Fingerprint, a device intelligence specialist, has rolled out Fraud Score, a predictive tool that builds fraud models from each customer’s own outcomes and, in one head-to-head test, flagged almost 50% more fraud than the client’s in-house system.

The new product combines Fingerprint’s device intelligence with machine learning to learn the specific fraud patterns affecting each business. This means organisations can sharpen their detection without needing to develop or maintain models internally.

During the comparative evaluation, the improvement in fraud capture came without any rise in the number of legitimate users being blocked.

Fingerprint argues that most fraud tools treat a varied set of threats as a single problem. Account takeover (ATO), payment fraud, promotion abuse and fake account creation each behave differently, and so each demands its own approach to detection.

Because broad risk scores miss these distinctions, engineering and fraud teams often lose significant time to manual investigation, building intricate rules and maintaining models. Security and risk teams are also contending with a growing volume of signals that must be turned into reliable decisions as online fraud grows more advanced.

To address this, Fraud Score merges Fingerprint’s wide range of device signals with outcome data supplied by each client, producing models shaped around that organisation’s particular traffic. Teams can use the predictions on their own or feed them into current decision engines through Fingerprint’s API responses. Whenever customers submit fresh outcomes, Fingerprint retrains the models on an ongoing basis.

The company sets Fraud Score apart from rivals that depend on generalised network rules or require in-house data scientists.

Source link