By Abhishek Gupta, Co-Founder, Dekrypt Labs
There is a quiet crisis playing out across India's lending ecosystem. NBFCs are disbursing loans faster than ever — digital onboarding, instant KYC, real-time transfers — but the intelligence underneath the credit decision has barely changed in a decade.
Bureau scores, income proofs, bank statements. The same three inputs that underwriters used in 2014.
Meanwhile, the borrower has changed completely.
A kirana owner in Coimbatore who runs ₹40 lakh in monthly GMV through a WhatsApp business account won't show up correctly in a CIBIL report. A first-generation entrepreneur in Jaipur who has built a profitable D2C brand but holds inventory on credit — her balance sheet looks terrible on paper and excellent in reality. A fleet owner in Nagpur with 14 trucks, all paid off, but no formal salary slip.
These borrowers are creditworthy. Traditional underwriting keeps calling them risky.
India's formal credit infrastructure covers roughly 220 million people. The country has over 900 million adults. That gap — roughly 700 million people with thin or no bureau files — is not a problem to solve later. It is the market.
NBFCs exist precisely to serve this segment. But most of them are underwriting it with tools built for a salaried, documented, metro-dwelling borrower.
The result? Two failure modes that destroy portfolio quality simultaneously:
False negatives — creditworthy borrowers rejected or under-served because their profile doesn't match the template. These borrowers often turn to moneylenders or informal credit at 36–48% per annum. The NBFC loses good business.
False positives — risky borrowers approved because they present well on conventional metrics. Salaried income, clean bureau, polished documentation — and then they default at month six.
Both failures are intelligence failures, not process failures.
The question is not whether to use alternative data. Every sophisticated lender already knows alternative data matters. The question is how to operationalise it without building a 40-person data science team.
Here's what AI-powered credit intelligence can surface that a traditional CAM (Credit Appraisal Memorandum) cannot:
Business network signals. Who does this borrower transact with? GST filings reveal supplier and buyer relationships. A small manufacturer whose largest customer is a Fortune 500 company has fundamentally different risk than one whose customer is an obscure shell entity. This is visible in data. Almost no NBFC is looking at it.
Regulatory footprint. MCA filings, ROC records, director histories, related-party structures. A director who has been associated with three struck-off companies in the last five years is a meaningful signal. Pulling this manually takes hours. Automated intelligence surfaces it in seconds.
Sector stress indicators. A borrower's risk is not just their own financial health — it is also the health of their sector. An MSME exporter in a category that has seen a 40% drop in US orders is categorically different from the same borrower six months ago, even if their own financials haven't caught up yet. Macro sector intelligence should be part of every credit decision. It almost never is.
Digital presence and consistency. For newer businesses, the consistency between what a borrower claims (turnover, employee count, business age) and what is publicly verifiable (LinkedIn, GST portal, trade directories, news mentions) is a surprisingly clean signal. Inconsistencies are not automatic red flags — but they warrant a question.
Even NBFCs that want richer intelligence face a practical constraint: the credit team can't spend four hours researching every application.
A branch-level credit officer handling 20–30 applications a week is not going to manually pull MCA records, cross-reference GST data, read sector reports, and check news mentions. They will look at the bureau score, verify the income documents, and move on.
This is not a failure of effort. It is a failure of tooling.
The entire value proposition of AI in credit underwriting is compression — taking the research that would take a senior analyst two hours and surfacing it in two minutes, at the point of decision, in a format the underwriter can actually use.
A well-structured credit intelligence system should hand the underwriter a CAM pre-populated with:
The underwriter's job then becomes judgment, not research. That's a better use of their time, and it produces better decisions.
India's NBFC sector reported gross NPA ratios of 6.4% as of March 2025 (RBI data). For smaller NBFCs in unsecured lending segments, the number is often higher — sometimes significantly.
A 1% improvement in NPA rates on a ₹1,000 crore book is ₹10 crore. That's not a rounding error. It is often the difference between a profitable quarter and a loss.
The arithmetic is straightforward. The implementation has historically been hard. AI is making it tractable.
Most NBFCs have invested heavily in the origination layer — apps, APIs, digital journeys. Very few have invested equivalently in the intelligence layer that sits underneath the credit decision.
BIOS is built to be that intelligence layer. It pulls structured intelligence from public regulatory sources, sector data, and business signals — and presents it in a format that credit teams can act on immediately, without hiring analysts or building internal data pipelines.
If your underwriting process still relies primarily on bureau scores and income documents, you are leaving signal on the table — and probably capital too.
See how BIOS works for NBFC credit intelligence → dekryptlabs.com