Payment Fraud Detection Dashboard

Overview

Sherlock is Adyen's machine-learning risk engine. The challenge: make the model's decisions understandable to people who aren't ML experts, so they trust it and act on it.

I led the design of its explainability features, working closely with risk officers and data scientists to learn how risk is assessed, what data matters and the language experts use.

Process

1. 01

   Discovery

   I interviewed risk officers and data scientists to understand the criteria they weigh, the data they rely on and the jargon they use — everything needed to explain the model to non-specialists.

2. 02

   Requirements

   I turned that into a vocabulary and a set of data types for explaining the model's "thinking" — simple, without losing accuracy.

3. 03

   Design & Prototyping

   I designed interactions that walk users from inputs to the risk score, so they understand why a payment was flagged and can trust the model.

4. 04

   Testing

   American Airlines — a large merchant with complex risk needs — tested the concepts in a real setting. Their feedback shaped the final design.

5. 05

   Development

   The result worked for a major airline and scaled down too: smaller merchants can now opt into ML-based risk assessment without a team of risk officers.

Explainability

Rule Editor