Use case #0003

Fair Lending AI Documentation: The Board Report That Satisfies Regulators

A fair lending programme without a board report is a measurement exercise. A board report without a fair lending programme is a document. The Fair Lending AI produces the report that bridges these two failures — a quarterly board fair lending report that contains the actual analysis, the actual findings, the actual remediation actions, and the actual evidence that the institution is actively managing its fair lending obligations rather than asserting that it is.

A fair lending programme without a board report is a measurement exercise. A board report without a fair lending programme is a document. The Fair Lending AI produces the report that bridges these two failures — a quarterly board fair lending report that contains the actual analysis, the actual findings, the actual remediation actions, and the actual evidence that the institution is actively managing its fair lending obligations rather than asserting that it is.

What Regulators Look for in a Board Fair Lending Report

An RBI inspection team reviewing an institution's fair lending governance asks a specific set of questions. Does the Board receive regular reports on fair lending metrics — and do those reports contain substantive analysis rather than declarations of compliance? Does the report identify specific disparities and their root causes rather than asserting that no disparities exist? Does it document remediation actions for any findings, with timelines and outcome tracking? Does it demonstrate that the Board has reviewed, questioned, and approved the fair lending programme — not merely received a document?

These questions are not currently codified as mandatory quarterly requirements for NBFCs. But they reflect the direction of RBI supervisory expectations — visible in examination findings at peer institutions and in the RBI's escalating focus on algorithmic fairness in credit. The institution that can already answer all four questions affirmatively is the institution that will pass this examination before it becomes mandatory.

"A board report that says 'we are committed to fair lending' is not a fair lending report. A board report that says 'the four-fifths ratio for SE Female applicants was 0.703 in Q3, below our 0.80 threshold, and here is the remediation plan with an 8-week timeline' — that is a fair lending report." — Fair Lending AI Documentation Module

The Quarterly Board Fair Lending Report: Live Example

Fair Lending Report — Q3 FY2026 · Board Risk Committee
Fair Lending AI · Auto-Generated · November 14, 2025
14Tests passed
3Under remediation
1New finding this quarter
18Total tests this quarter
IN PROGRESSOverall programme status

This report presents the results of the Q3 FY2026 fair lending analysis across all home loan originations (17,155 applications). The analysis covered 18 disparate impact tests across 7 protected groups and 4 metrics, plus a proxy variable audit across all 42 model variables. One new finding emerged this quarter: SE Female applicants failed the four-fifths threshold at a ratio of 0.703 (threshold: 0.800). Three previously identified findings remain under active remediation. Fourteen tests returned clean results. The Board is asked to note the new finding and approve the remediation plan presented in Section 4.

Gender approval rate Salaried Female 0.904 vs baseline — within threshold. Q3 improvement from Q2 (0.891). Pass
SE Female (approval) Ratio 0.703 vs threshold 0.800. New finding Q3. 1,284 applicants in scope. Matched-pair residual gap 11.4pp after income and bureau controls — indicates structural model bias, not risk difference. Fails — New
Gender loan size SE Female approved loans 28.6% smaller than matched Salaried Male approvals. Four-fifths on loan size: 0.714. Under active remediation — credit policy amendment in progress. Remediation
Pricing parity (Band B+) SE Female Band B+ applicants priced 18bps above matched Salaried Male Band B+ — not risk-justified. Pricing rule amendment approved by CRO, effective December 1. Remediation — Dec 1
Geography — Tier 2 Ratio 0.897 — within threshold. Slight improvement from Q2 (0.881) — attributed to expanded DSA network in Tier 2 cities. Pass
Name cluster disparity Ratio 0.926 — well within threshold. Proxy correlation on name-adjacent features: none detected. Model version v5.0 shows further improvement (0.944) — additional support for challenger promotion. Pass

The quarterly proxy correlation audit was run across all 42 model variables against religion, caste-proximate, and community geographic data. Two pincodes in the Mumbai portfolio (400008, 400012) showed proxy correlations above the 0.70 threshold with religious community composition data. The credit policy variable using raw pincode has been flagged for replacement with an income-quartile-adjusted property market index, which preserves the legitimate property-risk content without carrying the proxy correlation. Implementation is scheduled for the Q4 model update. Two variables added in the September model recalibration (business registration state and mobile carrier) were tested pre-deployment and returned clean proxy results.

The matched-pair analysis confirms that 11.4 percentage points of the SE Female approval gap are attributable to model structure rather than legitimate risk difference. The root cause is the combined weight of the employment sector variable (CSI: 0.31, currently drifting — separate issue), the income documentation type variable (ITR penalised vs salary slip), and bank balance volatility variable. Proposed remediation: an SE-Female-aware model recalibration that re-weights the income documentation variable to treat ITR on equal terms with Form 16 where the GST turnover trend is strong. Estimated timeline: 8 weeks to recalibration, 4 weeks challenger test, Board approval December BRC meeting. In the interim: all SE Female declinations in Band B−/C+ to be routed for manual underwriter review before final rejection — effective immediately.

The Regulatory Framework the Report Addresses

Regulatory Basis Specific Requirement How the Board Report Satisfies It
RBI Fair Practices Code 4 Fair and non-discriminatory lending — no differentiation based on caste, religion, gender Monthly disparate impact analysis provides positive evidence of non-discrimination — not just an assertion
RBI Corporate Governance Circular Board must oversee fair practices — not delegate entirely to management Quarterly Board report with Board sign-off required — creates documented governance trail
DPDP Act 2023 11 Automated decisions must not produce discriminatory outcomes based on protected characteristics Proxy variable audit confirms model variables are not proxies for protected characteristics — documented quarterly
Constitutional Arts. 14–16 Non-discrimination in contractual dealings — fair lending as constitutional obligation Four-fifths analysis across all constitutionally protected categories — documented evidence of compliance
RBI Model Risk Management Circular Model validation must include fairness assessment — not just performance metrics Fairness findings from board report feed into model card and annual validation report — integrated governance

What the Board Must Do With This Report — and Why It Matters

Receiving a fair lending report is necessary but not sufficient for board governance. The Board Risk Committee must actively review the findings, ask questions about the remediation plans, set timelines, and formally approve or reject the recommendations. The minute of that meeting — recording that the Board reviewed specific findings and approved specific remediation actions — is the governance document that demonstrates the Board is discharging its oversight responsibility.

The Fair Lending AI generates the board resolution template alongside the report: a specific resolution for each finding requiring Board action, drafted in the language appropriate for board minutes, ready for the company secretary to adapt for the meeting record. This template ensures that the governance record matches the governance substance — the board meeting where the fair lending report was discussed produces a minute that an inspector can read and confirm that the Board engaged substantively with the findings.

QuarterlyBoard report cadence — auto-generated from live fair lending analysis, never assembled manually
18Tests covered in Q3 report — 4 metrics × 7 protected groups, plus proxy variable audit
BoardResolution template included — ensures governance record matches the governance substance
5Regulatory frameworks addressed — Fair Practices, Corporate Governance, DPDP, Constitution, MRM

The Board Report Is the Evidence That Governance Happened — Not the Governance Itself

Fair lending governance happens when an institution measures disparate impact, identifies proxy variables, investigates findings, and remediates them with accountability. The board report is the record that all of this happened — in a form the board can review, approve, and be held accountable for. An institution that can present a quarterly board fair lending report containing specific findings, matched-pair analyses, proxy audit results, and concrete remediation plans with outcome tracking is demonstrating a governance culture that no declaration of commitment to fair lending can substitute for. That is the report that satisfies regulators — not because it says the right things, but because it proves the right practices.

← Back to Fair Lending Agent AI