Not all credit exceptions require the same human attention. A CIBIL score of 672 against a 680 floor, on a borrower with 28 consecutive zero-DPD months on an existing loan, strong GST income, and a well-secured property, is an exception with so many compensating factors that routing it to the ZCM for a 45-minute review is an inefficient use of senior credit officer time. A combined CIBIL and FOIR exception on a new borrower with no existing relationship, a single compensating factor, and a source from a DSA whose exception cohort DPD has been flagged as elevated — that one requires careful human review. The Credit Exception Agent AI distinguishes between these cases and routes each exception to the appropriate decision-maker: auto-approval, designated credit authority, or CRO escalation.
The routing framework: what determines the level
The routing decision is a function of three inputs. The first is exception severity — the magnitude and number of policy parameters breached. A single parameter breach of less than 10% is lower severity; a combined breach of two or more parameters, or a single parameter breach exceeding 20%, is higher severity and requires higher authority. The second is compensating factor strength — a quantitative score computed from the presence, verifiability, and historical correlation with performance of each compensating factor cited. Strong compensating factors (existing zero-DPD relationship, verified GST growth, property security with LTV below 60%) reduce the routing level; weak or absent compensating factors raise it. The third is risk modifiers — flags that override the severity-and-compensating calculation regardless of its outcome. An elevated DSA-channel DPD flag, a concentration warning (this exception would breach the portfolio exception limit), or a governance flag (this officer has had elevated exception approval rates flagged) all trigger escalation regardless of the individual exception's profile.
The routing decision engine: live case examples
The auto-approval category: what it means and what it does not mean
Auto-approval routes the exception to a designated approver with a pre-computed recommendation — it does not approve the exception autonomously
The term "auto-approval" in the Credit Exception AI context does not mean the AI approves the loan without human review. It means the exception is routed to a designated approver (typically ZCM for standard exceptions within their authority) with a system-generated recommendation of "approve" based on the compensating factor analysis, and the approver's expected review time for this profile is shorter because the documentation is complete and the recommendation is supported. The approver still makes the decision. The AI's role is documentation, routing, and recommendation — not approval.
→ Every exception requires a human approver at the appropriate authority level · "Auto-approval path" = pre-documented, pre-recommended · Not autonomous approvalThe routing engine enforces the Board-approved authority matrix — it cannot route exceptions to approvers whose authority does not cover the exception type
The authority matrix is encoded in the Credit Exception AI's routing rules. A Branch Manager cannot approve a combined CIBIL + FOIR exception of greater than ₹10L — the routing engine will not route to a Branch Manager for such a case, regardless of the RM's preference. If the RM submits an exception directly to the Branch Manager without AI routing, the exception record flags the approval as outside the authority matrix — a governance exception that is reported to the compliance function independently of the credit exception.
→ Authority bypass detection: if an exception is approved outside the AI routing path, the governance flag is automatic · Cannot be suppressed by the approverCorrectly routed exceptions with complete documentation are approved 3× faster than manually routed ones — because the approver does not need to gather information
Before the Credit Exception AI, the ZCM's review of an exception typically took 45 to 90 minutes — most of which was spent gathering information: calling the RM for the GST data, requesting the property valuation reference, confirming the existing loan DPD history from CBS. With the Credit Exception AI pre-populating all 9 documentation fields from system data at the time of submission, the ZCM review takes 12 to 20 minutes — spent actually assessing the credit case rather than assembling it. Average exception turnaround time has fallen from 3.2 days to 1.1 days since the Credit Exception AI was deployed.
→ Exception TAT: 3.2 days → 1.1 days · ZCM review time: 45–90 min → 12–20 min · Same credit quality · Better use of senior credit officer timeA denial recommendation from the Credit Exception AI triggers a clear escalation path — it does not end the borrower's application
When the Credit Exception AI recommends denial (as in Case C above), the RM receives a structured notification: the specific reasons for denial (compensating factor score too low, portfolio concentration breach), the escalation option (CRO override with documented reasons), and the alternative path (advise the borrower on what would change the credit profile to make them eligible — CIBIL improvement timeline, additional security, or income documentation that would close the compensating factor gap). The denial is not a closed door — it is a documented position with a clear pathway to reconsideration.
→ Denial recommendation + CRO escalation path + borrower rehabilitation guidance · Not a black box rejection · Always a documented rationaleThe Credit Exception AI does not make credit decisions — it makes credit decisions faster, better-documented, and at the right authority level
The distinction is important. Every approval in the credit exception process is made by a human authorised by the Board to make that type of approval. The Credit Exception AI's role is to ensure that the human who makes the decision has complete, source-attributed information at the moment of review; that the decision is being made at the correct authority level; that risk modifiers that should trigger escalation actually trigger it, even when the individual RM or credit officer does not notice them; and that the decision — whatever it is — is recorded in a complete, immutable, inspection-ready record within minutes of the approval. The institution that deploys the Credit Exception Agent AI does not reduce its credit governance — it makes its credit governance operate the way it was designed to, consistently, at every exception, with no gaps, no documentation shortcuts, and no authority bypasses that only appear in the next RBI examination.