Most boards evaluating AI are relying on vendor-reported accuracy metrics rather than interrogating the underlying evidence base. That distinction matters. Accuracy alone does not capture how a tool performs across diverse populations, workflows, or real-world conditions.

This white paper synthesizes findings from 24 independently verified studies into a practical decision-making framework. It clarifies where AI capabilities in behavioral health are well-supported, where evidence remains limited, and what the current landscape makes strategically achievable. It also examines failure modes that accuracy metrics do not reveal, including operational, clinical, and governance risks that can undermine otherwise promising initiatives.