A Fraud Prevention Platform That Scores People, Not Transactions

Fraud in after-sales service was spread across many surfaces: duplicate claims, inflated repair costs, invoice manipulation, wrong bank details on payouts, and repeat offenders. Each was caught (or missed) in its own silo, with no unified way to detect or score risk across them.

• Designed a risk engine with behavioural analytics that scores entities (customers, technicians, vendors) rather than individual transactions.
• Built a persistence layer that tracks each entity’s risk history over time, not just a single event.
• Introduced risk decay: historical flags lose weight unless new signals reinforce them, so nobody is penalised forever for one bad week.
• Added dedicated detection for repeat offenders, duplicate claims, invoice mismatches, repair-cost inflation and payout-detail fraud.

Every month, 3–4% of claims are stopped because the engine identifies them as fraudulent, saving roughly ₹35–40 lakh. A further ₹30–40 lakh a month is protected from wrong claims and payouts to wrong bank details. That is ₹65–80 lakh in monthly losses prevented, from a single explainable system.

1. 01

   Signal mapping

   Catalogued fraud vectors: duplicate claims, repair fraud, invoice fraud, payout fraud, repeat offenders.

2. 02

   Risk engine

   Built behavioural analytics and entity-level risk scoring on top of a persistent risk history.

3. 03

   Decay model

   Introduced risk decay so scores reflect recent behaviour, not permanent flags.

4. 04

   Next

   Extending toward AI-driven pattern detection across the full fraud surface.

1. Signal collectors (claims, repairs, invoices, payouts)
2. Behavioural analytics + entity risk scoring
3. Persistence layer (risk history)
4. Risk decay model → automated claim controls
5. Next: AI pattern detection