Cutting fraud losses 32% with a real-time graph
How a tier-one bank moved from rules to a real-time graph plus ML platform — and delivered explainable decisions at 3,500 TPS.
Results at a glance
Where they started
- Fraud losses had grown 19% year-over-year despite three vendor tools.
- Investigation teams worked across five disconnected systems.
- Every declined transaction required a documented, explainable reason.
How we did the work
- 1
Discovery
Two weeks with fraud, tech and risk to map the current pipeline, data, and decisions.
- 2
Graph + ML foundation
Built a customer-merchant-device graph in Neo4j, hydrated in near real time from Kafka.
- 3
Explainable models
Combined graph features with XGBoost, wrapped every prediction with SHAP explanations.
- 4
Investigator workspace
Replaced five tools with one workspace giving one shared view of the network.
- 5
Regulator engagement
Involved regulators early with documented controls, evaluations and monitoring.
What changed
Fraud losses down 32% within six months.
False positives cut in half — 1.1M legitimate transactions rescued per year.
Every decision ships with a regulator-ready explanation.
Tools of the trade
"Dataamps brought both the ML depth and the banking discipline our regulators expected. It's the strongest partnership we've had."
Bring us your hardest problem
30 minutes with our senior specialists. No decks — just a working conversation.
Ready to amplify your data?
Book a 30-minute strategy call. We'll map your data landscape and identify the three highest-leverage moves.