Credit-Risk Signals from Web Data for a Fintech SME Lender

the Client

A fintech lender providing working-capital loans to small and mid-sized businesses — restaurants, retailers, D2C brands, service firms — across India and one GCC market. Their underwriting used bank statements and bureau data competently; their problem was everything those sources can't see: whether the business is actually alive and trending, this week. Bureau data lags by months. Bank statements arrive at application, then go dark. Between disbursal and default sits a visibility gap the entire SME-lending industry knows and few instrument.

This is the quiet half of the BFSI web-data story: alongside the trading desks, lenders and insurers now feed credit models with scraped business-activity signals — the segment-anchoring demand we've documented across our finance work.

The Challenge

The brief had two halves:

• Underwriting enrichment. At application time, verify and enrich: does this business exist as claimed, how long has its digital footprint existed, what do customers say, is it operating at the claimed scale?
• Portfolio early warning. Post-disbursal, monitor thousands of borrowers continuously for deterioration signals — cheaply enough per merchant that monitoring the whole book made economic sense, not just the largest exposures.

The constraints were sharp: signals had to come from public business data only (no personal credit data, no individual profiling — both a compliance line and the client's policy), work for businesses with thin formal footprints, and ship with the lineage documentation their risk committee and regulator-facing audits would demand.

The Actowiz Solution: The Business-Vitality Panel

We designed a per-merchant panel of public web signals, collected on a weekly cadence (daily for flagged accounts), organized in four layers:

1. Existence & footprint

Business listings presence and consistency (name/address/phone across directories and maps), website liveness and change activity, domain age, social-profile activity cadence. At underwriting this is verification; in monitoring, footprint decay — a dead website, delisted maps entry, silent social accounts — is among the strongest simple distress markers.

2. Customer-demand proxies

Review velocity and rating trends across maps and category platforms (food aggregators for restaurants, marketplaces for sellers); for restaurant borrowers, menu availability and platform "temporarily closed" status on delivery apps; for online sellers, listing counts, stock-out rates, and marketplace seller ratings. These are the same demand-nowcasting techniques from our alternative-data practice, pointed at small borrowers instead of tickers.

3. Operational signals

Hiring activity (postings appearing/disappearing — our job-postings stack at merchant scale), announced-hours changes, branch/outlet count changes on maps, delivery-radius changes on aggregators.

4. Adverse-event stream

Public regulatory and legal notice boards, tender blacklists where applicable, and news mentions — entity-matched with the flag-don't-guess discipline of our quant news work, since small-business name collisions are rampant.

Each layer rolls up to a vitality score with explainable components — a hard requirement, because "the model said so" doesn't survive a credit committee. Score deltas, not levels, drive alerts: a 3.8 that was 3.9 last month is a business; a 3.8 that was 4.6 is a case.

The Outcome

The vitality panel entered production in two places. At underwriting, footprint verification and demand proxies became standard enrichment — catching a small but expensive class of applications (claimed scale unsupported by any public footprint) pre-disbursal. In portfolio monitoring, the early-warning layer changed the collections posture from reactive to preemptive: relationship managers began outreach on score-delta alerts weeks before missed payments, converting a share of would-be delinquencies into restructures.

The client's risk team reported the culturally hardest and most valuable shift was whole-book monitoring — because the per-merchant economics of public-data signals made it viable to watch every borrower, the long tail stopped being invisible. The engagement has since expanded to pre-qualification scanning of prospect lists and a sector-level dashboard (restaurant-sector vitality by city) their credit strategy team uses for exposure planning.

Why This Pattern Repeats in Lending

SME lending's core information problem — formal data lags, businesses fail faster than bureaus report — is structural, and public web signals are the only continuously refreshing source that covers small borrowers. The design principles that made it work transfer directly: public business data only, explainable composite scores, deltas over levels, entity-matching discipline, and per-merchant economics cheap enough for whole-book coverage.

Frequently Asked Questions

Is this personal credit scoring?

No — the panel monitors public business signals (listings, reviews, operations, adverse notices) about commercial entities. No personal credit data, no individual profiling; PII encountered incidentally is masked at the edge.

How early do web signals lead delinquency?

In this engagement's structure, footprint decay and review-velocity collapse led delinquency by roughly 5–9 weeks in the representative distribution — enough lead time for preemptive outreach to change outcomes.

Does it work for businesses with thin digital footprints?

Coverage varies by segment; the panel reports footprint depth as its own field, so thin-footprint merchants are flagged as low-observability rather than falsely scored. For restaurant and online-seller segments, coverage is strong by construction.

Can this run alongside our existing risk models?

That's the standard integration — vitality scores and deltas delivered as features into the client's models and casework queues, with full lineage for model-governance review. Contact Actowiz Solutions to scope a pilot on a portfolio segment.