Fingerprint, a US-based device intelligence platform for fraud prevention, has announced the addition of AI-powered recommendations to its Suspect Score solution.
The improvement is set to enable fraud teams to train a machine learning model on their own labelled fraud data, generating optimised signal weights tailored to their specific traffic patterns rather than relying on static, manually configured scoring models.
Suspect Score is built on Fingerprint's suite of Smart Signals, real-time device intelligence insights, and is used by enterprise fraud and security teams to assess transaction risk. The new AI-powered layer allows customers to upload labelled fraud data to train the system. Afterwards, it analyses that data alongside Smart Signals to adjust signal weights, reduce false positives, and maintain detection accuracy as fraud patterns evolve.
Customers can preview all recommended changes before applying them with a single click, retaining full visibility and control over their scoring configuration.
Limitations of static models and adaptive detection
Static fraud scoring models struggle to keep pace with dynamic, traffic-specific fraud patterns that vary by business and evolve continuously. Fraud teams often lack the time and resources to manually analyse signal interactions and retune model weights for their individual use cases. According to Fingerprint, the AI-powered recommendations address this by automating that process against each customer's own data.
The update also responds to the growing sophistication of AI-driven bots and agents capable of bypassing conventional detection, as well as the increasing adoption of privacy tools such as VPNs by legitimate users, which complicates traditional signal weighting.
AI-powered Suspect Score recommendations are available to all Fingerprint customers with access to Smart Signals and can be activated through the Fingerprint dashboard.
Commenting on the launch, Valentin Vasilyev, CTO and co-founder at Fingerprint, said fraud patterns vary by business and evolve constantly, and that AI-powered recommendations remove the manual tuning bottleneck by training on each customer's labelled data.
