EITX builds bespoke machine-learning models for insurance fraud. Our probabilistic graph network connects every car, policy, person and incident — and weights every link by the probability fraud travels through it.
Most fraud systems were built for the last fraud. We build for the next one — bespoke, with our insurers, against the rings they actually face.
EITX is a small team of fraud scientists and graph engineers. We work directly with claims and SIU leaders to design models tuned to each book — no generic risk scores, no black boxes you can't reason about.
Every deployment starts with a problem statement, not a feature list: the rings you're losing money to, the patterns you've stopped seeing, the staged incidents your incumbent flags two months late.
Click any node. We surface the people, vehicles, policies and incidents connected to it — and the probability fraud travels along each edge. Hover for context. Filter for paths above 60% confidence to see only the rings worth investigating.
Click any node to drill in · Filter chips top-right · Detail panel slides in on focus
Three books, three sets of patterns. One graph tuned to each.
Staged collisions, phantom passengers, exaggerated low-speed shunts, cash-for-crash rings. We weight links between vehicles, policies, incident locations and bodyshops.
Claim farms, organised whiplash rings, claimant–solicitor link patterns. We model relationships across claim chains over months, not days.
Provider over-billing, phantom procedures, identity-sharing across clinics. Our graph traverses providers, patients and procedure codes simultaneously.
We don't sell a SaaS dashboard and walk away. Each engagement is a tuning exercise against your data, your rings, your underwriters' tolerance for false positives.
Working sessions with your SIU and claims leads. Which rings are slipping through? Where is the noise loudest?
Schema-mapping your exports. Entity resolution across systems. The probabilistic graph is built from your real claim history.
Edge weights tuned to your fraud taxonomy. We benchmark against your incumbent on a held-out ring set.
Investigator UI, alerts, an explainable trace for every flagged ring. Quarterly re-tuning included.
The incumbent told you it was 'configurable'. We built ours for you.
Send us a 90-day export of claims and policies. In two weeks we'll come back with a graph of suspicious rings we'd never seen before — and the maths behind why.