The Settlement Agent AI can approve settlements within defined parameters without human involvement — and in the majority of early-stage cases, it does. But the settlement function is the one area of lending operations where the consequences of a wrong decision are both material and irreversible. A credit that is declined can be re-approved. A settlement that is over-generous cannot be reversed — the waiver is gone. The escalation logic that determines when the AI hands the decision to a human credit officer is therefore the most important design decision in the system.
The three-tier approval structure
Settlement authority follows the same principle that governs all authority delegation in regulated lending: the authority to approve a decision must be proportional to the financial consequence and the complexity of that decision. A settlement of a ₹3 lakh Sub-Standard personal loan that waives only penal charges is a routine operational decision that a trained system can approve correctly every time. A settlement of a ₹2.8 crore Doubtful D2 MSME loan that waives 40% of the principal requires credit officer judgment — not because the eligibility calculation is wrong, but because the decision has consequences that require a human to own and is of a magnitude that demands human accountability.
The three tiers are: Settlement AI auto-approval (for cases within defined parameters, where the economics clearly support settlement and the waiver is within policy limits); credit officer approval (for cases above the auto-approval threshold, involving principal waivers, or where the Settlement AI's assessment flags a complexity that warrants human review); and credit committee or Board approval (for large exposures, loss assets, or settlements involving connected parties or regulatory sensitivity).
The four-tier escalation framework
System approves without human review — within defined policy parameters
The Settlement AI approves the settlement and generates all documentation automatically. No human action is required. The credit officer receives a notification that a settlement was approved — not a request for approval. The notification contains the settlement terms, the eligibility calculation, and the enforcement NPV comparison that justified the approval. The credit officer can review and overturn within 24 hours if they identify an error.
→ Parameters: Sub-Standard or Watch · Total waiver ≤₹5L · No principal waiver · Enforcement NPV clearly exceeded · No connected partySettlement AI prepares the brief — credit officer makes the decision
The Settlement AI computes the eligibility, compiles the brief (account history, enforcement NPV, settlement terms, waiver justification), and routes to the designated credit officer. The credit officer's decision task is 15–20 minutes of review — not assembly. The brief contains everything needed to make the decision: the AI's recommendation, the dissenting considerations (why the AI flagged for escalation), and the one-click approve or modify interface.
→ Parameters: D1 Doubtful · Any principal waiver · Total waiver ₹5–25L · Enforcement NPV margin <₹5L · Any ambiguity in borrower capacity assessmentFull committee review — Settlement AI provides the analysis pack
For large or complex settlements, the credit committee reviews the full analysis pack compiled by the Settlement AI. The pack includes: account history since origination, the complete enforcement value analysis (what SARFAESI has returned, what DRT proceedings have cost, what the auction reserve price was set at), the settlement terms and waiver justification, and the credit officer's recommendation (who reviewed it at Tier 2 first). The committee's role is governance, not analysis — the analysis is already done.
→ Parameters: D2 Doubtful or worse · Total exposure >₹1Cr · Total waiver >₹25L · Any settlement with a connected party or co-borrower disputeSettlement of loss assets, very large exposures, or regulatory-sensitivity cases
Loss assets, exposures above ₹5 crore, or any settlement that involves a write-off of more than 50% of the principal outstanding require Board or MD sign-off. These cases are presented to the Board with the full Settlement AI analysis, the credit committee's recommendation, and a comparison of what the institution's expected recovery is under every available option. Board approval is a governance obligation — the Settlement AI's role is to ensure the Board receives analysis that is complete, accurate, and presented in a format that enables a governance decision rather than requiring the Board to assemble the facts itself.
→ Parameters: Loss asset any amount · Total exposure >₹5Cr · Principal waiver >50% · Connected party settlement · Regulatory sensitivityThe 8 escalation triggers — cases where the AI cannot approve regardless of financial parameters
Connected party settlements carry conflict of interest risk — always escalate regardless of amount
A settlement offered to a company where a director of the institution holds a significant interest, or where the borrower is a relative of an employee in the credit chain, carries reputational and regulatory risk that cannot be assessed algorithmically. All connected party settlements escalate to credit committee minimum, Board if above ₹50 lakhs.
→ Escalation: credit committee minimum · Board if >₹50L · Conflict of interest declaration requiredSettlement of a fraud-flagged account requires legal review before financial concession
Settling with a borrower who may have defrauded the institution waives rights that the institution may need to preserve for FIR or civil litigation. The Settlement AI flags accounts in the fraud-suspect register and escalates to the legal team before any settlement terms are discussed. No settlement offer is made to a fraud-flagged account without legal clearance.
→ Escalation: legal team first · No settlement offer without legal clearanceWhen settlement barely beats enforcement, human judgement should confirm the preference
A settlement that recovers ₹17L when enforcement NPV is ₹16.7L has a margin of only ₹30,000. The margin is positive — settlement is technically better — but it is so thin that the estimation uncertainty in the enforcement NPV (±₹2–3L depending on auction outcomes) makes the comparison unreliable. The Settlement AI flags this as a thin-margin case and escalates to credit officer.
→ Escalation: credit officer · Settlement margin <₹3L over enforcement NPVIf the borrower claims capacity beyond what the field assessment shows, escalate before committing terms
A borrower who says they can pay ₹28L but whose field assessment shows a business that appears dormant and a property that shows signs of neglect is claiming capacity that the evidence does not support. The Settlement AI flags the discrepancy and escalates to credit officer — who may decide to pursue enforcement rather than offer settlement to a borrower whose stated capacity may be overstated.
→ Escalation: credit officer · Verify capacity before settlement offer is madeA borrower who has settled once and defaulted again requires human review before a second settlement
A second settlement for the same borrower creates both an adverse precedent (the borrower learns that default is followed by a settlement offer) and a pattern of credit behaviour that the credit officer must assess. The Settlement AI flags repeat-settlement cases and requires credit officer review regardless of the financial parameters.
→ Escalation: credit officer mandatory · Prior settlement history disclosed in review briefSettlement terms that contradict active litigation positions require legal coordination
If the institution has filed a DRT case and the borrower has filed a counter-claim, the settlement terms must be consistent with the institution's litigation position — or the litigation must be formally withdrawn as part of the settlement. The Settlement AI flags active litigation cases and routes to the legal team for coordination before settlement terms are finalised.
→ Escalation: legal team · Settlement terms must be consistent with or resolve active litigationSettlement terms based on a stale security valuation may over-waive if the property has appreciated
The enforcement NPV calculation depends critically on the forced-sale value of the security. A security valuation more than 6 months old may not reflect current market conditions — particularly for real estate, where values in Bengaluru have appreciated 15–18% over the last 18 months. The Settlement AI flags stale valuations and requires a fresh valuation before finalising settlement terms.
→ Pause: fresh valuation required · Settlement terms finalised after revaluationSettlement requires all borrowers' consent — a dissenting co-borrower creates legal complexity
A settlement agreement signed by the primary borrower but not the co-borrower may be legally incomplete — particularly for home loans where both spouses are co-borrowers. The Settlement AI flags cases with a co-borrower where only one party has expressed settlement intent, and escalates to the legal team to ensure the settlement agreement is structured to bind all borrowers.
→ Escalation: legal team · All borrowers must consent before settlement is finalisedThe AI does not escalate because it lacks confidence — it escalates because the decision requires accountability
The Settlement Agent AI can compute the optimal settlement amount for Bharat Agro's account with high confidence. What it cannot do is own the decision's consequences if the computation was based on incorrect field data, an unusual property situation, or a regulatory nuance that was not in the policy at the time the system was trained. The escalation logic is not a limitation of the AI — it is the institutional architecture that ensures every settlement has a human who has reviewed the key facts and is accountable for the outcome. The Settlement AI's auto-approval authority is calibrated exactly to the cases where the decision is genuinely routine. For everything else, it prepares the analysis — and hands the decision to someone who can own it.