Why automated rationale is harder than automated decision
Generating a credit decision from 40 rule checks is a computation problem. Generating a rationale that a borrower can understand, that a regulator would accept, and that a human underwriter can act on is a communication problem. The Credit Decision Agent AI solves both simultaneously: the rule execution produces the decision; the rationale generation layer converts the rule outcomes into the appropriate communication for each audience — borrower, underwriter, or compliance record.
The borrower communication must be in plain language without technical jargon, must name the specific adverse factor rather than citing a generic "credit policy" reference, must be accurate enough to satisfy the ECOA / Regulation B disclosure obligation, and must include — where possible — a constructive path forward. The underwriter brief must be precise and evidence-referenced, providing the information needed to make a judgment without requiring the underwriter to reassemble it from raw data. The compliance record must contain the complete rule-by-rule output with scores and thresholds for every rule that was evaluated.
The three rationale outputs — and what each one contains
Dear Priya,
We are pleased to confirm that your home loan application for $28,00,000 has been approved in principle. Here is a summary of your approved terms.
Scorecard: 87/100. Decision: Auto-Approve. Rule execution: 40/40 rules evaluated. KO failures: 0. Refer triggers: 0. Positive factors: income trend +37% (12 months), FICO 736 improving, 0 DPD (24 months), cash flow score 84, collateral coverage 146%. Neutral factors: savings rate 21% (adequate, below high-savings threshold of 25%). No negative factors above soft-flag threshold. All 40 rules documented in execution log. Decision time: 41 seconds. Underwriter sign-off required for formal approval letter.
Dear Vinod,
Thank you for applying for a personal loan of $8,00,000. We have completed our assessment and are unable to approve this application at this time. We want to explain specifically why.
This decision was made using an automated credit assessment system. You have the right to request a review by a human underwriter within 30 days. Data sources used: FICO, Chase bank statement (12 months), application form. For grievances: [contact] | For data queries (CCPA / state privacy laws): [contact]
Score: 44/100. Decision: Decline. KO failures: Rule 17 (DPD 90+ in 12 months — FICO confirms SBI credit card DPD 94 days, July 2025), Rule 09 (DTI at proposed amount: 52.4% — exceeds 45% ceiling). Scorecard: KO cascade — KO failures override scorecard total. FICO: 612 (below 650 floor — would also trigger KO under Rule 16 independently). All 40 rules executed. Decision time: 34 seconds. Human review right: 30 days. Disclosure: ECOA / Regulation B–compliant borrower communication generated and dispatched.
Counter-offer analysis: At $4.5L loan amount, DTI reduces to 44.1% — within policy. However, Rule 17 (DPD 90+) remains a KO regardless of loan amount. Counter-offer not viable at this time due to independent KO on Rule 17. Counter-offer pathway opens when DPD event ages beyond 12-month window (July 2026). Recommended re-application window: August 2026, subject to clean payment record maintenance until then.
The rationale is not a formality — it is the most important output of the decision
An approval without a clear rationale produces a borrower who does not understand what their relationship with the institution is based on. A decline without a specific, actionable rationale produces a borrower who cannot improve their application and a regulator who cannot verify compliance. The Credit Decision Agent AI generates all three outputs — borrower communication, underwriter brief, compliance record — as an integral part of the decision, not as a downstream administrative task. The decision and its documentation are produced simultaneously, in 41 seconds, for every application.