Skip to content
Today’s Financial Crime Compliance (FCC) systems almost universally produce alerts and triggers of such low quality that 95% to 99% of outputs are routinely deemed false positives. In nearly every other industry, such dismal model performance would be grounds for demising a model and starting over. The adoption of model risk management (MRM) principles in the Anti-Money Laundering (AML)/Bank Secrecy Act (BSA) space afford market participants an opportunity to more closely scrutinize the operational effectiveness of AML/BSA FCC solutions. As true “AI” solutions enter the marketplace regulators and financial institutions alike must grapple with how best to assess the systems and technical processes supporting traditional FCC functions.
The market’s response to U.S. federal model validation requirements has to date been a largely “cookie-cutter” approach focusing on IT, data lineage, and systems-centric principles that assess whether data is passing to FCC systems, whether system logic meets technical requirements, and whether model governance exercises have been completed. These aspects of model testing, while admittedly foundational, do little to explain to FCC risk owners whether the model design and resulting outputs adequately mitigate inherent risk, are operationally sustainable, and align with a financial institution’s risk appetite.
In this white paper, Kurt Drozd discusses how the use of technology and machine learning continues to gain hold in the compliance arena, and how internal audit and model risk management teams should intensify their focus on model efficiency and effectiveness. Doing so would shift the conversation from whether a model works properly to the more fundamental question of whether the model design strikes the proper balance between risk coverage and operational efficiency.
Key Topics
- False Positives: how to reduce 95-99% false positive hit rates
- Shifting from Validating to Evaluating Models: making a business case for your FCC systems
- AI-Powered Technology: the role of model validation and the need for systems to be ‘explainable’
Please complete the form below to read the white paper.
Experiencing difficulties completing the form? Please contact lreid@exiger.com.
