Credit-Underwriting Safeguards for Banks
Scorecards, bureau data and control-maps all assume the officer applies the rule correctly — none of them generate evidence of what happens when a credit officer overrides a flag, before that override shows up as an NPA.
The situation
Every bank measures borrower quality through bureau scores, scorecards and risk-control self-assessments. None of these tools generate pre-loss evidence of what happens at the point a credit officer misinterprets a qualitative input, verifies a file incompletely, rationalises a policy exception, suppresses a risk escalation, or overrides a bureau flag — before that decision failure crystallises as an NPA. This report reads that unobserved layer.
In India, 51.9% of new retail non-performing assets in H1 FY25 originated in unsecured books — the segment where officer-level override decisions concentrate most — yet no bank publicly discloses officer-level pre-loss decision data. The underwriting decision surface is defined by five distinct failure modes, each with a direct route to IRACP provisioning and, for Material Risk Takers, to compensation clawback. A scan across seven major banking supervisory jurisdictions (EU, UK, US, Singapore, Hong Kong, Australia, India) confirms a clear regulatory trajectory toward individual accountability and override tracking.
The scale of the unmeasured judgment layer is not hypothetical: across 26 major European banks, management overlays on expected credit losses — the post-model adjustments banks apply when statistical models miss novel risk — ranged from 7% to 35% of total ECL, averaging 12% at year-end 2023. That is human judgment applied at portfolio scale, with no equivalent instrument measuring the calibration of the individual officer making the call.
Key findings
What the origination file already showed.
51.9% of new retail NPAs in H1 FY25 originated in the unsecured segment — the segment where officer-level overrides concentrate most — and no bank publicly discloses officer-level pre-loss override data to explain why.
Misinterpretation of qualitative inputs, incomplete verification, policy-exception rationalisation, escalation suppression and bureau-flag override each map to IRACP provisioning outcomes (15% secured / 100% unsecured slippage) and, for Material Risk Takers, conditional malus or clawback under the RBI's November 2019 Compensation Guidelines.
Across 26 major European banks, management overlays on expected credit losses averaged 12% of total ECL at YE2023, ranging from 7% to 35% bank to bank (Forvis Mazars, 2024). The overlay is judgment at scale; nothing measures whether that judgment is calibrated.
Of the seven jurisdictions scanned, three mandate individual accountability frameworks (UK's SM&CR, Singapore's MAS IAC Guidelines, Hong Kong's HKMA ECF-CRM certification) and two explicitly mandate override/deviation tracking (the EU's EBA/GL/2020/06 and the US Shared National Credit programme).
A gold-loan NBFC's assets under management contracted 53% within five months of an RBI supervisory action over collateral-certification deviations; microfinance borrowers holding five or more active loan accounts rose 17.2% between 2022 and 2024, and sector gross NPAs rose from 8.8% to 16% of gross loan portfolio by March 2025. In each case the override behaviour was observable before the loss — the gap was instrumentation, not data.
The portfolio signal was visible for three years before anyone measured the officer.
The US Shared National Credit programme tracks a portfolio-level non-pass rate — loans regulators tag Special Mention, Substandard or Doubtful — across large syndicated bank loans. The rate rose for three straight years without any accompanying officer-level override data to explain which decisions drove it.
Non-pass includes Special Mention and Classified categories; leveraged loans comprised 81% of all non-pass loans in 2025. Source: Federal Reserve / FDIC / OCC Shared National Credit Program Reports, 2022–2025.
On 4 March 2024, the Reserve Bank of India barred IIFL Finance, a large gold-loan NBFC, from sanctioning any new gold loans after a supervisory examination found 'serious deviations in the process of certifying purity and net weight' of gold collateral. The company's gold-loan assets under management contracted 53% within five months — from ₹26,081 crore to ₹12,162 crore — alongside a ₹618 crore swing in consolidated profit after tax. The pre-loss signal that was never collected: an inter-rater consistency score across collateral assessors. Officers were applying divergent assay standards at branch level; a calibration test run across branches would have surfaced the inconsistency months before the supervisory notice.
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 run a pilot process stress-test across three of your retail lending hubs — a 200-300 loan cohort, scenario-based simulation, and a Judgment Risk Map with an inter-rater consistency score and adjudication cycle time — with a first override-distribution readout in about 30 days.
RBI Credit Risk Management Directions 2025, RBI Digital Lending Directions 2025, RBI Master Circulars on Basel III Capital Regulations and IRACP, RBI Compensation Guidelines 2019, the RBI Risk-Based Supervision (RBSA) Framework and RBI press releases on supervisory action — set alongside the BCBS Principles for the Management of Credit Risk (2025), the EBA Guidelines on Loan Origination and Monitoring, the UK FCA's Senior Managers & Certification Regime, MAS Individual Accountability and Conduct Guidelines, HKMA's Enhanced Competency Framework, APRA CPS 220, US OCC/Federal Reserve credit-risk guidance and Shared National Credit Program reports, and the audited annual reports and Pillar 3 disclosures of SBI, ICICI Bank and Kotak Mahindra Bank. Full reference list in the report.