Start a project

Case 04 · Fintech · AI

Every transaction scoredin real time — and explained.

A risk engine for a payments company that combines machine-learning models with rules the risk team controls, explains every decision and gets better with every review.

Client name withheld under NDA
Industry
Fintech · Payments
Region
Europe
Engagement
Product Team · Discovery to support
Services
AI & ML · Fintech & Digital Assets

The challenge

A payments company relied on static rules to stop fraud. The rules caught yesterday’s patterns, blocked too many good customers, and every change took weeks to test safely. Fraud losses and false declines were both rising, and analysts couldn’t easily see why a transaction had been stopped.

What we built

  1. A real-time scoring service that rates every transaction inside the payment flow, in milliseconds.
  2. Machine-learning models trained on transaction, device and behavior signals, combined with rules the risk team manages themselves.
  3. An explanation for every decision: the main factors behind each score, visible to analysts.
  4. A case management tool for reviews and chargebacks, with analysts’ decisions flowing back into model training.
  5. Monitoring for model drift and performance, with new models running in shadow mode before they make decisions.

How we did it

We built a feature pipeline from historical transactions and validated models offline first, then in shadow mode against live traffic. Only then did they start making decisions.

  • A rules interface, so the risk team can respond to a new pattern the same day
  • Model governance documented for partner banks and auditors
  • Latency and availability targets agreed with the payments team before launch

The outcome

  • More fraud stopped before settlement, with fewer good customers declined
  • Analysts see why each decision was made
  • New fraud patterns addressed in hours, not weeks
  • A governance trail ready for partner and regulator questions

Stack

  • Python
  • LightGBM
  • Go
  • Kafka
  • ClickHouse
  • Redis
  • PostgreSQL
  • Kubernetes
  • Grafana

[ Next case ]

All case studies

Case 05 · Fintech · Payments

Cross-border payments platform

Multi-currency accounts, FX and payouts across corridors — on one ledger with automated reconciliation.

Read the case

[ Contact ]

Have a complex product in mind?Let’s talk.

✓ reply within one business day · NDA on request

Start a project

or write to {{CONTACT_EMAIL}}