Ringfence detects organized fraud by its structural fingerprint — the same people, vehicles, and addresses recurring across "unrelated" claims — on an identity graph built from entity-resolved vehicle and person data. Output: ranked, explainable investigative leads for your SIU.
Organized rings coordinate across many claims, policies, and often multiple insurers at once. Reviewed one file at a time, every individual claim can look ordinary — the fraud only exists at the network level. Incumbent tools score in isolation or link on fuzzy name matches. The layer that's missing is identity.
A rear-end collision. Three occupants. Soft-tissue injuries, an attorney letter, a clinic bill. Nothing anomalous. Paid.
The "victim" in this file was the at-fault driver in two others. Same garaging address as four claimants across three files. Same attorney on all of them. All filed inside one 60-day window, in one venue, days after policy inception. That's not a claim. That's a node.
Uber and Liberty Mutual sued a 14-person, 8-provider ring that staged collisions against rideshare drivers in New York — at least eight incidents over 19 months, $312,979 paid out before it was caught. The evidence that broke it open is exactly what Ringfence scores on every claim, automatically.
// It took 19 months and a federal lawsuit to see this ring. The topology was machine-visible by claim three.
Twelve randomized synthetic books, each ~600 claims and ~1,000 people, seeded with everything that trips naive fraud models: families sharing one address, apartment buildings collapsing strangers onto one street, a handful of clinics handling huge legitimate volume, honest repeat claimants — and six rings per book hiding among their members' ordinary claims. Precision held at 100% on every book: not one household or building was ever flagged. Mean ring recall: 99% with the full signal set — 96% on structural signals alone. Every lead ships with the signals that produced it, so an investigator sees why before spending an hour on it.
Ringfence is an analysis layer, not a data broker. Claims stay under your control; we bring the identity-resolution layer that makes the ring topology visible.
Claim ID, date, venue, loss/paid, parties, attorney, clinic, shop. Flat file or API. Under a data-processing agreement, inside your security requirements.
Parties and vehicles are entity-resolved — one node per real person, VIN, and address — so "J. Smith" at three addresses stops being three strangers.
Communities scored on seven structural signals. Your investigators get a ranked, fully-explained lead file — and make every call themselves.
Insurance anti-fraud investigation is an explicitly permitted use of motor-vehicle records under the Driver's Privacy Protection Act. This lane is the clean one.
Output is an investigative lead for a licensed SIU professional — probabilistic, explainable, human-in-the-loop. Ringfence never recommends denying, delaying, or rescinding anything about anyone.
Processing under DPA, scoped to the engagement. No cross-customer pooling of your claims. Deliverables are leads, not datasets.
You bring a claims extract under NDA + DPA. We run the full ring scan and hand your SIU a ranked, explained lead file. In exchange: case-outcome feedback and, if the leads hold up, a reference. A denied staged claim is a number your CFO can name to the dollar — you'll know exactly what this was worth.