Using AI to create dynamic, risk-based Radar rules

The introduction of adaptive rules is enabled by two key factors: Stripe’s underlying AI infrastructure, which allows us to continuously improve our models for fraud prevention, and our collaboration with issuers through the Enhanced Issuer Network.
The Enhanced Issuer Network allows issuers to securely access Radar’s AI-powered fraud scores for transactions. This additional data helps issuers make more informed decisions about whether to decline or authorize transactions. As a result, when an issuer authorizes a transaction despite an incorrect CVC or postal code, we know they’ve taken Radar’s scores into account, which signals to us their confidence in the payment’s legitimacy.
Now, Radar’s new rules treat the issuer’s response as an additional input, allowing for some transactions to go through that would have otherwise been blocked. First, our AI models use data across our network of millions of businesses and tens of billions of transactions to develop a risk score that predicts whether a payment is likely to be fraudulent. Based on this score, we block the transaction, trigger a manual review, or send it to the issuer.
The issuer then either authorizes or declines the transaction, and sends us new, additional information with their decision—such as whether the CVC or postal code was incorrect (which is not something we know at the time Radar’s models initially go to work). At this point, Radar combines the issuer’s response and our original risk score. For example, depending on your other Radar rules, we would block higher risk transactions with an incorrect CVC, while allowing lower risk transactions with an incorrect CVC—ultimately helping you increase payment success rates.