Giftly · 2024 – 2026
Fraud review as a decision surface
Rebuilt fraud investigation as a decision surface rather than dashboards: throughput up 40%, the review backlog halved.
Public- +40%
- investigation throughput
- −50%
- review backlog
The situation
Fraud review was a manual process fighting a growing queue with general-purpose tools — admin pages built for support work, not for deciding whether an order is fraudulent.
The constraints
- A public write-up of fraud tooling is read by fraudsters too, so this study stays at the decision level: what the tool is for, not what it looks for.
- A false positive punishes a real buyer, so throughput could not come from lowering the bar of evidence a reviewer sees before deciding.
The call
Build a decision surface, not dashboards. A reviewer claims a case, sees what the decision needs in one place, decides, and the decision feeds back into the queue — instead of assembling context from admin pages for every order.
Automate the evidence assembly around chargebacks, so contesting one is a review step rather than a research project.
What happened
Investigation throughput rose 40% and the review backlog halved.
Chargeback evidence became something the system assembles rather than something a reviewer compiles by hand.