The platform
FraudBrain is an intelligence platform in the literal sense: it fuses records into evidence, scores what it sees, and puts a ranked queue in front of the people with the authority to act.
The evidence graph
Applications, quotes, payments, commission ledgers and post-sale carrier evidence are fused into one graph of producers, applicants, instruments and origins. Fraud hides in connections. The graph is where connections live.
Scoring
Learned models score the patterns that repeat. Engineered signals encode what investigators already know. They work together, and they are allowed to disagree.
The whole book is rescored around the clock. New evidence lands in the next walk of the graph, not in the next audit cycle.
Scoring changes are evaluated against a held-out set of confirmed cases before release. If precision on confirmed cases drops, the change does not ship.
The production metrics on this site are aggregate counts, refreshed on a schedule and published only when the refresh fully succeeds. Nobody types a statistic here.
Humans in the loop
FraudBrain's specialists maintain the models and review what the system surfaces for quality. Decisions about producers belong to your organization: your investigators, your procedures, your record.
A score determines no action. The system never clears, declines or terminates anyone. It builds the case file and ranks it; a person makes every call.
Confirmed human verdicts become training signal, so the engine sharpens against threat patterns that will not sit still.
Every organization sees exactly its own book. Boundaries are enforced server-side on every request and proven by deploy-blocking isolation tests.
Beyond insurance
Entities, origins, instruments, relationships. The graph does not care which industry the ledger came from. Life insurance distribution is deployment one, chosen because fraud hides best where the money moves fastest. Everything else is a data feed away.
Architecture reviews are how every deployment starts. Bring your data people.
Request an architecture reviewA human reads every request.