An AI governance framework a board could actually use

The situation

Models were being built in pricing, claims and customer service, but nobody could say how many were in production or who owned them. A supervisory review was scheduled for the following year.

What we did

1. Built the first complete inventory of models in use, including those bought from vendors.
2. Introduced three risk tiers, each with its own testing, monitoring and approval path.
3. Designed a one-page quarterly board report on model risk, and trained the risk committee to read it.

What changed

• 47 models inventoried, owned and risk-tiered
• No findings in the supervisory review
• One page quarterly board report on model risk

The lesson

Boards don’t need to understand models. They need to know which ones exist, what they decide, and who owns them.