Banking Circle: federated AML across two continents — without moving a single transaction.
Robert Norvill · Senior Data Scientist · Banking Circle
Flower Summit 2025 · Cambridge, UK · 18 min
+65%
Precision uplift
+25%
Recall uplift
+10%
Accuracy gain
Reach the same outcome
Two paths. Same first-class status. Both end with you running it independently.
Path 1 · Tools
Build it yourself
Open the recipe Robert used — Flower SuperLink, secure aggregation, the FLAME pipeline. Community-supported, on your timeline.
Path 2 · With us
Build it with our FDE team
Scoped Pilot or full FDE engagement to deliver a similar outcome — without doing the cryptographic and AML pipeline engineering yourself.
The build
Banking Circle processes more than 10% of Europe's B2C e-commerce flow. As it expanded into the US, its AML team needed a way to test whether a US transaction-monitoring model could learn from European patterns without moving EU transaction data across borders. Robert Norvill built FLAME, a Flower-based federated learning pipeline, mapping each bank to a SuperNode so models could train across jurisdictions while transaction data stayed local. In validation, the federated setup outperformed the local baseline, with the largest uplift in precision.
Flower was exactly the solution we were looking for.
Robert Norvill · Senior Data Scientist, Banking Circle
Stack
Flower Framework · Secure aggregation · PyTorch · Custom FLAME orchestration
Timeline
From prototype to deployed cross-border model in one quarter.