Stripe Radar is the fraud-detection system built into Stripe, the payments company. When a card payment is attempted, Radar scores it with a machine-learning model and decides in real time whether to allow, block, or flag it for review. Stripe’s own engineering guide on machine learning for fraud describes a model that draws on hundreds of features, many of them aggregates computed across the entire Stripe network rather than just the individual merchant’s history.
That network scale is the central advantage Stripe claims. Because Stripe processes payments for millions of businesses and handles hundreds of billions of dollars in volume, a fraud pattern seen at one merchant can immediately inform scoring for every other merchant. The guide notes that Stripe uses sophisticated techniques including neural networks and deep learning to exploit the size of this dataset, and frames fraud detection as a probability problem balanced on precision-recall and ROC trade-offs - catching fraud without falsely declining good customers.
Radar is a good illustration of supervised machine learning operating at industrial scale on a live financial system. Most of its training signal comes automatically from the outcomes of past payments - which were charged back, which were disputed, which were legitimate - so the model continuously updates as fraud tactics shift.
Why business readers should care: Radar shows the dominant pattern in modern payments fraud defense. The competitive edge is not a single clever algorithm but the breadth of the network behind it. A platform that sees more transactions can spot emerging fraud earlier, which is why fraud scoring has consolidated toward the largest payment networks and processors.