How do I test a fraud or AML detector without real transactions?
Describe a bank and the share of customers who launder money, for example 0.6%, and Misata Studio simulates ordinary customer behaviour by segment, injects typologies such as structuring and mule accounts as labelled ground truth, evaluates alert rules against the data, and produces cases. You can then measure a detector's precision and recall against the truth.
Best for: Fraud, AML and compliance teams, and vendors validating detection models. Updated 2026-09-21.
A regional bank with 20,000 customers where 0.6% of them launder money through structuring and mule accounts over 12 months.
- Labelled typologies
- Rule-based alerts and cases
- An exact suspicious rate
- Legitimate behaviour that varies by segment
Why it holds up
Ground truth
Every laundering customer and pattern is known, so results are measured, not guessed.
Realistic background
Legitimate activity follows the customer's segment, so a detector is tested against realistic noise.
No real accounts
Nothing comes from real customers, so the data can be shared freely.
Questions
- Which typologies are included?
- Structuring, mule accounts, rapid in-and-out movement, high-risk geography, dormant-account spikes and more, each labelled.
- Can I set the suspicious rate?
- Yes. The rate is applied exactly and shown in the certificate.
- Does it produce alerts and cases?
- Yes. Alert rules are evaluated on the generated data and cases are opened from them.