How an NBFC Cut First-EMI Default by 36% — by Rejecting Just 3% of Applicants
A mid-sized NBFC lending partner reduced first-EMI default on its personal loan book by 36% across two consecutive loan disbursal cycles. The only change: declining the riskiest 3% of an applicant pool that had already cleared bureau and income checks. Approval criteria, ticket size, tenure, and pricing stayed identical across both cycles — the decline decision was the sole variable.
First-EMI default is a portfolio health signal, not just a bad loan
Most underwriting reviews focus on overall default rate. First-EMI default — a borrower missing their very first repayment — deserves its own line item, because it behaves differently from default further into a loan’s life. It shows up almost immediately, so it’s the fastest signal a lender gets that a cohort was mispriced or misjudged at origination. It’s disproportionately expensive to collect against, since there’s no repayment history to work with and no established contact pattern with the borrower. And it compounds: a first-EMI default is a strong predictor of full default later in the loan, not an isolated miss.
For an NBFC running thousands of personal loans a month, first-EMI default rate is one of the cleanest early-warning numbers available — cut it, and portfolio quality improves before collections ever gets involved.
What changed between cycle 1 and cycle 2
The NBFC’s bureau-and-income-based BRE (business rules engine) didn’t change between the two cycles. What changed was a single addition: every applicant who cleared the existing BRE was also scored by Tenacio Sense on signals the bureau doesn’t capture — contactability across channels, digital footprint consistency, and housing-strata/location risk. The bottom-scoring 3% of otherwise-approved applicants were declined. Everyone else moved through disbursement exactly as before.
That’s the part worth sitting with: this wasn’t a policy tightening. Approval criteria didn’t move. The BRE didn’t move. Pricing didn’t move. One layer was added after the existing rules engine, and it touched 3 applicants out of every 100.
Why 3% was enough
It’s tempting to assume default reduction scales with rejection rate — reject more, default falls further. In practice, most of a loan book’s risk concentrates in a small, identifiable slice that bureau data alone can’t separate from the rest of the pool. The other 97% of applicants who cleared the bureau check were, by and large, genuinely creditworthy — rejecting them would only have cost approval volume without improving the number that matters.
This is the same logic behind Sense’s swap-out capability on Tenacio’s Lending & Banks page: identify the applicants who look acceptable on bureau data but carry hidden risk — inconsistent contact information, address signals that don’t line up with financial stability, or a digital footprint that doesn’t match the stated profile — and remove only them. In a separate lending book, the same swap-in/swap-out approach has expanded approvals by 22% by rescuing borderline applicants the bureau score alone would have rejected. The mechanism runs both directions: it can tighten a book by 3% or widen one by 22%, depending on where the risk actually sits.
Where Sense fits
This result came from Sense’s fraud and identity risk signal, contactability scoring, and location intelligence — the same signal set Sense uses as a second line of defence after a BRE approval, applied here to credit risk rather than fraud. The underlying idea is consistent across both use cases: a bureau score and an income check tell you whether an applicant looks acceptable on paper. They don’t tell you whether the phone number is reachable, whether the declared address matches the applicant’s actual digital and location footprint, or whether the neighbourhood’s repayment history suggests hidden risk. Sense scores that layer, continuously, and hands the BRE one more input before disbursement — not instead of the bureau pull and Connect’s income verification, but after them.
How this compares to other underwriting approaches
A few other India-focused platforms work adjacent territory here, and each is genuinely strong at what it does.
Perfios has arguably the deepest bench in bank-statement and financial-document analysis at scale, serving over a thousand financial institutions. Its strength is reading what an applicant reports — parsing bank statements, GST, and ITR data into a creditworthiness view. Sense works from a different angle: independently observable signals — contactability, location, digital-footprint consistency — that don’t depend on a document the applicant submitted. The two are closer to complementary than competing.
FinBox is probably the closest true peer in alternate-data underwriting. DeviceConnect and the Inclusion Score take a genuinely broad, device- and app-behaviour-centric approach to scoring thin-file borrowers, and FinBox’s embedded-lending infrastructure is well regarded in the market. Where Sense differs is in the signal mix — location and housing-strata intelligence and multi-channel contactability sit alongside device signals as first-class inputs — and in reach: the same signal set follows a customer past underwriting into collections-stage recovery, rather than stopping at the origination decision.
SurePass launched Finpass this year, a credible and fast-moving move into AI underwriting worth taking seriously rather than dismissing — Finpass Lens’s multi-bureau portfolio monitoring is a genuinely useful layer for tracking risk after disbursement. This case study sits at a different moment in the loan lifecycle: the pre-disbursement decision, not post-disbursement monitoring.
IDfy brings the strongest identity-verification and DPDP-compliance credibility of the group, and to their credit, most of their public positioning leads with that rather than overreaching into underwriting claims. This result is squarely a credit-risk-scoring outcome, which isn’t IDfy’s core lane — they play a different, adjacent role in the same lending stack, at the KYC and consent layer.
What to check before you try this
A few things worth confirming before layering a similar approach onto an existing BRE:
- Isolate the variable. The credibility of a result like this comes from changing exactly one thing. If approval criteria, pricing, or ticket size move at the same time, the result becomes unreadable.
- Measure across at least two cycles. A single cycle can be noise. Two consecutive cycles showing the same direction is a much stronger signal than one.
- Check the decline rate against the result, not just the result. A 36% improvement means something different at a 3% decline rate than it would at a 15% one — the ratio is the actual proof point, not the headline number alone.
- Confirm the signal set doesn’t duplicate the bureau check. The value here came from signals the BRE didn’t already have — contactability and location risk, not a second bureau pull.
FAQ
What is first-EMI default and why does it matter for NBFCs? First-EMI default is when a borrower misses their very first loan repayment. It matters because it’s the earliest available signal that a cohort was mispriced or misjudged at underwriting, and because it’s a strong predictor of full default later in the loan.
How can alternate data reduce early loan delinquency? By scoring signals a bureau report doesn’t capture — contactability, digital footprint consistency, and location or housing-strata risk — after a bureau-based BRE has already approved an applicant, and removing only the highest-risk slice before disbursement.
What’s the difference between bureau-only and alternate-data underwriting? Bureau-only underwriting relies on credit history and reported income. Alternate-data underwriting adds independently observable signals on top of that — contactability, device and behavioural data, or location intelligence — to catch risk the bureau report doesn’t reflect, particularly for thin-file or new-to-credit applicants.
How much of a loan portfolio should you decline to cut default meaningfully? There’s no fixed number — it depends on where the risk concentrates in a given book. In this case, 3% was enough because the risk was concentrated in a small, identifiable slice; a book with more diffuse risk might need a different rate. The ratio of decline rate to default improvement matters more than either number alone.
Does rejecting more applicants always reduce default risk? No. Most of a loan book’s risk typically concentrates in a small slice that bureau data can’t separate out. Rejecting broadly beyond that slice mainly costs approval volume without a proportional improvement in default — precision matters more than volume of rejections.
Curious what your bottom 3% looks like? Talk to us about running a portfolio sample through Sense →
Don’t wait for a bad cycle to find your risk concentration. Get in touch to score your first applicant cohort →