Vintage analysis is the single most powerful diagnostic tool in consumer and SME lending — and the most consistently under-resourced. Most lenders do it annually, if at all, with a team of analysts spending two weeks in Excel. The Portfolio Monitor AI runs full vintage analytics monthly, automatically, for every cohort across every product segment, and delivers insights that would take a human team a quarter to produce.
What Vintage Analysis Actually Tells You
A vintage is a cohort of loans originated in the same time period — typically a quarter. Vintage analysis tracks the credit performance of each cohort as it ages: what percentage of loans originated in Q1 FY24 had bounced EMIs by month 3? By month 6? By month 12? How does the default curve of Q1 FY24 compare to Q1 FY23 at the same age? Is the current book performing better or worse than historical norms at every point in the repayment lifecycle?
These questions are not academic. They are the earliest available signal of whether underwriting quality is improving, deteriorating, or stable. A vintage that shows elevated bounce rates at month 6 is a warning that will become visible as NPA somewhere between month 12 and month 24 — giving a lender 6 to 18 months of lead time to tighten underwriting, raise provisions, or adjust pricing before the stress crystalises.
The challenge is computational volume. A mid-sized NBFC with 8 product segments, 15 active quarterly vintages, and 6 borrower cohorts has over 700 active vintage cells to track every month. No human team does this comprehensively. The Portfolio Monitor AI does it automatically.
The Monthly Vintage Report: What Gets Tracked
30+ DPD Curve
First-payment-default rate and 30+ DPD trajectory at each month-on-book milestone, compared to prior vintage benchmarks.
90+ DPD & NPA Entry Rate
Loans crossing the 90-day threshold — the RBI NPA trigger. Tracked by cohort and compared to expected entry rate from credit model assumptions.
Prepayment & Foreclosure Rate
Early prepayments signal high-quality borrowers refinancing elsewhere. Elevated rates reduce expected yield; the AI adjusts ROA projections automatically.
Roll Rate Matrix
Movement between DPD buckets: current→30+, 30+→60+, 60+→90+. Worsening roll rates are the earliest leading indicator of emerging vintage stress.
Cure & Recovery Rate
Stressed accounts that self-cure back to current. Low cure rates indicate structural borrower distress, not temporary liquidity disruption.
Model vs Actual Deviation
Each vintage is compared to what the credit model predicted at origination. Systematic outperformance or underperformance triggers model recalibration flags.
The Vintage Heat Map: Seeing the Whole Portfolio at Once
The most powerful output of the Portfolio Monitor AI's vintage module is the heat map — a single view of every active vintage, every product segment, and every age bucket, colour-coded against performance benchmarks. A risk officer can look at this table for 90 seconds and see exactly which vintages are performing well, which are deteriorating, and which require immediate management attention.
The heat map below shows a typical mid-tier NBFC's LAP portfolio. The pattern visible to a trained eye — and immediately flagged by the AI — is the progressive deterioration in self-employed vintages from Q3 FY24 onwards, coinciding with the rate hike cycle impact on borrower cash flows. A human analyst would take two weeks to produce this view. The AI refreshes it monthly, automatically.
| Vintage · Segment | MoB 3 | MoB 6 | MoB 9 | MoB 12 | MoB 18 | MoB 24 | Trend |
|---|---|---|---|---|---|---|---|
| Q1 FY23 · Salaried | 0.4% | 0.9% | 1.4% | 1.8% | 2.1% | 2.3% | Stable |
| Q2 FY23 · Salaried | 0.3% | 0.8% | 1.3% | 1.7% | 2.0% | 1.9% | Strong |
| Q1 FY23 · Self-Empl. | 0.8% | 1.6% | 2.8% | 3.4% | 3.9% | 4.6% | Watch |
| Q3 FY23 · Self-Empl. | 0.9% | 2.1% | 3.3% | 4.8% | 5.2% | — | Stress |
| Q1 FY24 · Self-Empl. | 1.4% | 3.2% | 5.6% | 6.9% | — | — | Alert |
| Q3 FY24 · Self-Empl. | 1.6% | 4.1% | — | — | — | — | Early Alert |
| Q1 FY24 · Salaried | 0.4% | 1.0% | 1.6% | 2.0% | — | — | On Track |
The Three Vintage Insights That Change Underwriting Decisions
Monthly vintage analytics produces three categories of insight that directly feed back into underwriting and policy decisions — this is the feedback loop that transforms portfolio monitoring from a reporting function into an underwriting intelligence engine.
The first is model deviation detection. When a vintage performs systematically worse than the credit model predicted at origination — for example, a vintage where the model predicted 2.5% NPA but actual NPA is tracking to 5.5% at month 12 — the AI flags the model for recalibration and identifies which borrower characteristics in that vintage drove the overestimation of creditworthiness. This is how underwriting scorecards get smarter over time rather than quietly degrading.
The second is underwriting period identification. The vintage heat map makes it possible to identify specific time windows where underwriting quality deteriorated — the quarters where approval rate was pushed up by business pressure, where a new product was launched without sufficient risk calibration, or where a geographic expansion was executed too aggressively. This institutional learning is invaluable and available nowhere else.
The third is forward NPA forecasting. By applying the observed default curves of mature vintages to the age distribution of the current book, the AI produces a 6 to 12-month NPA forecast with confidence intervals. This gives finance, treasury, and the board a provision planning tool grounded in empirical vintage data rather than management judgement.
Vintage Analysis Is the Audit Trail of Underwriting Quality
Every underwriting decision made today will show up in vintage data 6 to 18 months from now. The Portfolio Monitor AI ensures that when it shows up, the institution sees it immediately — and that the learning from what it sees is fed back to the underwriting team within weeks, not quarters. The vintage curve is the most honest feedback mechanism that lending has — the AI finally makes it fast enough to act on.