Commerzbank

Commerzbank, one of Germany's largest commercial banks, faced a structural challenge common to large financial institutions: its rules-based AML monitoring system generated high volumes of alerts that required manual review by compliance teams. Static rule sets, while auditable and controllable, are inherently backward-looking — they flag transactions matching known patterns but struggle to detect novel money laundering typologies that fall outside predefined thresholds. The result was a resource-intensive review burden skewed toward low-risk alerts, with compliance staff spending significant time on cases unlikely to yield actionable findings. As financial crime typologies evolve and regulatory expectations around effective monitoring systems continue to rise, the cost of maintaining an alert-heavy, low-precision system — in both operational spend and detection gaps — became unsustainable.

Commerzbank partnered with Hawk, a financial crime technology firm, to deploy the Hawk AML AI Extended Risk Model alongside its existing rules-based monitoring infrastructure. Rather than replacing legacy systems, Hawk's integration layer connects directly to Commerzbank's established tooling, allowing the machine learning model to supplement rule frameworks without requiring repeated changes to underlying infrastructure each time the AI model is updated or refined. This additive deployment approach — layering AI on top of existing controls — allowed Commerzbank to introduce machine learning and predictive analytics into production without dismantling established compliance architecture. Hawk's model identifies patterns in transaction behaviour and customer risk that fall outside static rules, enabling the bank to prioritise higher-quality alerts. Critically, the solution was designed with explainability at its core, a non-negotiable requirement for regulatory approval when AI influences case generation and prioritisation.

The deployment produced measurable improvements across both alert quality and detection breadth:

• Higher targeting accuracy — the AI model improved precision in flagging alerts that merit investigation, reducing noise from low-risk transactions.
• Fewer false positives — routine review work on low-risk alerts declined, freeing compliance staff for higher-value investigations.
• Novel typology detection — Commerzbank reports identifying more new money laundering and fraud cases since implementation, including patterns not captured by predefined rules.
• Expanded model governance — software validation processes were extended to cover AI model governance, supporting regulatory approval workflows.

Viktor Kraus, Cluster Lead Global Financial Crime Prevention Platform at Commerzbank, described AI as a "high strategic priority" for the bank's compliance architecture, signalling institutional commitment beyond a single deployment.

• Additive AI deployment — layering machine learning on top of existing rules engines — allows large banks to improve AML detection without dismantling established compliance infrastructure or legacy tooling.
• Explainability is a regulatory prerequisite, not a product differentiator. Where AI influences case generation or alert prioritisation, supervisors expect banks to demonstrate how decisions are made.
• Reducing false positives is as strategically significant as detecting new typologies: both lower compliance operating costs and reduce risk of investigator fatigue on genuine cases.
• Integration layer design matters — connecting AI to legacy systems without requiring repeated infrastructure changes enables faster iteration as models are refined.
• AML AI programmes require executive sponsorship and a clear compliance architecture strategy, not just a point solution.