Something happened that neither side mentioned in the meeting but both noticed. The founder reversed an AI tool decision the delegate had already called. The delegate said nothing. The founder said nothing. But in the next planning session, the delegate checked before acting on anything AI-related. The programme was losing speed through an informal approval loop that had appeared without anyone deciding to create it.
The missing piece was a document.
What is a joint AI decision-rights map?
A joint AI decision-rights map is a one-page agreement, built in a single sitting between founder and delegate, that names which AI calls the founder retains, which the delegate owns outright, and which require both to decide together. The map covers the decision types the business will realistically face and is agreed before any of those calls lands, not after one has already caused friction.
The document runs to one page with three columns. One for the decisions the founder retains. One for the decisions the delegate owns outright. One for the decisions that require both. Against each decision type, you mark which column applies and agree what triggers a reconvene. Completing it takes less than an hour.
The exercise keeps both sides whole. The founder retains strategic oversight. The delegate operates with genuine room to move. Both can be true at the same time only if the boundary between them has been explicitly drawn, and that drawing is exactly what the joint map does.
Why does one-sided ownership fail?
When the delegate writes the decision-rights framework alone, it reads as a power grab. When the founder writes it alone, it reads as a leash. Both are trying to solve the same problem, but each one-sided version produces a resentment the other side cannot quite name. The version that holds is the one both sides built together, against the same list of decisions, in the same sitting.
The underlying dynamic is well documented in delegation research. When a founder hands off a mandate verbally but continues intervening on key decisions, the organisation draws its own conclusion. The AI programme belongs to the founder regardless of what was said. The delegate’s authority is conditional. Fruto.design describes this as the line between delegation and abdication, and it appears in AI mandates as reliably as in any other leadership handoff [43].
A rights document imposed by one party starts life as a fait accompli. Research on decision-making between co-leaders shows that frameworks work when they emerge from a shared process rather than being handed down [22]. A document produced jointly has already survived its first negotiation. Both sides gave something to reach it, which means it carries a weight that a unilateral document never can.
The process matters as much as the content. If the founder hands a written framework to the delegate for sign-off, the power dynamic of the handoff itself undermines the agreement. Build it together.
Which decisions actually need mapping?
The map only needs to cover the decisions that will create tension if made without prior agreement. For owner-managed businesses deploying AI, four categories cover the territory. Tool spend above a threshold, anything that touches customer data, decisions that change how a team works day to day, and anything that will appear in a board report or carry a public footprint are the categories worth mapping explicitly.
Tool spend is usually the easiest boundary to agree. Pick a threshold, say £500 per month per tool or £2,000 per year, and anything above becomes a joint call regardless of who identified the tool. Below the threshold, the delegate decides. This single rule heads off many of the “why did you sign up for this?” conversations before they start.
Customer data is where risk concentrates. Any tool that processes, stores, or acts on customer information carries regulatory and reputational weight that warrants the founder’s involvement. NACD’s AI governance guidance treats this as a board-level consideration rather than purely an operational one [37]. In practice, the delegate assesses the tool and recommends; the decision to proceed is joint.
Role impact is the most politically sensitive category. When AI changes what a team member’s day looks like, it touches employment, morale, and the founder’s relationship with the people in the business. The delegate should surface these calls early rather than managing them alone.
Board-visible and public-facing decisions carry the founder’s name regardless of who made them. Anything appearing in investor reporting or visible to clients is a joint call by default.
Three columns, four categories, one sitting. The document does not need to anticipate every AI decision the business will ever face, only the ones where an unchecked call would cause a rupture.
How does each side hold the line once the map is set?
Once the map exists, both sides have a different job. The founder’s job is to resist the reach-back, honouring the delegate’s column without reopening decisions that belong there, even when the instinct is to check. The delegate’s job is to escalate early rather than decide in the grey, surfacing joint-call items before they become fait accomplis rather than after.
For the founder, the reach-back is the harder discipline. Research on perceived loss of control shows it carries a real psychological cost, and founders accustomed to deterministic decisions often find AI’s unpredictability genuinely unsettling [30]. The map helps because it converts ambiguous territory into named categories. The founder defines where their authority ends rather than relinquishing it, and the delegate operates with explicit permission on the other side.
For the delegate, the temptation runs the other way. Grey-area calls feel faster to absorb than to escalate, particularly when the delegate is building credibility and does not want to appear indecisive. The risk is that a run of absorbed calls teaches the organisation that the delegate decides everything, which is the condition that triggers founder overreach. Escalating early is a sign of confidence in the map rather than a sign of weakness.
The reconvene trigger is worth agreeing in writing alongside the map itself. Against each joint-call category, name the mechanism beforehand, whether that is a fifteen-minute call before committing, a brief written note with a forty-eight hour response window, or a standing slot in the weekly rhythm. The mechanism does not need to be elaborate. It needs to be agreed and used consistently.
What sits alongside the decision map?
The decision-rights map is one part of a broader set of agreements that make a delegated AI mandate work. Related disciplines sit alongside it, and knowing where each fits prevents the map from being asked to carry more than it should. Three in particular come up in the same conversations, and understanding how they connect is worth a few minutes.
Getting decision rights in writing is the delegate’s solo move, the step where a mandate handed over verbally is documented so it is not subject to reinterpretation later. The joint decision-rights map is a different exercise, with the founder as an active party in building it rather than a signatory presented with something the delegate drafted alone. The two documents complement each other and are worth having both.
Verbal delegation interference is the pattern that emerges when the map does not exist or is not honoured. When a founder who handed off the AI mandate keeps reaching back into delegate decisions, the business gradually learns that ownership is ambiguous. Recognising this pattern after it appears is useful. The joint map built up front is what prevents it from appearing in the first place.
Founder engagement with AI is the third related question. BCG’s research on the AI adoption puzzle finds that companies stuck at proof-of-concept stage frequently share an absence of sustained leadership engagement [42]. The delegate cannot substitute for that engagement, but the decision-rights map creates a defined space for it. The map tells the founder where their presence matters and where the delegate operates freely, which is a cleaner basis for involvement than open-ended oversight.
If you have been running with a verbal handoff and no map, one honest conversation and an hour is enough to produce a document that makes the next twelve months materially easier. If you would like to work through the allocation together, Book a conversation.
