Real-time fraud scoring
A streaming fraud model that scores every payment in milliseconds and explains each decision to the risk team.
- Client
- Payments provider for online merchants
- Industry
- Fintech
- Services
- Data & analyticsCloud & DevOps
- Timeline
- 5 months
Results
fewer false declines
per decision
markets live in one quarter
Static rules were blocking good customers while new fraud patterns slipped through, and the risk team couldn't see why a payment had been declined.
- 01
Built a streaming feature pipeline from payment and device signals
- 02
Trained gradient-boosted models with a clear explanation for every decision
- 03
Shadow-tested against live traffic before switching over
- 04
Gave analysts a review console to label cases and retrain safely
Built with
- Kafka
- XGBoost
- Python
- Redis
- AWS
Next case study
Want results like these?
Tell us what you're building. A senior member of our team will reply within one business day with honest next steps, even if we're not the right fit.