Account takeover, payment fraud, fake sign-ups, promotion abuse: No two types of fraud look alike. And none of them look the same at your business as they do at another company.
Yet most fraud scores judge every attack with one generic model trained on everyone’s traffic. The fraud specific to your business slips through, or good customers get turned away.
That’s why today, Fingerprint is launching Fraud Score. It trains a separate model for each type of fraud you fight, using the fraud your team has already confirmed, so you catch more of the fraud that’s specific to your business.
How Fraud Score works
Fraud Score starts with what you already know: The fraud your team has confirmed. You share a file of labeled fraud outcomes, such as confirmed account takeover or promo abuse, along with events you know are legitimate.
We combines your labeled outcomes with a broader set of Fingerprint device intelligence signals and train a separate model for each fraud type.
That could be account takeover, payment fraud, fake account creation, promo abuse or any other fraud you track: You choose the label, and the score comes back under that name.
Fingerprint reviews every model, and nothing goes live until you approve the model in your Fingerprint dashboard.
From then on, every visit gets a Fraud Score from 0 to 1 for each fraud type, in the same Fingerprint API response as the visitor ID and Smart Signals. The closer to 1, the more confident the model is that a visit is fraud, and you set your own threshold for each fraud type.
You can act on the prediction directly, or your own business logic, or pass Fraud Score to your fraud orchestration platform as one more input. As you send new outcomes, Fraud Score retrains, so your models keep up as fraudsters change tactics.
More fraud caught, with less upkeep
The payoff starts with accuracy. Because each model learns from your own outcomes, one fraud type at a time, Fraud Score picks up patterns that a generic score can miss.
You don’t need to rip out your existing tools. Fraud Score sits alongside the systems you already use, adding a specialized prediction for each fraud use case, while Fingerprint handles the training, review and upkeep of every model. That frees your team to spend its time on decisions rather than on maintaining models.
You also don’t have to take the results on faith. Fingerprint runs a proof of value on your own fraud outcomes, for the fraud types that matter most to you, so you can see how Fraud Score performs before you commit.
Fraud Score: Availability details
Fraud Score is available today on our Enterprise plan for customers using Smart Signals. Contact our team to get started and talk through which fraud types to model first.
If you’re not yet a Fingerprint customer, you can start for free today or talk to our team about how Fraud Score fits into the fraud stack you’re building.
Fraud isn’t one problem, so it shouldn’t get one score. With Fraud Score, every type of fraud you fight gets its own model, trained on the fraud your team has already confirmed.
