A borrower who will miss their EMI in 90 days is already telling the story through their behaviour today. Their GST filings are getting later. Their bank balance is thinner than it was three months ago. They are drawing down more on their overdraft. They have queried a balance transfer on their credit card. No human team can watch all of these signals, for every borrower, every day. The Early Warning AI watches all of them, continuously.
Why Human-Monitored Early Warning Systems Miss What Matters
Most lending institutions have some form of early warning system. It typically consists of a monthly MIS report highlighting accounts that have missed a payment or bounced an ECS. This is not an early warning system — it is a delinquency tracker. The event it reports has already happened. The borrower is already in the collections queue. The opportunity to intervene before the bounce — to contact the borrower while they still have options and while the institution's intervention has the highest probability of success — has passed.
Genuine early warning means detecting stress before it manifests as a missed payment. That requires monitoring signals that precede the first bounce by weeks or months — signals that are individually subtle, often ambiguous, but powerfully predictive when seen together. No monthly report surfaces these patterns because they require daily data, cross-system correlation, and pattern recognition across thousands of signals simultaneously. These are machine capabilities, not human ones.
The Early Warning AI monitors 12 categories of pre-delinquency signals for every active borrower, every day, updating a stress score that moves before the borrower's payment behaviour does.
The 12 Signals
Declining Average Bank Balance
Sustained month-over-month decline in the average bank balance across 3 consecutive months. Not a single low month — a trend. A borrower whose average balance drops from ₹48,000 to ₹31,000 to ₹19,000 is exhibiting structural liquidity erosion, not seasonal variation.
Lead time: 60–90 days before first bounceBalance Below 1x EMI on Due Date
The account balance falls below the EMI amount on the scheduled deduction date for 2 or more months. The EMI still gets paid — via overdraft or by scraping from elsewhere — but the margin for error has disappeared. This is a structural stress signal, not a liquidity event.
Lead time: 30–60 days before first bounceRising Overdraft / Credit Line Utilisation
Overdraft or credit line utilisation rising above 70% and remaining elevated for 60+ days. The borrower is funding day-to-day operations or personal expenses with revolving credit — a pattern that precedes inability to maintain EMI deductions once the revolving capacity is exhausted.
Lead time: 45–75 days before first bounceCheque Returns on Other Instruments
A cheque or NACH mandate returned on a different obligation — a supplier payment, a utility bill, another loan's EMI. This institution's EMI has not yet bounced, but the borrower's payment infrastructure is already failing on adjacent obligations. A leading indicator, not a concurrent one.
Lead time: 20–40 days before first bounceGST Filing Delays
A borrower who consistently filed on time begins filing late — 8, 12, 18 days after the due date. For self-employed borrowers, late GST filing is one of the strongest single predictors of incoming cash flow stress. It signals either business disruption or administrative capacity collapse.
Lead time: 60–90 days before first bounceGST Turnover Decline
GST-declared turnover declining for 2 or more consecutive quarters. For self-employed and MSME borrowers, this is the most direct measure of business income — and its decline directly predicts reduced loan repayment capacity with a 30 to 90-day lag as working capital buffers deplete.
Lead time: 60–120 days before first bounceCIBIL Score Drop of 20+ Points
A decline of 20 or more points in 60 days on the CIBIL score, without a new loan origination explaining the drop. This reflects a deterioration in the bureau's view of the borrower — typically driven by increased utilisation, a late payment at another institution, or a new derogatory mark.
Lead time: 30–60 days before first bounceNew Credit Enquiries — Refinancing Attempts
Multiple bureau enquiries in a short window from personal loan or balance transfer lenders suggests the borrower is actively seeking to refinance or obtain additional liquidity. This is often the final attempt to stay solvent before payment failures begin.
Lead time: 15–45 days before first bounceDeclining UPI Business Inflows
For MSME and self-employed borrowers with UPI-linked accounts, declining business receipt volumes are a direct proxy for falling revenue. Monthly UPI inflow declining by 20% or more for 2+ months indicates genuine business contraction, not a seasonal pattern.
Lead time: 45–90 days before first bounceIrregular UPI Transaction Timing
Borrowers under financial stress often change their payment patterns — moving from regular, scheduled payments to irregular, last-minute transactions. The clustering of outgoing payments at month-end rather than through the month indicates cash flow management pressure.
Lead time: 30–60 days before first bounceSector-Level Stress Index Rising
When the borrower's sector (construction, textiles, agriculture) shows rising NPA rates at the portfolio level, all borrowers in that sector are assigned a proportional sector stress uplift to their individual EWS score. Individual borrowers in stressed sectors require earlier intervention even if their personal signals are not yet triggered.
Lead time: 60–120 days before first bounce (portfolio-level signal)Contact Pattern Changes with RM / Collections
A borrower who previously responded to RM calls within hours begins taking 2–3 days to respond. Or stops responding to digital communications and only responds to direct calls. Avoidance behaviour is a well-documented precursor to delinquency — borrowers who know a missed payment is coming often begin avoiding the lender in advance.
Lead time: 10–30 days before first bounceHow the 12 Signals Combine into a Stress Score
No single signal is determinative. A declining bank balance in January might be explained by school fees. Late GST filing might be explained by a holiday. The Early Warning AI looks for convergence — multiple signals pointing in the same direction simultaneously. Three or more signals from different categories, all trending negative over the same 30-day window, produce a stress score that is materially more predictive than any single signal.
The stress score runs from 0 to 100. A borrower at 0–30 is in normal range. A borrower at 31–55 is flagged for Watch — passive monitoring intensified. A borrower at 56–74 is Amber — proactive outreach initiated within 5 days. A borrower at 75–89 is Red — relationship manager intervention required within 48 hours. A borrower at 90–100 is Critical — restructuring assessment initiated immediately.
Early
GST Filing 11 Days Late + Bank Balance Declining (Month 2)
Two signals converge: a self-employed LAP borrower files GST 11 days late for the second consecutive month; average bank balance has dropped from ₹52,000 to ₹34,000 over 2 months. EWS score rises to 42 — Watch flag raised. Passive monitoring intensified; RM flagged for awareness at next review.
Amber
Bureau Enquiry + GST Turnover −24% QoQ + OD Utilisation 78%
Three additional signals arrive: a balance transfer enquiry at two other lenders; GST turnover for Q2 shows 24% decline versus Q1; overdraft utilisation has risen to 78% and held there for 6 weeks. EWS score rises to 68 — Amber alert. RM scheduled for relationship call within 5 days. Restructuring option assessment triggered.
Red
NACH Return on Another Lender + Balance Below 1x EMI
Two more signals arrive: a NACH return detected on a personal loan at another institution; account balance is now consistently below the monthly EMI on the due date (ECS has been going through on borrowed OD capacity). EWS score rises to 84 — Red alert. Restructuring offer made proactively. If accepted, this borrower never enters the NPA register. If rejected, collections team pre-positioned for Day 0 intervention.
The Signal You Are Not Watching Is the One That Will Become Your NPA
Every lender monitors payment behaviour. The competitive advantage of Early Warning AI is not that it monitors payment behaviour earlier — it is that it monitors the 12 categories of behaviour that predict payment behaviour before payment behaviour has changed. The borrower who hits your NPA register today did not become risky today. They became risky 60 to 90 days ago — and their signals were there to see. The Early Warning AI sees them.