Every credit override — every loan sanctioned outside the parameters of the credit policy — creates a regulatory obligation. The RBI's supervisory expectations under the Internal Ratings-Based approach and the Fair Practices Code require that exceptions to credit policy be documented with the reason, the authority level of the approver, the compensating factors that justified the exception, and the outcome data that allows the supervisor to assess whether the exception population performed differently from the standard portfolio. The Credit Exception Agent AI creates this documentation automatically, at the moment of override, for every exception — so when the RBI inspection team asks for the exception register, it already exists, is complete, and contains every field the examination team needs.
What RBI examiners look for in a credit exception file
A credit exception, in the RBI supervisory context, is any loan where the sanctioning authority approved terms that deviate from the Board-approved credit policy. This includes CIBIL score exceptions (loan sanctioned below the policy minimum), FOIR exceptions (loan sanctioned where the borrower's fixed obligation-to-income ratio exceeds the policy ceiling), LTV exceptions (security value lower than the policy minimum), tenure exceptions, and product-combination exceptions (lending to a borrower who has existing delinquency at the institution in a different product).
RBI examination teams assess exceptions on four dimensions. First, documentation completeness: does the exception file record the specific policy parameter that was breached, the magnitude of the breach, the compensating factor(s) cited, and the authority level of the approver? Second, authority appropriateness: was the override approved at the level the Board has delegated for that type and magnitude of exception — or was it approved below that level? Third, concentration: is a disproportionate share of exceptions originating from a single geography, product line, officer, or DSA — suggesting systematic policy avoidance rather than case-by-case judgement? Fourth, performance: how did the exception population perform versus the standard portfolio? An institution that has high exception rates and higher exception DPD relative to standard portfolio DPD has a credit governance problem that the examiner will escalate.
The exception record: every field, populated at the moment of override
The documentation fields that make the difference in an RBI inspection
Three fields in the exception record are the ones that RBI examiners most consistently flag when they are absent or inadequate. The first is the compensating factor — not the conclusion ("creditworthy despite low CIBIL") but the specific, verifiable evidence ("GST revenue grew 61.9% YoY, verified against GSTN data"). The examiner does not take the credit team's word for the compensating factor — they look for the data source that supports it. The Credit Exception Agent AI pulls the compensating factor data from the institution's systems at the time of the override and embeds it in the record with source attribution: GSTN data as of Nov 14, CBS account data as of Nov 14, valuation report reference number.
The second is the authority check — confirmation that the approver's authority level covers this type and magnitude of exception. Many institutions have well-designed authority matrices that are inconsistently applied in practice — a Branch Manager approves a combined exception that requires ZCM authority because the ZCM is not available and the Branch Manager did not realise the exception required escalation. The Credit Exception Agent AI runs the authority check automatically before routing — preventing authority breaches at the point of approval rather than discovering them in an audit.
The third is the performance tracking flag — the linkage between the exception record and the subsequent DPD data that allows the institution to demonstrate to the examiner that it monitors how exceptions perform. An exception register that contains 847 approved exceptions from the last 12 months but no DPD outcome data for any of them is a register that tells the examiner the institution approves exceptions but does not know whether those exceptions were good decisions.
The examiner who finds 847 complete exception records with source-attributed compensating factors, timestamped approval chains, and 90-day DPD outcomes has found a credit governance programme that is working — not one that is hiding something
The RBI's supervisory concerns about credit exceptions are fundamentally concerns about whether the institution knows what its exceptions are, why they were approved, who approved them, and how they performed. An institution that can answer all four questions instantly — because the Credit Exception Agent AI has been building the exception register automatically since the last inspection — demonstrates a credit governance function that is continuously operating, not one that was assembled in the 6 weeks before the inspection visit. The Credit Exception AI's documentation function is not about compliance theatre. It is about building a credit governance infrastructure that is inspection-ready at all times because it is genuinely being operated at all times.