the problem
Merchants fight fraud one order at a time, with the information from a single store. The fraudster, meanwhile, works across dozens of stores with the same cards, devices and addresses. Rule engines tuned for one shop block good customers and still miss the organised actor, and every chargeback arrives weeks after the lesson could have been learned.
what we built
BonesFraud is a multi-tenant fraud platform for merchants. At its core is an entity graph: orders, cards, emails, devices, phones and addresses become nodes, and the links between them surface the clusters a flat rules list cannot see. A chargeback feedback loop feeds every dispute outcome back into scoring, so the model learns from the merchant’s own losses instead of a generic dataset. A private cross-merchant reputation network lets participating stores benefit from each other’s signals without exposing customer data.
what it changed
Fraud decisions stop being a per-store guessing game. The same actor hitting a second store is recognised by the network, chargebacks improve the model rather than just costing money, and reviewers get a graph they can reason about instead of a score they have to trust.