The lag problem
Bureau reporting in India runs on cycles. A facility opened today may surface to other lenders weeks later; utilisation and repayment behaviour later still. None of that is a bureau defect — it is how periodic reporting works. But it means the window in which a borrower accumulates credit fastest is exactly the window in which each individual lender knows least. Loan-stacking, evergreening through refinance, drawdown sprints ahead of stress — all live inside that lag.
Velocity is a comparison, not a count
Raw counts mislead. Three new facilities in ninety days is alarming for a borrower who historically opens one a year, and routine for a trader who cycles working capital constantly. Velocity only means something against the borrower's own baseline— and honest systems refuse to judge when no baseline exists. Counts get reported; “elevated” is only ever declared against evidence.
The second comparison is capacity. Debt growing 20% in a year is healthy when revenue grew 30%, and a warning when revenue grew 5%. Utilisation racing ahead of operating inflows tells the same story from the cash side. Capacity-adjusted velocity — growth measured against the borrower's demonstrated ability to carry it — separates expansion from accumulation.
Drift: the quiet signals
Between velocity spikes sit slower deteriorations that periodic reviews are designed to catch and usually catch late:
- Repayment drift— mean payment delay creeping from a day to two weeks, measured against the borrower's own payment history, not a portfolio average. A bounce or a restructuring is loud; drift is quiet, and earlier.
- Evidence drift— statutory filings slipping, a fresh charge registered by an unfamiliar lender, sources starting to contradict each other. The borrower's paperwork often deteriorates before the borrower does.
What makes a velocity alert usable
An early-warning framework fails in one of two ways: it misses, or it cries wolf until nobody listens. The second failure is a design problem, and the fix is explainability as a hard requirement. Every alert should carry its metric, window, baseline, threshold, source classes, freshness, confidence and — critically — its limitations. A reviewer who can see “30-day count 3 vs baseline 1, lender-known plus bureau-reported sources, medium confidence, no AA data in scope” can act on it or dismiss it with a recorded reason. A bare red icon teaches people to ignore red icons.
Signals, then a worked case — never a decision
Velocity tells a lender where to look, earlier. It does not say “lend” or “don't” — that remains the lender's call under the lender's policy, with the alert flowing into an assigned, evidenced case whose closure (or override, with a mandatory reason) is on the record. This is the model AssurePulse implements, at design-partner stage today: the live simulation shows a velocity alert forming end-to-end on a synthetic borrower — elevated facility velocity, debt outgrowing revenue, repayment drift, and the fully explained alert underneath.