Leading UAE Bank · BFSI — Predictive Analytics

Reduction of False Positives in SAS AML Data

Rule-based AML alerts pruned with ML — fewer Maker reviews, sharper risk signal.

The problem

One of the leading banks in the UAE runs SAS as a post-facto anti-money-laundering platform. Alerts are rule-based — transactions get flagged as suspicious against various rules, scenarios, and parameters. Every flagged alert is checked manually by a Maker, creating a heavy operational bottleneck and diluting the signal with false positives.

The system we built

TerraEdge layered an ML intelligence tier over the incumbent SAS rules engine. The model learns historical Maker decisions, scores alert likelihood in real time, and routes low-risk alerts to a fast-review queue while escalating high-risk patterns. The Maker pipeline keeps working — it just spends its hours where they matter.

What this means commercially

In AML, operating cost lives in the Maker review queue — every false positive consumes the same minutes as a genuine alert. A 50–70% false-positive reduction (typical on rule-based SAS pipelines after TerraEdge's ML layer is wired in) compounds into significant FTE displacement without weakening the regulator-facing audit trail. The SAS-compatible architecture means no rip-and-replace — the bank's existing SAS investment is preserved, and the value is purely additive.