Risk Modeling

CHAPTER 6Machine Learning Model Deployment, Implementation, and Making Decisions

As we have established in previous chapters, a step change in the use of AI, machine learning, and automation in risk and compliance functions, the question is not whether it has transformational potential—that's a given—but rather how to better operationalize AI and machine learning in an ethical, responsible, and sustainable way.

The opportunities that these innovative models present are broad: it is said to give better accuracy in the quantification of risk, deeper insights into big data, and efficiency gains from the automation of repetitive tasks.

AI and machine learning only deliver value when organizations take actions based on the insights. Regardless of how impressive an AI or machine learning looks in the laboratory, it is only when the AI is deployed in the real world that it delivers tangible benefit to impact the bottom line.

With the increased sophistication in algorithms, deployments are an increasingly challenging hurdle organizations are facing. Many of these modern models do not make it into production. According to a Gartner report in 2021, only 53% of POC (proof of concept) models made it beyond the lab to production systems, and this process still took on average 9 months.¹ Similarly, research from SAS indicates that only half of AI models built make it to production. In other words, only half of developed models end up generating business value, and often these models take ...