Case Study Insurance

Reducing claims fraud losses by 34% through pattern recognition at full volume

A global insurance underwriter reduced claims fraud losses by 34% in the first year by applying LLM reasoning to 100% of claims events — surfacing organized fraud rings that were invisible when only high-value claims were analyzed.

34%

Reduction in claims fraud losses, Year 1

$89M

Fraud prevented in first 12 months

12% → 100%

Claims volume under active fraud reasoning

847

Organized fraud network connections identified

$28K

Average value of fraudulent micro-claims in largest scheme

1000×

Inference cost reduction enabled full-volume deployment

The Organisation

Global Insurance Underwriter · Insurance

The Challenge

The underwriter's fraud analytics team focused their resources on high-value claims — those above $50,000 — which represented 12% of claim volume but a large share of visible fraud. The remaining 88% of lower-value claims were processed with minimal fraud scrutiny. Organized fraud rings had identified this threshold and were operating sophisticated schemes just below it, using large networks of claimants to aggregate significant losses from individually unremarkable claims.

The Approach

Full claims event coverage with LLM reasoning applied to every claim regardless of value. The model identified network relationships between claimants, solicitors, medical providers, and adjusters — surfacing organized schemes that individual claim analysis could never detect.

"We had been looking at the top of the iceberg for years. The organized fraud was happening below our visibility threshold. Full coverage changed everything."

Global Head of Claims Fraud Intelligence

Key Finding

The largest scheme identified involved 312 connected claimants, 8 medical providers, and 3 legal firms operating a coordinated soft-tissue injury fabrication ring. Individual claims averaged $28,000 — below the manual review threshold. The network had submitted $68M in claims over 26 months. The LLM identified the network topology within 72 hours of full coverage.

Results at a Glance
Reduction in claims fraud losses, Year 1 34%
Fraud prevented in first 12 months $89M
Claims volume under active fraud reasoning 12% → 100%
Organized fraud network connections identified 847
Average value of fraudulent micro-claims in largest scheme $28K
Inference cost reduction enabled full-volume deployment 1000×
Get in Touch

Talk to us about your data.

Tell us about your event stream and we'll show you what full LLM reasoning coverage looks like for your environment.

Or book a call directly →