Fingerprint, the Chicago device intelligence company, launched Fingerprint Fraud Score on 10 November 2026, a predictive signal trained on each customer’s own fraud outcomes, and said that in a head-to-head test it caught nearly 50% more fraud than one customer’s in-house model with no increase in good users blocked. Fintechbits’ archive shows Fingerprint’s own October 2026 survey of 300 US financial services leaders found 95% said they could tell genuine AI assistants from spoofed traffic but only 45% detected and mitigated spoofed AI traffic in real time, the gap Fingerprint Fraud Score is now pitched at. The 50% figure comes from a single unnamed customer.

How Fingerprint Fraud Score works

According to the Fingerprint release, Fraud Score combines Fingerprint’s device signals with fraud outcomes supplied by each customer, then trains separate models for different threats: payment fraud, account takeover, multi-accounting, fake account creation, promotion abuse, platform abuse and content abuse. Fingerprint retrains the models as new outcomes arrive.

Customers can act on the scores directly or feed them into existing decision systems through Fingerprint’s API. The pitch is custom machine learning without an internal data science team to build and maintain it.

Catherine Woneis, Fingerprint’s vice president of product, said: “Fraud teams have a signal-to-decision problem.” David Fennessey, a principal software engineer at the grocery delivery company Flink, said blocking a visitor ID alone was not enough to stop its savvier fraudsters.

What the 50% claim does and does not show

The headline figure is honest about its limits if read closely. It is one evaluation against one customer’s existing model, with no detail on the size of the sample, the fraud type or how good the in-house model was to begin with. A weak baseline makes a large improvement easy.

The more useful claim is the design. Fraud is not one problem, and a single generic risk score blends very different attacks. Training separate models per threat on a client’s own labels is how large banks already work, and offering it as a service to firms without data scientists is a reasonable product.

For a fintech lender or a payments company, the value is in the separation. An account takeover attempt and a promotion abuse ring leave different traces, and a model tuned for one will miss the other. Fingerprint’s pitch is that its device data, which identifies a browser or phone even when cookies are cleared, is the common input that makes each specialised model sharper.

Fintechbits analysis

Why Fingerprint Fraud Score Is a Step Up the Value Chain

Fingerprint Fraud Score moves Fingerprint from selling signals to selling decisions, and that changes its competition. Device intelligence vendors usually feed data into someone else’s fraud platform. A scoring product puts Fingerprint alongside Sardine, Sift, Featurespace and the fraud teams inside processors, which all claim per-client modelling.

Fingerprint’s advantage is the signal. Its own survey found most financial firms struggle to spot bots posing as AI assistants, and device-level data is one of the few reliable ways to tell a real phone or browser from an automated one. Fintechbits has covered the wider shift toward AI fraud models in banking.

The catch is that per-customer models need per-customer labels. Firms with poor fraud tagging, which is most of them, will get less from Fingerprint Fraud Score than the test customer did. Buyers should ask for a trial on their own data rather than rely on the headline.

Frequently asked questions

What is Fingerprint Fraud Score?

Fingerprint Fraud Score is a fraud prediction product launched on 10 November 2026 that trains separate models for each customer and threat type, using Fingerprint device signals and the customer’s own fraud outcomes.

How much more fraud does Fraud Score catch?

Fingerprint says Fraud Score caught nearly 50% more fraud than one customer’s in-house model in a head-to-head evaluation, with no increase in good users blocked. The release does not give sample sizes.

What did Fingerprint find about AI traffic?

Fingerprint’s October 2026 survey of 300 US financial services leaders found 79% had seen a malicious bot slip past defences by posing as an AI assistant or crawler.

What to Watch Next

The next proof point would be named financial institution customers and results across several of them, not one. If Fingerprint publishes those, along with pricing, buyers can compare Fingerprint Fraud Score with the scoring layers they already pay for. Account takeover is where many of those signals start: phishing drove 55% of signals at US banks in Q2’s account takeover report.