Most organisations now have an AI policy. Few have decided where their AI should make decisions, who can change the rules it follows, or who can stop it. This playbook turns those questions into seven decisions an executive can make, record and audit.
Professional advisory firms are not required to disclose when AI generated the analysis instead of expert judgment. The accountability structure your engagement letter assumes may not have applied to the analysis shaping your last capital decision.
AI governance and product operating model governance are the same design problem – clear accountability, defined decision boundaries, empirical quality standards, persistent teams. Only 21% of enterprises have mature agentic AI governance while 74% expect to use AI agents at least moderately by 2027. Bolting an AI-specific framework onto the existing operating model funds the appearance of governance while the accountability gap keeps widening.
The December 2026 ADM disclosure obligation tests the data underneath the AI application, not the application layer most security investment has hardened. Closing the gap requires bridging the structural divide between application-layer security and data-layer governance before the obligation takes effect.
A passing AI readiness assessment confirms your organisation can run pilots. It does not confirm it can convert them into returns. Here is the layer the scorecard was never commissioned to test.
AI agent prompt governance is the missing control layer inside most enterprises. Four controls that apply to every other governed document in your organisation — version identifier, named owner, change process, audit trail — apply to prompts too.
Every AI program your organisation has run started governance from scratch. The organisations that retained their AI governance learning now make decisions yours cannot reach in a single cycle.
AI agents commit real-time spend against budget lines governed by annual review cycles. Three control gaps leave the CFO with no visibility, no spending caps, and no accountability over the fastest-growing line of technology expenditure.
AI pilots stall at scale because the AI operating model they need to land in — roles, workflows, decision rights, and incentive design — was never redesigned. The next executive act is not another pilot.
Only 5% of AI programs produce measurable value. The 60% majority deployed the tools and are producing nothing that shows up in the P&L. What separates the tiers is process redesign before deployment — not budget, not model choice.
© 2019 All rights reserved
