Score transactions, logins and payouts in milliseconds.
What it is
Streaming models flag suspicious activity at authorization time — card-not-present fraud, account takeover, payout abuse — with explanations the risk team can act on.
Real-world examples
·Decline a $4,000 e-commerce charge that doesn't match the cardholder's pattern
·Catch an account takeover from a new device + new geo
·Hold a payout that matches a known mule signature
What to look for
·Sub-100ms scoring latency
·Reason codes for every decision
·Continuous model retraining
1 model in this category
Stripe Radar
Stripe
AIDB85
ML-based fraud detection trained on the global Stripe payments network.