Most leadership teams are asking the wrong questions about AI. Not stupid questions. Wrong ones. Which tool should we use? Is it safe? What can it do? Should we be worried about the new model? Those are reasonable things to want to know, and every one of them has the same flaw: someone outside your organisation could answer it for you. A vendor could. A consultant could. A well-prompted chatbot could. And a question that can be answered from outside the building is not a leadership question, however senior the people asking it.
The questions that decide whether AI works for you are the ones nobody else can answer, because they are about your priorities, your people and your appetite for change. This article gives you the ten I use with leadership teams, the filter that separates them from the rest, and the way to work through them in a single meeting.
The Outsource Test
The Outsource Test is the one filter I ask leadership teams to run every AI question through before they spend a meeting on it: could someone outside the organisation answer this for you? If they could, it is not a leadership question, and it should not take up the room's time until the questions only you can answer have been dealt with.
I have watched this filter clear a whole agenda. A senior team arrives with a page of questions about tools, models and risk, most of them gathered from webinars and supplier decks. Run them through the test and perhaps two survive. The rest can be delegated, bought in or looked up. What is left is uncomfortable, because it is about the organisation and not the technology, and the organisation is the thing the people in the room are responsible for.
The principle underneath is one I return to constantly: outsource the doing, not the thinking. Leadership teams have been quick to hand AI the doing. They have been far slower to notice that they have been handing the thinking to vendors too.
Why the questions at the top matter more than the tools
The evidence says that what a leadership team pays attention to predicts what it gets from AI, more than which tools it chose. Protiviti and BoardProspects surveyed 772 board members and executives in late 2025 and published the findings in March 2026: only 26% of boards discuss AI at every meeting, but among organisations reporting a high return from AI the figure is 63%, against 13% among those reporting a low return. Attention at the top is not a nice-to-have. It is the variable that separates the two groups.
PwC's 29th Global CEO Survey, published in January 2026 from 4,454 chief executives in 95 countries, shows what happens without it. More than half (56%) say AI has yet to produce a financial benefit they can point to, and only one in eight (12%) report gains on both sides of the ledger. The same survey found chief executives spending 47% of their time on issues less than a year out and 16% on anything beyond five years. AI is being led as an operational issue by people with almost no time for the strategic one.
Education is not exempt. Teach First and Accenture's report, AI in Schools: What School Leaders Need to Know (30 June 2026), found that only 2% of schools in England have a formal AI strategy and 12% have any policy at all, while AI is already embedded in daily teaching. Their conclusion, that confidence, capability and organisational capacity rather than technology will limit adoption, is a finding about leadership, and it matches what I see.
The ten questions only a leadership team can answer
The ten questions below are the ones that pass the Outsource Test, in the order I put them to a leadership team, so that each answer makes the next one easier. Work through them in sequence. Skip one and the later answers get vague.
1. What would we be trying to change this year if AI did not exist?
Start here because AI attached to nothing changes nothing. Every organisation I work with already has two or three things it is desperate to move: outcomes for a particular group, the time senior staff lose to reporting, the quality of a service that has drifted. AI is useful in proportion to how tightly it is anchored to one of those. A weak answer names AI as the priority. A strong answer names the existing priority and only then asks whether AI can move it. If your team cannot answer this question without saying the letters A and I, the strategy has been written backwards.
2. Which of those changes could AI move, and which could it only decorate?
Be honest about the difference between doing what you do slightly faster and doing something you could not do before. Both matter, and they need different decisions. The first kind (drafting, summarising, first passes at analysis) is where most organisations should begin, because it builds confidence with low risk. The second kind (a different model of feedback, a different way of allocating expertise) needs a sponsor, a pilot with an exit and a willingness to stop something else. A weak answer lists twenty use cases. A strong answer names two of the first kind and one of the second, and says which is which.
3. What will we not use AI for, and could we say so in public?
The line you draw is the fastest way to earn the trust that adoption depends on, and no vendor can draw it for you. I ask teams to write the sentence they would be comfortable putting on the website: "We do not use AI to..." For a school it might be decisions about a child's future, or communication with a family about a serious concern. For a business it might be final decisions about people. A weak answer says "we will always keep a human in the loop", which means nothing until you say where. A strong answer names the decisions, and the people affected have been told.
4. Who owns this, and what have we taken off their plate to make room?
Ownership without capacity is a name on a slide. Across the leadership teams I work with, the most common pattern is a capable person given AI as an addition to a full role, with enthusiasm as their only resource. Six months later the enthusiasm has gone and so has the momentum. A weak answer names a person. A strong answer names a person and a deputy, names the thing they have stopped doing to make room, and gives them the authority to decide something without asking the whole team first.
5. Who is already using AI here without telling us, and why did they not tell us?
Somebody in your organisation is already using AI for work, and the reason they have not mentioned it tells you more about your culture than any survey. In my experience the answer is rarely that people are hiding something. It is that nobody asked, or that the last message they heard about AI was a warning. A weak answer treats quiet use as a compliance problem. A strong answer treats it as free information: these are your early adopters, and their reasons for silence are the first thing your approach has to fix.
6. What are we asking people to stop doing?
Adoption fails when AI is added to workloads and nothing is removed from them. This is the question leadership teams most want to skip, because it means choosing. If AI drafts the report, does the person still write the report? If it produces first-pass feedback, is the old expectation withdrawn? I make the point in the Sponsorship Test, and it bears repeating: have we removed what it replaces? A weak answer says people will find their own efficiencies. A strong answer names the task, the date it stops being expected and who has told the people who used to do it.
7. Which decisions may AI inform, and which must a person still make and be seen to make?
Decision rights are the part of governance that a policy document cannot settle for you, because they depend on what your organisation is for. A model can inform a judgement about a pupil, a patient, a candidate or a customer. It cannot own it, and the people on the receiving end need to know that a person did. A weak answer relies on the phrase "human oversight" and leaves it at that. A strong answer lists the decisions where AI may draft, the decisions where it may recommend, and the decisions it is kept away from entirely, and the list is short enough for staff to remember.
8. What would we need to see, by when, to know it is working?
Usage is not a result. Logins, licences and the number of staff who have tried a tool tell you nothing about whether the thing you named in question one has moved. I ask teams to write the evidence they would accept before they start, with a date, because evidence chosen afterwards always flatters the decision already made. A weak answer measures activity. A strong answer measures the change from question one, on a date that has been put in the diary, and names who will say out loud whether it happened.
9. What happens the first time it gets something wrong, and who has rehearsed it?
It will get something wrong, and the organisation that has rehearsed its response is the one that keeps its permission to continue. The failure is rarely dramatic. It is a confidently wrong letter, a summary that missed the one thing that mattered, a piece of feedback a student took at face value. A weak answer points to a policy nobody has read. A strong answer names the person who is told first, what they do in the first hour, how the people affected are informed, and when the team last walked through it.
10. What will we be asked in a year that we cannot answer today?
The questions coming towards you from governors, boards, parents, regulators, customers and staff are already visible, and a leadership team that lists them now can choose its answers instead of improvising them. Why did you let AI do that? Why did you not? What did you tell the people affected? Who decided? Ask each member of the team to write down the question they would least like to be asked in twelve months' time. Then read them aloud. A weak answer assumes the questions will wait. A strong answer treats the list as next year's agenda, written early.
What I Tell Leadership Teams
The ten questions are not a survey to fill in. They are a single meeting, taken in order, with the answers written down and dated. Since I started recording a daily podcast, the questions leaders send in have changed noticeably: fewer about which tool, more about what to say to a board, a governing body, a union or a room of anxious staff. Those are Outsource Test questions in reverse. Nobody else can answer them, which is exactly why they are the ones leaders lie awake over.
The same pattern shows up at the level of a whole system. In advisory work with government bodies, including the Department for Education, KHDA in Dubai and the Ministry of Education in Kazakhstan, the questions I found myself returning to were never about technology. They were about who would decide, what would stop, and what the public would be told. The scale changes. The questions do not.
Three things I say in every session. First, the quality of your AI approach is set by the questions you refuse to delegate, not by the tools you select. Second, if a question can be answered by a supplier, let a supplier answer it, and spend your time on the ten that cannot. Third, the discomfort you feel at question six is the point. That is what it feels like when a leadership team is doing its own thinking.
How to use the ten questions
Use the ten questions as the agenda for one ninety-minute meeting of the senior team, in the order above, with one person writing the answers as they are agreed. Do not aim for perfect answers. Aim for honest ones, with the gaps named. The gaps are your plan.
Then put the answers through the Readiness Test, the three questions everyone in the organisation must be able to answer consistently before it can call itself AI ready, which I set out in what it actually means to be AI ready. If the leadership team cannot answer its ten, the organisation will not be able to answer its three. If the ten are answered and written down, you have most of what a strategy needs, and an AI strategy that people actually use is a short document away. And if a pilot is already running, take question eight straight to it and ask the Exit Question before the next one starts.
If your leadership team is working through these questions, this is the kind of work I support through AI strategy sessions and executive advisory work, where the ten questions become the first hour and the answers become the plan. You can find out more about how I work with leadership teams.
Dan Fitzpatrick is the founder of The AI Educator and works with leadership teams, schools and organisations on what to do about AI. More about Dan.
