Banks Need to Boost Transparency in Machine Learning Models
Machine learning (ML) is becoming increasingly prevalent in anti-money laundering (AML) transaction monitoring, but banks are under pressure to improve transparency and explainability of their models. Model risk management teams and regulators are demanding better methods of interpreting “black box” ML models, which develop and learn directly from data without human supervision or guidance.
Best Practices for Adopting Machine Learning
To successfully adopt machine learning for AML transaction monitoring, leading financial institutions can follow these three best practices:
1. Align Stakeholders on Vision and Design
Engage stakeholders from the beginning of the project to align on vision, make architectural design choices, and consider trade-offs for all processes from end to end. This ensures that business-as-usual activities and ongoing regulatory actions are considered.
2. Develop a Safe Technology Transition Plan
Transformations require an intentional approach, collaborative mindset, and rigorous execution. To minimize risks, banks can run existing rule- and scenario-based risk scenarios in parallel with ML-based scenarios during the pilot phase.
3. Enhance Model Risk Management Framework
Model risk management (MRM) teams need to expand their capabilities to work closely with data science teams, educate data scientists about potential risks, shape validation standards, policies, and frameworks to address specific risks associated with ML models, and define precise performance and monitoring requirements.
Industry Insights
The use of machine learning in AML transaction monitoring has the potential to significantly improve effectiveness by better capturing risk and generating high-quality alerts for downstream investigation. However, banks need to invest significant time and resources to build a talent pool, create reliable data sources, and leverage subject matter expertise.
Experts Weigh In
- “The fight against money laundering is a constant cat-and-mouse game,” said PK Doppalapudi, consultant at McKinsey. “Machine learning with network analytics can improve transaction monitoring dramatically by reducing false-negative and false-positive rates.”
- “To realize the full benefit of machine learning in AML, institutions will need to build a talent pool, create reliable data sources, and leverage subject matter expertise,” said Pankaj Kumar, partner at McKinsey.
Conclusion
As ML becomes increasingly important in AML transaction monitoring, banks must prioritize transparency and explainability of their models. By following best practices, banks can adopt ML safely and effectively, improving their ability to detect and prevent money laundering activities.