An AI Decision Rights Framework for Executives

Most delays in AI programs are decision delays: a project waits weeks because nobody is sure who may approve the autonomy change, the vendor, or the deployment, and the question travels upward until it reaches someone willing to own it. This guide gives executives a decision rights framework that assigns each recurring AI decision to a level, names the evidence it needs, and lets decisions be made once and fast.

Why do AI decisions stall?

Because they are new, cross-functional, and consequential, and because the organization has no precedent for them. Who decides whether an agent may issue refunds alone? The head of support, the CFO, the CISO, legal, or the CEO? Without a framework, all of them, sequentially, and the answer is usually to wait. FISTA's executive guide to AI agent governance describes the governance structure; this piece assigns the decisions within it.

What does the framework look like?

Decision	Board	CEO	Accountable executive	Functional executive	Business and technical owners	Engineering
AI thesis and measures	Oversees	Decides	Proposes	Consulted
Risk appetite and policy lines	Oversees	Decides with risk, security, legal	Proposes	Consulted
Operating model and funding structure	Informed	Decides	Proposes	Consulted
Use-case selection and committed outcomes	Informed	Approves top outcomes	Recommends	Decides within function	Proposes with baseline
Funding tranches	Informed		Decides at gates	Consulted	Presents evidence
Autonomy changes	Informed		Consulted	Decides within appetite	Proposes with evidence	Implements
Model, architecture, tools					Consulted	Decides within gateway and evaluation
Vendor selection	Informed if material		Approves	Consulted	Evaluates on cases	Recommends
Deployment approval				Approves tier 3	Approves tier 1 and 2 on pass rate	Releases
Incident response	Escalated if material	Informed	Directs tier 3	Directs tier 2	Runs tier 1	Executes
Retirement	Informed		Approves	Decides	Recommends with evidence	Executes

The exact placement varies by company; the principle is that every row has exactly one "decides."

What principles shape the assignment?

1. Decide at the lowest level with the accountability and the evidence. Owners run agents; executives change authority; the CEO sets direction.
2. Escalate by consequence tier, not by anxiety. The how much autonomy should AI agents have guide defines the tiers.
3. Name the evidence for each decision in advance, so the decision is about numbers.
4. Separate what from how. The business decides what agents do and what counts as acceptable; engineering decides how.
5. Risk, security, and legal hold lines; they do not approve every agent. Their policy is enforced by tier, and their review is reserved for the high tier.

What does the board decide?

Nothing operational. The board oversees the thesis, the appetite, and the governance structure, reviews evidence that the structure works, and receives escalation of material incidents. The board director's guide to AI and agentic AI describes what oversight looks like in practice.

What does the CEO decide?

The thesis, the risk appetite with risk, security, and legal, the operating model and funding structure, and approval of the top committed outcomes. The CEO's guide to AI and agentic AI sets out these four decisions and why they cannot be delegated.

What do functional executives and owners decide?

Functional executives decide which outcomes their function commits to, approve autonomy changes within appetite on evidence presented by owners, and approve high-tier deployments. Business owners propose outcomes with baselines, run agents, handle exceptions, and propose autonomy changes; technical owners run the systems and release on pass rates. The AI operating rhythm for leadership teams guide places these decisions in the monthly and quarterly reviews.

How is the framework recorded and used?

In one table, approved by the executive team, referenced in every review, and revised annually. When a decision arrives, the first question is which row it belongs to; the second is whether the evidence that row requires is present. Decisions that have no row are added at the next quarterly review. The AI program RACI template provides a starting artifact.

What should executives ask?

• For each recurring AI decision, can we name the one level that decides?
• What evidence does each decision require, and is it defined in advance?
• Which decisions were delayed last quarter because nobody knew who could make them?
• Are autonomy changes decided by executives within appetite, or by project teams?
• When was the framework last reviewed?

How can FISTA Solutions help?

FISTA Solutions helps executive teams build the decision rights table, the tiering, and the evidence definitions through its AI enablement practice, and delivers AI agents whose permissions, gates, and metrics implement and inform those decisions. Since 2017, FISTA has delivered 150+ projects for 50+ companies across 12+ countries.

To remove the decision delays from your AI program, talk to FISTA on WhatsApp, or read how to run an AI steering committee for the group that administers the framework.