principle, manually assess a thin-file customer's UPI transaction history and utility payment record. But at the volume a bank needs to make inclusive lending a real business line rather than a token CSR programme, manual assessment does not scale: the cost of a manual review often exceeds the margin on a small-ticket loan.
The Path to Progress
That customer is not necessarily a bad credit risk. They are simply invisible to a scoring model built for a different kind of income pattern.
The Business Case
Every published estimate of India's thin-file, creditworthy population is a model, not a census; useful for planning purposes, but not a figure this page will present as fact. What can be said with real numbers behind it: RBI's own Financial Inclusion Index, a composite measure of access, usage and quality of financial services, rose from 64.2 in March 2024 to 67 in March 2025, genuine progress on RBI's own measure that still leaves meaningful room to run. PMJDY (Jan Dhan) has opened 56.16 crore accounts as of August 2025, and roughly 23% of them sit inoperative. The gap between "banked" and "actively creditworthy" is not a marketing framing; it is a measured, persistent fact about India's own flagship inclusion scheme.
The gig-worker example this page opens with already has a real-world precedent. Swiggy's delivery-partner lending programme, run with NBFC partners including Indifi, InCred and IIFL, disbursed ₹102 crore to delivery partners over a twelve-month period, with more than 150,000 loan applications a month: a live, large-scale demonstration of exactly the alternative-signal underwriting (platform earnings and activity data standing in for a salary slip) this page argues for. Alternative-signal scoring turns a segment that was previously uneconomical to even evaluate, not merely unprofitable to lend to, into one a bank can serve at the same marginal cost as its existing book. That is a new revenue line, not a CSR cost centre.
Sources: RBI Financial Inclusion Index releases (2024, 2025); PIB, PMJDY account statistics (Aug 2025); Business Standard, coverage of Swiggy delivery-partner loan disbursals. Figures as publicly reported.
Where AI Agents Fit
Role: Enabler
This is the same Credit Decisioning Agent used in mainstream underwriting, pointed at a different signal set. Instead of bureau history and salary slips, it perceives UPI transaction velocity, utility-payment regularity, and GST filings for micro-merchants, scores repayment capacity against policy tuned for thin-file behaviour, and hands the case to a Reasoning Agent that produces the plain-language reason for the decision, approval or decline, so the customer understands why and not only the credit committee. The scoring itself happens entirely in the backend; what the customer experiences is simply an answer they could not previously get from a formal lender. The enabler work is invisible, but the outcome — a loan a previously invisible customer now qualifies for — is the most directly felt social and business impact of any category in the Socio-Economic pillar.
Managing Risk in a New Segment
Extending credit to thin-file customers is not the same as relaxing underwriting standards. The agent applies a different set of signals, but the same discipline: a policy engine that tracks default patterns against alternative-signal scores continuously, recalibrating thresholds as real repayment behaviour on the new segment accumulates.
The Skeptic's Answer
A portfolio manager's strongest counter-argument runs like this: a customer with no bureau file and no salary slip was excluded from mainstream underwriting for a reason, and relabeling that gap as an "inclusion opportunity" risks dressing up higher default risk as social good. Whether this segment is being lent to responsibly, or simply aggressively, is a fair question to press.
It comes down to sequencing, not sentiment. Shadow-scoring a thin-file cohort against how comparable bureau-scored customers actually repaid, before extending real credit, is what tells a bank whether UPI velocity and utility-payment regularity genuinely predict repayment for this segment, rather than assuming it and calling the assumption a CSR target. Swiggy's ₹102 crore in delivery-partner disbursals did not happen by lending on faith; it happened on the same alternative-signal discipline this page describes, at a scale that would have surfaced a bad model quickly. Widening the credit box on evidence, one occupation segment at a time, is the opposite of relaxing standards.
Treating Inclusion as an Underwriting Problem
Financial inclusion is frequently treated as a regulatory obligation to satisfy rather than a genuine growth opportunity. Appice's view is that the same infrastructure that makes a bank's mainstream credit decisioning real-time and auditable is the infrastructure that makes inclusive lending profitable at scale.