Every fraud vendor makes the same promise. Faster detection. Smarter models. AI that thinks like an analyst. The words change; the evidence rarely follows. So one of Europe's largest retail banks decided to stop listening to promises and start running a test of its own.

It pulled a set of production fraud cases its senior analysts had already investigated and closed. Then it stripped away every note, every verdict, every hint of the correct outcome — and handed them, blind, to Cleafy's Autonomous Fraud Operations platform, Nyx. The question was simple: can an autonomous system investigate fraud with the same rigour as an experienced fraud team?

Nyx reconstructed each investigation from scratch, building the relationships between customers, accounts, devices, sessions and transactions, testing competing hypotheses and producing a full evidence trail. Each case took three to four minutes. And it didn't just match the bank's senior team — it caught what they had missed, and wrote a new detection rule the bank promoted into production.

Investigation, it turned out, no longer had to be constrained by analyst capacity.

Read the whole story to see how the bank tested Nyx blind, what the evidence revealed, and why its Head of Fraud reopened closed cases on the strength of it:

  • 100% independent agreement with senior analyst verdicts — every case, zero notes or hints.
  • 9% additional confirmed fraud recovered from investigations that had already been closed as legitimate.
  • A new production detection rule, authored by Nyx from the evidence and promoted through the bank's own governance review.