Airbnb alum shares fraud detection lessons: feature signals, sequence and velocity checks that catch bad actors

jeff_weinstein · x · 2026-09-24

An ex-Airbnb engineer shares a fraud detection lessons thread.

Base features: using a mock transaction, he shows what to ask first — does the IP location (Lisbon) match the billing address (San Francisco)? Name, email, and email provider reputation? A good start but not enough.
Time dimension: examine both sequence and velocity — account created → card added → purchase → card changed → more purchases; a 4-hour-old account with 2 card updates and 3 transactions in an hour warrants scrutiny.
He says this is the opening of a longer series from his Airbnb days.

More from coding & agent