PayPal: Real-Time Fraud Detection at Massive Scale
How PayPal built a system to detect fraudulent transactions in milliseconds, without slowing down millions of legitimate payments.
The challenge
PayPal processed millions of transactions daily, and even a small fraud rate translated into massive absolute losses. Fraud detection needed to happen in real time — before a transaction completed — not after the fact, since reversing a completed fraudulent payment was far harder than blocking it upfront. The system also had to avoid excessive false positives, since blocking legitimate customers' payments would destroy trust and drive users to competitors.
The strategy
PayPal built a layered fraud detection system combining rule-based checks (obvious red flags like mismatched billing addresses) with machine learning models trained on historical fraud patterns, scoring every transaction in milliseconds before approval. The strategy explicitly accepted that no system would be perfect, and instead focused on continuously tuning the tradeoff between fraud caught and false positives…
Want the full story, outcome and quiz?
Read how PayPal executed it, the results, key lessons and test yourself with a quiz. Free on CaseLearn.
Try the full case free →Key lessons (preview)
- Real-time fraud prevention (before completion) is fundamentally more valuable than after-the-fact detection and reversal.
- Layering rule-based checks with machine learning catches both obvious and subtle fraud patterns that either approach alone would miss.
- Routing suspicious-but-uncertain transactions to additional verification, rather than outright blocking, preserves legitimate customers.