By the time a borrower misses their first EMI, the stress that caused it has been building for 60 to 90 days. The signals were there — in their banking behaviour, their bureau activity, their GST filings, their mobile payment patterns. Most lenders never see them because they only look when the bounce happens. The Portfolio Monitor AI looks continuously, and it looks everywhere.
The 90-Day Window That Changes Everything
In Indian retail and SME lending, the average borrower who eventually goes 90+ DPD shows a predictable stress arc. It begins with subtle behavioural shifts 3 months before the first bounce: reduced savings account balances, increased revolving credit utilisation, late GST filing, changed UPI payment patterns. It escalates through 60 days before default: a returned cheque on a different loan, a credit inquiry suggesting refinancing attempts, reduced business receipts. And it crystallises in the 30-day window: the EMI bounce, the call to the collection team, the account beginning its journey toward NPA.
A lender that detects stress at the EMI bounce has one option: collections. A lender that detects stress at 90 days before the bounce has three: proactive restructuring, relationship-based resolution, or controlled exit. The difference in recovery rate between these intervention points is not marginal — it is the difference between a restructured account that performs and a written-off NPA.
The Signal Architecture: What Gets Watched and When
The Portfolio Monitor AI monitors over 200 individual signals across four data domains for every active borrower. Signals are weighted by their predictive lead time — how far ahead of default they typically appear — and aggregated into a borrower stress score that updates daily.
Early Behavioural Shifts — Macro & Financial Signals
Escalating Stress — Cross-Institution & Behavioural
Imminent Default — Direct Payment Signals
A Borrower's Stress Journey: What the AI Sees That Humans Miss
The timeline below reconstructs the stress arc of a typical self-employed LAP borrower, showing side-by-side what the Portfolio Monitor AI detected and when — versus when a traditional monitoring process would have flagged the same borrower. The contrast illustrates not just the speed difference but the quality-of-intervention difference.
GST Filing 8 Days Late + CIBIL Drop of 22 Points
Two T−90 signals converge. Borrower stress score rises from 18 to 41. Flagged as Watch. RM notified to make a relationship call within 5 days.
Bureau Inquiry at 2 Other Lenders + Revolving Utilisation 74%
Refinancing attempt confirmed. T−60 signal cluster. Stress score rises to 67. Escalated to CPO as Amber. Restructuring assessment recommended.
GST Collections Down 31% QoQ + Trade Payable Overdue
Business distress confirmed through financial signals. Stress score reaches 82. CRO briefed. Proactive restructuring initiated with borrower consent.
Account Balance Below 1x EMI on Due Date
Imminent default confirmed. Stress score 94. With restructuring already in progress from Day −61 intervention, this is managed — not a crisis.
EMI Bounce — Account Enters Collections
Without AI EWS, this is Day 1 of the problem. With AI EWS, restructuring was agreed 33 days ago. The account is being managed, not chased.
Account Crosses 90 DPD — Classified as NPA
Without early intervention, this account joins the NPA book. With the AI-triggered intervention at Day −94, the restructured account is current.
From Signal to Action: The Intervention Workflow
Detecting a stressed borrower early is only valuable if the organisation has a workflow to act on that detection. The Portfolio Monitor AI does not generate alerts and stop — it generates alerts with prescribed actions, routes them to the right people, and tracks whether the action was taken and what the outcome was.
200+ Signal Monitoring Across 4 Data Domains
Bureau, banking behaviour, GST/tax data, and internal payment history monitored daily for every active borrower. Each signal scored, weighted by lead-time predictive power, and aggregated into a borrower stress score from 0 to 100.
Score Threshold Routing & Severity Classification
Score 0–40: Monitoring only. Score 41–65: Watch — RM relationship call triggered. Score 66–80: Amber — CPO review, restructuring assessment. Score 81–100: Red — CRO escalation, active intervention protocol.
Borrower Stress Brief Generated Automatically
For every escalated account, a structured brief is generated: which signals triggered the alert, borrower financial summary, collateral position, exposure at risk, recommended intervention type (restructuring, partial prepayment, enhanced monitoring), and comparable past cases and their outcomes.
Intervention Tracking & Model Feedback
Every intervention is logged with outcome: did the borrower cure, restructure, or default despite intervention? This outcome data feeds back to improve the signal weights and stress score model — making the EWS progressively more accurate over time with every resolved case.
The Portfolio-Level View: Early Warning as a Risk Management System
Beyond individual borrower alerts, the Portfolio Monitor AI aggregates EWS data into a portfolio-level stress dashboard. At any point in time, the CRO and CPO can see the full distribution of borrower stress scores across the book — how many accounts are in Watch, Amber, and Red bands, what the total exposure in each band is, and how the distribution has moved over the past 30, 60, and 90 days.
A rising proportion of accounts in the Watch band — even with few accounts yet in Amber or Red — is an early indicator of a deteriorating credit environment that has not yet shown up in NPA numbers. This portfolio-level signal gives the institution 3 to 6 months of advance warning before the NPA cycle peaks, allowing provisions to be front-loaded, origination standards to be tightened, and liquidity buffers to be strengthened before the stress is visible to the market.
The Only NPA That Can't Be Prevented Is the One You Didn't See Coming
The Portfolio Monitor AI's early warning system does not eliminate defaults — some borrowers will default regardless of how early you intervene. But it eliminates the preventable NPA: the account that defaulted because nobody saw the stress building, nobody made the relationship call, nobody offered the restructuring option before the borrower had no other choice. That category of NPA — the one produced by institutional blindness rather than irreversible borrower distress — the AI makes structurally impossible.