SME Underwriting: The 13 Signals Banks Miss Before a Default

Data, case, signal — 50 verified defaults fused with system-wide MSME stress data into one case-derived early-warning-signal (EWS) codebook.

The situation

This report fuses two evidence layers most SME risk work keeps separate. Layer 1 reads system-wide lending data and regulatory data and finds MSME stress concentrated in the smallest tickets, just over half public-sector-originated (52%). Layer 2 reads 50 verified court cases and finds the same concentration, plus what specifically failed. Layer 3 turns both into a 13-signal codebook a bank's own CAM review or EWS system can use.

System-wide lending data alone points at the risk: delinquency in the smallest MSME loans (under Rs 10 lakh) roughly doubled, from 2.0% to 4.1% (Mar-22 to Mar-24 cohorts), even as the overall MSME loan book improved — and public-sector banks hold 52% of that origination. The bureau's own reading names PSU underwriting of micro-MSME as the driver.

The court case evidence corroborates the risk directly, though only for public-sector lenders: all 50 verified defaults with an identifiable lender are public-sector, an SBI-associate, or a regional rural bank. That is not because private banks are exempt — private banks recover through SARFAESI, which produces no court record, and neither their internal write-off decisions nor RBI's divergence assessments for them are made public the way CAG audits and court filings are for public-sector guarantee schemes.

Fusing system-wide and regulatory lending data with court case evidence produces a 13-signal codebook, organised by loan-lifecycle stage: five signals to check before sanction, two governance signals at the sanction decision itself, and six signals for after disbursement. A peer-reviewed decision-fatigue study (Baer & Schnall, 2021) independently corroborates the governance-stage finding — credit decisions vary by the clock, not just the file.

Key findings

What the origination file already showed.

13 signals, ranked by how often they recur across 50 verified cases.

Case-level presence — a signal counted once per case if it appeared at least once, not an occurrence count. The five largest are a strong within-sample pattern; the smaller signals are documented occurrences, not system-wide rates.

Sources: 50 individually-cited court and disciplinary case judgments (indiankanoon.org). Analysis: ZenLearn Research.

In Sudhir Kumar Arora's case, a Corporation Bank loan-processing officer testified that "there was a lot of pressure from Pavan Arya to process the CVPOD loan within a day and not to put many objections" — an overdraft later found sanctioned on grossly inflated collateral. The same Pavan Arya was separately convicted in his own case at the same centralised processing centre: a Rs 5 crore cash-credit facility sanctioned after a CIBIL score of minus one was struck off the file without any reason recorded, and after the loan amount was reduced from Rs 7 crore specifically to stay under a higher-scrutiny audit threshold (CBI vs Pavan Arya & Ors., CC No. 61/2019). Two independently adjudicated cases, naming the same individual, from the same centralised processing centre — this is not a single anomalous file, and it is exactly the kind of decision an override-justification trail would have caught at the time it was made, not years later in a courtroom.

What's in the report

Ten sections, 43 pages, every case cited to a public document.

The next step is in your own book.

We can benchmark 20 of your own recently-sanctioned sub-Rs 10 lakh SME files against this codebook's most frequent gaps — document verification and valuation mismatch — and give you a written pass-rate readout for your credit committee.