Risk insights
Stripe Radar’s adaptive machine learning system determines the risk score and risk level for a payment and uses them to decide when to block or mark payments for review. The system evaluates hundreds of signals about each payment, using data from Stripe’s network across millions of businesses. The risk insights feature, available with Stripe Radar for Fraud Teams, identifies some of the highest impact signals that power Radar’s machine learning system, to clarify why Radar scored a payment a particular way.
The Top Fraud Factors section notifies you with fraud signals when the payment has values that commonly indicate fraud. If you don’t see any of the top fraud indicators for a payment, then fraud wasn’t detected. Because Radar’s machine learning detects complex patterns across hundreds of signals, it’s still possible for a charge to be correctly identified as fraud, even if none of the signals appear suspicious on an individual level.
Risk insights also includes information about the customer, such as matching the cardholder’s name with the provided email, and the success rate of transactions on the Stripe network associated with the email address. A low authorization rate may indicate suspicious behavior, because previous declines sometimes suggest past attempts at fraudulent transactions.
We also highlight geography-based information, including the billing, shipping, and, IP address locations associated with this payment.
Risk insights dialog
If you want to see a more complete list of Radar’s highest impact signals, click the Show all insights button from the risk insights section. This opens a dialog with several of the most important signals to Radar’s machine learning engine.
Understanding fraud factors
Some of the signals in the risk insights dialog have badges with numbers next to them. These badges show the fraud factor for a signal on this payment. A fraud factor represents the likelihood of fraud for charges with a value similar to this signal when compared to the average transaction on Stripe. A fraud factor of 3.5x means that charges with a similar value for this signal are 3.5 times more likely to be fraudulent than average. In a higher risk payment, we expect to see some fraud factors greater than 1, and in a lower risk payment we expect to see some fraud factors less than 1.
Hover over a fraud factor to see more information about the possible values for it. These factors will change over time as the data in our network changes. This data provides context for the distribution of fraud factors for a signal. This dialog also provides the network distribution of values for a signal, letting you know whether the current payment has a value that’s common or if it’s rare or unique in the Stripe network.
Related payments
You can also view the network of related payments, which includes any other payments made to your business using the same email address, IP address, or card number as the payment you’re currently viewing. This can help identify common fraud patterns, such as card testing (many different cards sharing a single IP address) or trial abuse (many “customers” share the same card).