Stripe Radar's adaptive learning fraud tool reduces chargebacks 30% on average across its global payment network
Stripe's Radar fraud detection tool — powered by adaptive machine learning continuously trained on $1.9T+ in annual payment volume from millions of businesses — reduces chargebacks by an average of 30% and overall fraud by an average of 32%, with real-time risk scores assigned to every payment.
Background
Chargebacks impose direct financial losses and processing fee penalties on merchants. Stripe, as infrastructure for millions of businesses across diverse categories and geographies, needed a network-level fraud solution protecting merchants without requiring each to build its own fraud stack.
What Was Implemented
- Stripe Radar: adaptive ML fraud scoring trained on all payment data across the Stripe network
- Real-time risk scores assigned to every payment; high-risk payments blocked automatically
- Continuously retrained on new payment and fraud pattern data
- Radar for Fraud Teams: optional merchant-configurable rule layer on top of Stripe's baseline ML
Results
- 30% average chargeback reduction for Radar users (confirmed by Stripe) - 32% average fraud reduction (confirmed by Stripe) - Applied across $1.9T+ in annual payment volume from millions of merchants globally
Lessons
- Network-scale training (millions of merchants, $1.9T+ in volume) provides fraud detection advantages that no individual merchant can replicate with isolated data
- Adaptive (continuous) retraining keeps detection current as fraud vectors evolve
- Merchant-layer customization captures business-specific fraud patterns that network-level models may miss