Rules don't scale, and fraud slips through
The main type of fraud StackCommerce deals with involves purchases made using stolen credit cards—and the most time-consuming and impactful type of fraud comes from the loss of digital goods that are distributed instantly. When this fraud occurs, it not only hurts cardholders, but also the merchants. StackCommerce needed to stop these transactions as quickly as possible, and they sought a solution that could prevent them in the first place.
Before Sift, StackCommerce was using a legacy, rules-based solution that didn’t include any machine learning. As the company’s order volume grew, they discovered the shortcomings of rules-based systems: they don’t learn and they don’t scale. The team found themselves reviewing hundreds—or even thousands—of orders per day, and fraud review became unmanageable. As a response to the increasing volume, StackCommerce began mass approving orders, which in turn increased disputes. There were times when their support queue was so backed up, they’d have to spend a day or more getting caught up.