The challenge

A top-20 global cryptocurrency exchange processing over $2B per month in trading volume came to Serixo in Q3 2025 facing a fraud crisis that was threatening their regulatory standing in the EU. Account takeovers, wash trading, and a sophisticated synthetic identity scheme had driven their fraud rate to 2.3% — well above the 0.5% threshold that triggers enhanced regulatory scrutiny under MICA Article 72.

Their existing fraud stack — a combination of a legacy rules engine and a three-year-old ML model — was generating 14% false positive rates, causing significant customer friction and churn. Every attempt to tighten the rules reduced fraud but increased false positives; loosening them did the reverse.

Deployment

Serixo deployed in shadow mode alongside the existing stack for 30 days, ingesting the same event stream and producing scores that were logged but not actioned. This allowed us to calibrate the model on the exchange's specific transaction distribution and validate score performance without disrupting live operations. At day 31, Serixo took over as the primary decision engine.

Results

By day 60, fraud losses had fallen 91% from baseline. More significantly for the business, the approval rate increased by 12 percentage points — meaning more legitimate trading volume was flowing through the platform. The false positive rate fell from 14% to 4.1%.

Technical details

The primary drivers of improvement were three capabilities that the legacy stack lacked: behavioural graph analysis (which identified the synthetic identity ring within 72 hours of deployment), agent classification (which correctly identified and separately scored the exchange's legitimate algorithmic trading clients), and continuous model retraining (which adapted to the exchange's evolving transaction distribution without manual intervention).

ROI analysis

The exchange's annualised fraud losses pre-deployment were approximately $28M. Post-deployment, the projected run rate is $2.5M — a saving of $25.5M annually. Serixo's annual fee represents less than 8% of the saving. The total ROI from the approval rate improvement (an additional $18M in annual trading fee revenue from previously-declined legitimate volume) brings the total return to over 40x the investment.