The Use of AI/ML in Risk Management

Recent news stories about platforms like ChatGPT and Bard have raised public awareness of artificial intelligence and machine learning (AI/ML). To better understand how and to what extent financial institutions are leveraging AI/ML for risk management, GARP recently surveyed our certified members about the use of AI/ML at their organizations.

Almost 700 FRMs responded to the survey, representing a broad range of risk managers across the globe, financial services firms, and risk disciplines. Roughly two-thirds of respondents work at large firms (over 1,000 employees) and more than 80% have more than 5 years of risk management experience.

On the following pages, we report the key insights from the survey results.

Insight #1 – Large Majority of Firms Use AI/ML in Risk Management to Some Extent

The use of AI/ML in risk management has been growing. Approximately 75% of survey respondents indicated that the usage of AI/ML in their organization over the past 3 years has increased significantly/moderately. Only 11% of respondents reported that AI/ML was not being used at their firm.

Insight #2 – Use of AI/ML in Risk Management Viewed Positively

Increased efficiency (86%) and improved accuracy (80%) were reported as a “very 
significant/significant” reason firms had adopted AI/ML and a full 75% of respondents indicated that AI/ML has had a positive impact on the overall effectiveness of risk management at their firm.

Insight #3 – Supervised Learning is the Most Widely Used Approach

Supervised learning was by far the most widely reported AI/ML approach with 69% of respondents indicating the prevalence of its use at their firm as either extensive or moderate. At 64%, supervised learning was also the most cited method respondents expect their organizations to increase their use of. Almost 40% of respondents also indicated moderate or extensive use of Natural Language Processing, another sign of significant AI/ML use in financial institutions.

Insight #4 – Credit Risk is the Most Common Domain for AI/ML Tools

Credit risk was the most cited domain for AI/ML usage with 57% of respondents reporting “extensive/moderate” use in that area, followed by 50% in market risk. Credit risk (64%) and market risk (51%) were also cited as the two domains respondents expected to see increased usage of AI/ML at their firms, followed by operational risk (40%), compliance (35%), trading (34%), and liquidity risk (31%).

Insight #5 – Vendor Models and In-House Developed Models are Being Used

There was significant diversity in the use of in-house versus vendor-developed AI/ML models with 46% reporting usage to be almost exclusively/mostly in-house developed models, 30% almost exclusively/mostly vendor models, and 24% evenly mixed.

Insight #6 – Top Concerns in Adopting AI/ML

Reflecting the adage of “garbage in, garbage out”, data quality was most often cited (54%) as a “very significant” concern in adopting AI/ML at their firm, followed by cyber risk (42%), explainability (38%), ethics (37%) and transparency (36%).

Insight #7 – Broader Risks of AI/ML

When asked about the risks usage of AI/ML posed to finance in general, poor model explainability was the most often cited response (72%), followed by unintended bias (68%), increased systemic risk (48%), and increased risk of financial loss (23%).

Insight #8 – AI/ML Risk Governance 
and Management is Evolving

In terms of managing the risks of AI/ML, 51% of respondents indicated that their organization did not have a dedicated team responsible for AI/ML governance and 49% reported that there were no clear guidelines for the usage and deployment of AI/ML models in their organization.

Further, for managing the risks of AI/ML within organizations, there was no clear majority in terms of methods indicated by respondents. Model performance monitoring (33%) was the most often cited technique.

Survey Demographics

Almost 700 certified members participated in this survey, representing a broad swath of risk managers across the globe, financial services firms, and risk disciplines, as shown in the respondent breakdowns by region, firm type, and primary risk area in the charts below.