Fraud, waste & abuse leaks 8–10% of claim payouts industry-wide — the number to lead with. This view sets out the motor-vs-health fraud types, the SIU case pipeline from flagged to recovered, the detection-method mix (SIU · rules · AI · network · IIB), and readiness for the IRDAI Fraud Monitoring Framework 2025. Company-specific save and catch figures are MODELED targets, peer-benchmarked.
Every point of leakage recovered flows straight to the loss ratio. Industry FWA runs 9% of claim payouts and ~15% of health claims carry a fraud element. The case table below is illustrative; the gold "modeled" figures (savings, catch-rate, SIU build-out) are peer-benchmarked targets, not claimed actuals.
BCG × Medi Assist (2025) put FWA leakage at 8–10% of payouts industry-wide; in health alone that is ₹8,000–10,000 Cr a year. Motor is the largest fraud-exposed general line; health the fastest-rising.
Staged accidents, inflated repairs, fake TP injury (MACT), phantom vehicles, total-loss/salvage, backdated e-policies. The Supreme Court flagged a 'wide racket' of fabricated motor-accident claims (Feb 2026).
Inflated/upcoded hospital bills, impersonation, provider collusion, PED non-disclosure, ghost billing, fake PMJAY/scheme claims. ~15% of health claims carry a fraud element.
An illustrative snapshot of the Special Investigation Unit book. Bars show ₹ claimed by stage with the recovered/prevented portion in green; amounts in ₹ lakh.
| Case | Line | Fraud type | Detection | Claimed | Saved | AI score | Status | |
|---|---|---|---|---|---|---|---|---|
| MOT-FR-2203 | Motor | Fake Third-Party (TP) injury | Network analytics | ₹12.5 L | ₹0.0 L | 0.88 | Investigating | |
| HLT-FR-3303 | Health | Provider collusion (network hospital) | Network analytics | ₹9.4 L | ₹6.0 L | 0.91 | Investigating | |
| MOT-FR-2205 | Motor | Total-loss / salvage fraud | Red-flag rules | ₹6.0 L | ₹4.5 L | 0.76 | Recovered | |
| HLT-FR-3305 | Health | Ghost / phantom billing | AI score | ₹5.2 L | ₹3.9 L | 0.79 | Flagged | |
| MOT-FR-2201 | Motor | Staged accident (organised ring) | SIU | ₹4.8 L | ₹4.8 L | 0.92 | Confirmed |
Sample of 12 cases: ₹54.1 L claimed, ₹34.2 L saved (recovered or prevented pre-payment). Claims scoring above the AI threshold (0.75) auto-route to the SIU. This table is illustrative — portfolio-level savings are the modeled ₹78 Cr below.
Catch-rate only moves if suspicion is raised BEFORE payment. This scores a live claim against the SIU case book above — band and routing computed in code from the fraud-score threshold, not left to the model.
Azure OpenAI scores the claim against the 12 seeded SIU cases; the band and SIU routing are derived in code from the threshold in fraud_metric — the model does not route claims.
Pick a scenario or write your own claim, then run the triage — you get a fraud score, the red-flag indicators it matched, what to collect before deciding, and the closest confirmed cases from the SIU book. Two presets are ordinary, genuine claims; they should score low.
Five detection engines: SIU field investigation, red-flag business rules, AI/ML scoring, link/network analytics, and cross-insurer IIB matches. Ranked by ₹ saved in the sample.
AI scoring and red-flag rules screen the widest net pre-payment; SIU and network analytics carry the high-value organised-ring cases.
Motor is the largest fraud-exposed general line by count; health carries the higher-value collusion and billing cases.
The FMF is effective 1 April 2026: a mandatory Fraud Monitoring Committee & Unit, insurer-specific red-flag indicators, and mandatory participation in the IIB Fraud Technology Framework (Caution Repository of blacklisted providers + a cross-insurer policyholder ID).
The prize: lifting fraud savings toward ₹78 → ₹120 Cr is a direct loss-ratio lever. The capability path is peer-proven — Generali Group has run Shift Technology AI fraud detection since 2016 (typically +20–30% detection uplift); Star Health reports +35% YoY on FWA savings. These are the benchmarks behind the modeled targets.