By Abhishek Gupta, Co-Founder, Dekrypt Labs
India's NBFC sector disbursed over ₹35 lakh crore in loans last year. A significant chunk of that — industry estimates put it between 4% and 7% — will turn into NPAs. That's not a rounding error. That's tens of thousands of crores written off because a credit officer looked at the wrong signals, or couldn't access the right ones in time.
The problem isn't bad intent. It's bad intelligence.
Every NBFC that lends to MSMEs follows roughly the same process: collect documents, run bureau checks, prepare a Credit Appraisal Memo, get approvals, disburse. A good CAM takes 3–7 days. A thorough one takes longer.
By the time that memo lands on the credit committee's desk, some of the information in it is already stale.
The borrower's GST filings reflect what happened two quarters ago. The bank statement analysis is a snapshot. The field investigation report captures one day's reality. Nobody looked at whether the borrower's key suppliers are under financial stress. Nobody checked if the industry segment is seeing rising defaults. Nobody mapped the promoter's other business interests against MCA filings.
This is not a failure of effort. Credit teams work hard. It's a failure of bandwidth — there's simply too much signal to process manually, and too little time.
A common defence for high NPAs is "thin file borrowers" — MSMEs with limited formal credit history. The logic is that without CIBIL history, you can't underwrite reliably.
This is increasingly wrong.
A business that's been operating for three years leaves a substantial digital footprint: GST return patterns, Udyam registration details, MCA filings if it's a private limited, director history, litigation records, trade references, and dozens of other data points that are either publicly available or consent-based accessible. The issue isn't that the data doesn't exist — it's that no analyst can manually aggregate it across 15 sources before the TAT deadline expires.
AI can do this in minutes.
Here's a concrete example. A mid-sized NBFC in Pune was evaluating a ₹80 lakh working capital loan for a textile exporter. The bureau checks came clean. GST filings showed consistent turnover. The CAM was positive.
What the credit officer didn't know: two of the borrower's three largest customers had filed for NCLT resolution in the prior six months. The borrower's receivables — nearly 40% of their assets — were effectively frozen.
That information was public. It was sitting in NCLT cause lists and court records. No one had the bandwidth to check.
The loan was approved. It turned NPA within nine months.
This isn't an unusual story. It's a structural gap in how Indian credit underwriting works.
What separates good underwriting from great underwriting is context — not just "can this borrower repay?" but "what is the environment they're operating in?"
That means:
Sector intelligence: Is the borrower's industry segment seeing payment cycle elongation? Are input costs rising faster than output prices? What's the historical NPA rate for this segment in this geography?
Ecosystem intelligence: Who are the borrower's key customers and suppliers? Are any of them under financial stress? Has there been any litigation between parties?
Promoter intelligence: What is the promoter's full business interest map? Are there related-party transactions that create risk? Are there prior defaults in entities they control?
Regulatory intelligence: Are there any RBI directives, sector-specific regulatory actions, or GST compliance flags that affect this borrower?
Individually, each of these requires hours of research. Collectively, most credit teams simply skip them — not because they're unimportant, but because there aren't enough hours in the day.
Modern AI systems can aggregate public data sources — NCLT records, MCA filings, GST compliance status, court records, RBI notifications, news signals — and surface the relevant signals for any borrower in minutes, not days.
This doesn't replace the credit officer. It amplifies them.
The analyst who used to spend four hours building a CAM from scratch now spends ninety minutes reviewing an AI-generated intelligence brief and adding their judgment. The credit committee sees more context, not less. Decisions improve. TAT drops. And the borrowers who were getting rejected for thin files — because no one had time to do proper due diligence — now get a fair assessment.
For NBFCs competing on speed, this is a competitive moat. For NBFCs competing on credit quality, it's a risk management imperative.
India's top private sector banks already have large analytics teams and proprietary data assets. They can afford to build this capability in-house. Most NBFCs cannot.
The opportunity is in giving mid-tier NBFCs — the ones disbursing ₹500 crore to ₹5,000 crore a year — access to the same quality of intelligence that their larger competitors have. Not as a consulting engagement that costs ₹50 lakh and takes six months. As a product they can use on every loan file.
That's the direction the industry is moving. The question for any NBFC leadership team is whether they're ahead of it or behind it.
At Dekrypt Labs, we're building BIOS — a Business Intelligence Operating System designed specifically for the Indian market. If your credit team is still spending hours aggregating data that should take minutes, we should talk. Visit dekryptlabs.com to learn more.