Syamil Fahmi Syamilabs.
← Home

Cutting fraud detection time by 80% for a digital bank

Cover image for Cutting fraud detection time by 80% for a digital bank

The starting point

Analysts were reviewing flagged transactions across seven separate internal tools, copying case IDs by hand and losing context at every hop. The average review took eleven minutes, and the backlog grew faster than the team could clear it.

Designing one surface

I consolidated the workflow into a single review canvas: the case, the evidence, and the decision in one column-based layout. Keyboard-first navigation let experienced analysts move through a queue without touching the mouse, while a progressive disclosure model kept new hires from drowning in signals.

Testing with the queue

We shadowed six analysts for two weeks and instrumented every interaction. The prototype went through four rounds, each one removing something rather than adding it. The final design shows fourteen fewer data points than the original and resolves cases faster.

Where it landed

Review time dropped from eleven minutes to just over two. The backlog cleared within a quarter, and the team redeployed two analysts to proactive investigation work instead of queue triage.

More work