Artificial intelligence (AI) and machine learning (ML) are evolving at lightning speed with ever increasing applications across many areas of a financial institution’s business activities. Whilst FS firms have actively pursuing maturation of their model risk management (MRM) practises over the last 10 years, the emerging analytical technology developments create new challenges and model risks that require a review of some key elements of the MRM framework. Given the enhanced predictive power of AI and ML algorithms compared to traditional statistical modelling techniques currently in practice, significant business value can be generated through improved quality of decisions informed by AI/ML model outputs. FS institutions that take a pro-active approach in comprehensively re-designing and re-tooling their MRM practices will be well prepared to mitigate AI/ML model risks, developed trust in these technologies and achieve the potential business benefits. However, given the inherent complexities in modelling practices compounded by increased complexity of AI/ML algorithms, adaptations of MRM practices will have to overcome many challenges. A focused approach, building on existing MRM practices holistically with enhancements to address the incremental risks introduced by AI/ML, can fast track implementation of the enhanced framework. Such a MRM framework adaptation programme should contain the following core elements:
* An AI/ML risk appetite policy
* Enhanced model definitions and risk tiering
* Minimum level of AI/ML MRM training enterprise wide
* Implementing tools and techniques to mitigate AI/ML incremental model risks
* Enhancements to the model development life cycle key affected activities

In this presentation, we examine some of the intricacies involved in these adaptation activities and offer recommendations to address issues.

Practical outcomes:
* An appreciation of differences in AI/ML algorithmic techniques compared to traditional statistical modeling techniques
* An awareness of the key incremental model risks inherent in AI/ML model solutions – such as bias, explainability and robustness
* An appreciation of the importance and value of an AI/ML risk appetite statement and the contents of such a statement
* Insight into criteria that can be used for AI/ML model risk tiering
* Awareness of independent model validation and change management practices required for AI/ML
* Appreciation of data quality and open source software risks
* Awareness of the importance of ongoing AI/ML model risk and performance monitoring and insight into required content of such monitoring system outputs