What is a fraud score?

‍A fraud score is a number or risk category your payment processor or fraud tool assigns to a transaction to estimate how likely it is to be fraudulent. A model produces it from signals gathered at checkout, before the charge is approved.

It's a composite output that you don't calculate yourself.

How a fraud score is calculated

A fraud score comes from several categories of signal, all read during the one checkout attempt. Stripe says the scoring happens in real time, "often within milliseconds of a transaction."

Stripe names eight categories its model weighs:

• Transaction details: the amount, currency, item type, timing, and frequency.
• Payment information: card metadata, the issuing country, and billing details.
• Customer history: whether this buyer has bought here before, and how that went.
• Account and identity signals: email quality, account age, and how complete the profile is.
• Device data: whether the device has been seen before, and how it behaved.
• Network and location signals: the internet protocol (IP) data that infers location.
• Behavioral patterns: odd checkout behavior or repeated credential attempts.
• Speed indicators: several login or payment tries in a short window.

Merchants tend to assume one bad signal kills an order.

A failed address verification (AVS) check is one input, not a verdict. So is a phone the model has never seen. Plenty of real buyers order from a new device, so that signal can't decide an order alone.

What a fraud score is used for

A fraud score sends each transaction one of three ways, approving it, blocking it, or holding it for review. A threshold on the score decides which one.

Stripe describes all three. Low scores "pass automatically, which keeps checkout fast for most customers." Scores above the threshold get blocked before the charge completes. Mid-range scores go to "human reviewers when the risk is unclear but worth a closer look."

Stripe's Radar for Fraud Teams scores each payment from 0 to 99. By default, 65 or above reads as elevated risk, and 75 or above as high risk.

Those two cutoffs are Radar's own defaults rather than an industry standard.

Stripe says teams adjust their thresholds as they watch fraud rates and false declines. Drop your block threshold and you catch more fraud. You also review more orders and turn away more good customers. Raise it and checkout stays fast while more risk gets through.

Why fraud scoring matters for merchants

Fraud scoring stops a stolen-card charge at checkout, which is the cheapest place to stop it. Catch it there and you ship no goods, answer no dispute, and pay no fee.

The score estimates fraud at the point of sale, and that's the whole job.

Two kinds of dispute fall outside that job. One is friendly fraud, where the cardholder did buy the thing and disputes it later. The other is a real purchase that nobody in the house recognized on the statement.

A low score says the transaction looked legitimate at checkout. It was, in both of those cases.

That leaves the transaction you scored, approved, and then lost to a dispute anyway. An alert network catches those disputes after checkout, when a refund can still settle things before the card network records a chargeback.

We built our alerts for exactly that gap.

Run your own scored-but-disputed count through our ROI calculator to see what that gap is costing you.

FAQ

Is a fraud score the same as a credit score?

No, they measure different things for different parties. A credit score rates a person's odds of repaying a loan, and a fraud score rates one transaction.

Can a merchant see the exact fraud score for a transaction?

Usually yes, though the format varies by tool. Some expose the exact number per payment, and others show only a risk band like low, elevated, or high.

Does a high fraud score mean a transaction is declined?

Only if it clears the block threshold. A score above your review threshold but below your block threshold sends the order to manual review instead.