You make a call on a vendor. Three weeks later, the founder overrules it in a meeting you weren’t invited to. You make another call, more conservative this time, and get told you’re moving too slowly. Nobody says you’re doing it wrong, exactly. They just keep adjusting what you’ve already decided.
That is what a missing decision-rights framework looks like from the inside. A slow drain on your credibility and your programme’s forward motion.
What are AI decision rights?
AI decision rights are the pre-agreed, written list of which calls belong to the person leading the programme and which require the founder or board. They define the boundaries before disagreements surface, not after them. Without that list, every decision sits in a grey area where whoever speaks last wins, and the delegate ends up managing contradiction rather than moving the work forward.
In practical terms, decision rights cover which operational calls the delegate makes independently, which need the founder’s sign-off before action, and where the regular check-in sits for anything close to the boundary.
The written part matters as much as the agreement itself. Verbal mandates feel sufficient when they are given. Both parties mean well. But when the first real disagreement arrives, each side recalls the mandate through the pressure they are under at that moment. The founder remembers a general direction; the delegate remembers permission to act. The gap between those two interpretations is what the programme runs into.
BCG’s 2025 research on AI adoption found that a persistent accountability gap is one of the main reasons organisations see AI usage rise while business impact stays flat. The gap does not always signal governance dysfunction. Often it is simply an arrangement that was never written down.
The delegate who has a written decision-rights agreement has something the delegate with a verbal mandate does not: a common reference point. When a disagreement arises, both parties can return to what was agreed rather than to what each recalls. That reference point is what makes the mandate stable enough to build on.
Why does a verbal mandate almost always break down?
A verbal mandate breaks down because both parties interpret it through the pressure they are under when they need to recall it. The founder who said “you’ve got this” in March means it less literally by June, when a decision touches something they care about. The delegate, who took those words at face value, finds themselves wrong-footed and their authority gradually eroded.
The Korn Ferry research on AI leadership identified a pattern they named the “AI readiness paradox”: organisations place strong operators in AI leadership roles without the accountability structures those roles require. The operator is selected for general delivery credibility, but the specific question of where their authority starts and the founder’s ends goes unresolved.
Spencer Stuart’s research on executive AI sponsorship found that founders who delegate AI leadership while remaining disengaged produce a specific failure mode. The delegate absorbs the risk, escalates cautiously to avoid overstepping, and the pace of the programme drops to whatever the founder is willing to approve on any given day.
Both findings point to the same gap. A mandate without written decision rights assigns accountability without a matching grant of authority. The programme slows not because the delegate lacks capability but because the structure needed to act at pace was never put in place.
Where do the gaps show up in practice?
Gaps in decision authority tend to cluster at three types of decision. Vendor selection is one, where the founder has existing relationships and views on suppliers. Team-facing communications about AI direction are another. Spending calls within the agreed budget are the third, where the scope of what that budget covers may never have been made explicit.
Each of these three areas shares a characteristic. The founder did not necessarily intend to retain authority over them, and the delegate did not necessarily intend to overstep. The conflict emerges from a category of decision that was never discussed.
Harvard Law School’s research on AI risk disclosures found that reputational risk is the leading AI concern for a significant share of large companies, yet many organisations lack an accountable lead with the authority to install safeguards. The pattern appears at a different scale inside owner-managed businesses. The person responsible for AI outcomes often has no formal authority over the decisions that determine those outcomes.
A useful test before acting: ask whether the founder would be surprised by the decision in a week’s time. If yes, it belongs in the conversation before action is taken, not after. If no, make the decision and note it at the next check-in. This handles the majority of ambiguous calls without requiring a formal escalation for every decision.
When should you act alone, and when do you bring the founder back in?
A working rule that holds up under pressure is this. Operational decisions within the agreed mandate are yours to make and own. Decisions that touch the founding team directly, commit the business to a new supplier relationship above a defined spend level, or change how AI is described to clients or investors belong to the founder before they happen, not as a report after.
The threshold amount matters less than the agreement that a threshold exists. Many founders will not overrule a decision made within a clear mandate, even if it was not precisely what they would have chosen. What they respond badly to is surprise. Putting something on the shared agenda before a decision is made, even briefly, lands very differently from reporting a done deal.
McKinsey’s 2025 research on AI programme effectiveness found that organisations producing measurable AI outcomes consistently had named ownership combined with cross-functional authority to act on it.
Giving the delegate enough autonomy to decide in real time, then aligning with the founder regularly to confirm the boundaries hold, is the working structure. A monthly check-in in the early months of the programme keeps the arrangement calibrated. The agenda is a two-way reset, short enough to stay on the founder’s calendar and specific enough to give the delegate a refreshed mandate to carry back into the work.
How do you start the conversation without it reading as a power grab?
Frame the conversation around the founder’s time, not the delegate’s authority. “I want to make sure I’m bringing the right decisions to you, not adding to your week with things I should own” lands very differently from “I need to know what my authority is.” The first version is already aligned with what the founder wanted from the delegation. The second sounds like a territorial negotiation.
The document that comes out of the conversation does not need to be formal. A one-page written summary, shared by email and confirmed as agreed by both parties, is enough. It just needs to exist outside anyone’s memory.
NACD’s AI governance guidance emphasises that accountability without corresponding authority is the condition that generates AI risk, not the technology itself. The delegate who agrees their decision rights in writing has addressed that gap at the programme level, without needing a governance framework built for an enterprise.
On paper, the useful categories are which calls the delegate owns outright, which decisions need the founder before action is taken, and the spend threshold above which anything escalates. Once those are agreed, the question about any specific decision shifts from “whose call is this?” to “does this fall inside or outside what we agreed?” That is a much shorter conversation, and one the programme can hold at pace.
The Paro research on effective delegation found that the delegation trap, in which a delegated lead absorbs risk without authority over the decisions that shape outcomes, is most reliably broken by a written accountability map. The map needs to be agreed, not elaborate.
The conversation is worth having before the programme starts. The later it runs, the more decisions have been adjusted in ways that shift the shape of the mandate, often without either party realising it until something breaks. An hour spent agreeing what is yours to decide, and writing it down, is the most direct investment you can make in a programme that keeps moving.
