Problem
Uber riders and Eats customers sometimes pay with stolen cards or otherwise bypass payment. Failed transactions then become losses for drivers and businesses. Rules and ML scores flag such users in real time, but the old response was a hard action: decline the ride and restrict the payment method or the user. Some flagged users are legitimate, and they had to contact customer service, which the author calls a resource-intensive process.
Example from the article: a ride request arrives from an account whose behavior suggests a stolen card. Instead of declining and banning, Uber asks the user to prove ownership of the card.
Solution in brief
Replace the ban with a risk challenge: a task the user must complete to prove legitimacy. The challenge here is penny drop verification. Uber places two small random authorization holds on the card, and the user must read both exact amounts from the bank statement and enter them in the app. Passing indicates the user is likely the legitimate cardholder. Honest users can recover on their own.
Questions this case answers
- How to act on a high fraud score without permanently blocking users who may be legitimate
- How to verify that a user controls a payment instrument when the standard action is a hard ban that overloads support
- How to stop payment fraud with stolen cards without blocking legitimate users?
- How to let a flagged user prove they own a payment card instead of banning them?
Tools and methods
- Rules engine and ML risk scores — decide when a challenge is needed.
- Penny drop verification — two random authorization holds the user must report back.
- Per-card challenge status — stored in a database so the challenge survives app exits and card switches.
Google or a one-time email code. No password.