Model Validation
The process of testing a model for accuracy, fairness, stability, and fitness for its intended use before and after deployment.
Model validation is the testing and review process that checks whether a model works as intended and remains fit for use. It typically covers accuracy, bias and fairness, robustness, and whether the model performs well on the population and use case where it will be deployed.
The NAIC Model Bulletin expects insurers to validate high-risk AI systems before deployment and to revalidate them on an ongoing basis. Validation should be documented, independent where possible, and proportionate to the risk the model can cause.
Validation is not a one-time check. A model validated at launch can degrade as data or behavior changes. Ongoing monitoring, drift detection, and periodic revalidation are required to keep the validation current. See our AI vendor risk assessment checklist for validation questions to ask vendors.