A half day with a leadership team does not open with content. It opens with questions, because seven people who each authorize spend have already decided what they came to ask, and no agenda survives the first twenty minutes of a room like that. What you want to know before buying a session is which questions arrive, in what order, and which ones the room will fight about.
The questions carry the session because the decision has landed on the people in the chairs. BCG’s AI Radar 2026, a survey of 2,360 executives across 16 markets including 640 CEOs, found that 72% of CEOs now say they are the main decision maker on AI, roughly double the share a year earlier. Half say their job is on the line if it does not pay off.
Owning the decision is not the same as being fluent in what you are deciding. That gap is why the room exists.
The first forty minutes belong to them
Run a session for a sales team and you can start working inside ten minutes. The reps want to see the thing done on a real call note. Open the same way with a leadership team and you get nowhere, because the COO has a question she has been carrying since the last budget review and she will not practice anything until it is answered.
So the opening block is handed over. Everyone says what they came to ask, the questions go on a board, and nothing is answered yet. People find that mildly irritating, which is fine: the list becomes the agenda, and seven versions of the same worry written down does more for alignment than a slide about alignment would.
The list is stable. Across sessions at companies of 50 to 500 people, the same nine questions come up with small changes in wording. Who speaks first shifts the order. The content does not move, which is the one dependable thing about AI for leadership teams as a category: the curriculum changes with every vendor, the questions do not.
The nine questions, and what sits underneath each
| What they ask | What is underneath it | What the room does with it |
|---|---|---|
| “Which tool should we standardize on?” | A procurement decision standing in for a capability decision | Park it. Work on whichever account is already paid for |
| “What am I allowed to put into it?” | Nobody told them, and they suspect they have been doing it wrong | Hands-on redaction on their own documents, 30 minutes |
| “Am I going to be wrong in public?” | Reputational fear, usually the real reason attendance is patchy | A verification habit, practiced until it is boring |
| “What do I tell the board?” | They have read two headlines that contradict each other | Read both numbers properly, write three sentences |
| “Who owns this?” | It has landed on someone as a second job | Name the owner and the hours in writing, in the room |
| “Is this a headcount decision?” | The question the CFO came to ask | Separate it from the session and schedule it properly |
| “Do I need to learn this myself?” | Whether delegation is available as an option | Yes, and the time cost gets a number |
| “What about the people who refuse?” | A management problem being handed to a vendor | Draw the line between skill and performance |
| “How will we know it worked?” | They have been asked for a return and cannot answer | Two baselines, taken before the room empties |
Six of those resolve in five minutes once said out loud. Three do not.
Our role-by-role view of what each function needs explains why leadership is the one group whose working material is decisions rather than deliverables. That is why the questions do the structural work here and drafting exercises do it everywhere else.
The three that take the most time
“What do I tell the board?”
This eats a block, because the room holds two numbers it has not reconciled.
The first is MIT Project NANDA’s GenAI Divide study, which reported that 95% of generative AI pilots produced no measurable P&L impact. It drew on 150 leader interviews, a survey of 350 employees, and 300 public deployments, and its bar was a production deployment with sustained profit impact inside roughly six months. A strict test, and a finding about measurement rather than about the tools.
The second is Wharton’s 2025 AI Adoption Report, where 74% of businesses that measure return from generative AI said they were already seeing a positive one. Two caveats: the denominator is firms that measure, and the sample is US companies above 1,000 employees and $50 million in revenue, larger than most teams in this room. Wharton’s Stefano Puntoni said of the contrast, “Our definition is very different from the MIT report. Theirs is much more stringent.”
Both are honest, and they answer different questions. A team that can say that in three sentences is readier for a board meeting than one that has memorized either figure, so the block ends with the group writing those sentences themselves.
“Who owns this?”
At this size there is no Chief AI Officer. The work has landed on a VP of Operations or a Chief of Staff who already had a job. The room usually knows and has been careful not to say it.
So it gets said, then written down: a name, and hours per week. The hours matter more than the name. Without them the role is a title that gets squeezed out by the first real deadline, which is how most AI enablement efforts quietly stop. The same pattern sits behind the difference between buying licenses and building capability.
“Do I need to learn this myself?”
The honest answer is the least popular one in the room, and the data backs it. The same BCG survey sorts companies into three groups by how far along they are, and in the most advanced of them, chief executives spend more than eight hours a week on their own AI upskilling. That group also puts 60% of its AI budget into upskilling and capability building, against 27% and 24% for the two groups behind it.
Eight hours a week is not a realistic ask for every leader. The workable version is smaller: one recurring task per leader, done with the tools for six weeks, long enough to form a judgment. A leader who has never drafted anything this way cannot tell a good proposal from a bad one, and will approve both.
The question nobody asks
Nine questions arrive. A tenth almost never does: what are we going to stop doing?
Every leadership team treats this as additive. New capability, same meetings, same reporting, same approval chain, nothing returned to the calendar. A plan with no subtractions stops in week three, because the people carrying it have no hours.
The fix is unglamorous. Before the room empties, each leader names one thing their function will stop doing, or do less often, to pay for the new habit. A weekly summary that becomes monthly. A report nobody reads. It is the only agenda item that reliably survives a busy quarter.
What the half day actually contains
| Block | Minutes | What happens in it |
|---|---|---|
| The questions on the board | 40 | Everyone says what they came to ask. Nothing is answered |
| What you can put in, in practice | 30 | Hands-on redaction and account settings on their own material |
| Working the three hard questions | 80 | The board answer, the owner, the personal-time question |
| One task each, done live | 45 | Every leader works a real recurring task of their own |
| Decisions, subtractions, two numbers | 30 | Written output, owner named, baselines taken |
About three and a half hours of working time plus breaks. A full day is worse, not better: leadership rooms empty after lunch as calendars reassert themselves, and the last two blocks decide whether anything changes.
The hands-on block is not decoration. A team that discusses AI enablement and never touches it will approve a program it cannot judge. Our account of what a half day with a people team contains shows the same pattern where the cost of getting it wrong is more visible.
What they bring in
Four things, requested a week ahead:
- One recurring task per leader, in their own words. The monthly board update, the pipeline review, the vendor comparison, the hiring scorecard. Real and unfinished.
- Whatever the company already pays for. Which of ChatGPT, Claude, Gemini, or Microsoft Copilot is approved, who administers it, what the retention setting is.
- The last thing they were asked for and could not answer. Usually a number, usually from a board member.
- The policy, if there is one. Often two paragraphs in a handbook nobody in the room has read.
Do not spend the morning choosing between products. Tool selection is the question the room most wants to answer because it feels like a decision, and it has the least effect on whether capability moves. The skill transfers between these products. The license negotiation does not transfer into anyone’s Tuesday.
Where the room goes quiet
Twice, usually.
The first is when someone pastes real material and sees what comes back. There is a particular silence when a leader watches a tool produce a serviceable draft of something that takes her an hour. Not delight. Recalculation, and worth leaving alone for a minute.
The second is colder. Someone says a number out loud about staffing and the room stops. The move is to name it and take it off the table: a real question, not one a working session can answer responsibly, and one that needs its own meeting with the people who hold the data. Leaving it half-discussed is worse than either, because the room then argues with a version of it nobody said.
Writer’s survey of 1,600 US knowledge workers, 800 of them C-suite, found that 68% of executives said AI adoption had caused division at their company, and 42% said that division was tearing the company apart. Half-finished conversations at the top are how that starts.
Three ways the session fails
It becomes a strategy offsite. Three hours of discussion, a deck, nobody touches a tool. The group leaves with vocabulary and no judgment. The most common failure and the hardest to see, because the room feels productive.
It becomes a demo. A facilitator shows impressive things on sample data. Everyone is entertained, nobody tries a task of their own, nothing transfers to Monday.
It leaves without an owner or numbers. Akkodis found executive confidence in AI strategies fell from 69% in 2024 to 58% in 2025 across 2,000 executives, with only 55% believing their teams fully grasped the risks and opportunities. Confidence drops when plans are made and not carried. The fix is assignment, not more planning.
What a session like this costs
Facilitated sessions for a leadership team sit in the $2,000 to $15,000 market range, depending on length, preparation, and whether a follow-up is included. The line nobody budgets costs more: seven senior people for half a morning is a real number before any invoice arrives.
Which argues for doing it once, on real material, rather than three times as an introduction. If you are weighing a facilitated session against university or vendor programs for individual leaders, we compared what each of those products actually buys on price and on the hours they hide.
How you know in ninety days
| Signal | How to read it at ninety days |
|---|---|
| Cycle time on one named recurring task per leader | Measurably shorter on at least half, or the habit did not take |
| Hours per week assigned to the named owner | Still on the calendar, or the role was a title |
| Subtractions actually made | At least one per function, visible on a calendar |
| Whether the board question came back | And whether the answer beat three sentences |
Deloitte’s Global Boardroom Program, covering 695 board members and senior executives across 56 countries, found 31% say AI is not on the board agenda, down from 45% in the prior edition. The question reaches more boards each year, so the three sentences get asked for again. The second asking is the real test.
Frequently asked questions
Who should be in the room for a leadership AI session?
The people who can authorize a change to how their function works, plus whoever holds the budget. Usually five to nine. Above about ten the question block becomes a queue and the hands-on block stops working, because people watch instead of typing.
How long should a session for a leadership team run?
About three and a half hours of working time in one sitting, with a shorter second session two or three weeks later. Full days lose the room after lunch, and ninety minutes fits the questions or the hands-on work, not both.
Should leadership be trained before or after the rest of the company?
Before, and not by much. Leaders who have not used the tools approve programs they cannot judge and set rules they cannot defend. Two to four weeks is enough: long enough to form opinions, short enough that nobody notices a pause before the wider AI enablement work starts.
What should a leadership team walk out holding?
Four written things: three sentences for the board, a named owner with weekly hours, one recurring task per leader to run for six weeks, and one subtraction per function. Less than that is a conversation, not a decision.
What if the CEO does not attend?
Reschedule. With 72% of chief executives naming themselves the main decision maker on AI, a session without the person holding the decision produces recommendations that need re-litigating. The exception is a functional leadership team inside a larger company, where the division head is the relevant decision maker.
How do we measure whether AI training for employees worked at this level?
Not through usage dashboards, which count logins. Baseline cycle time on one named task per leader, count the subtractions made, and check whether the named owner still has the hours ninety days later. Three numbers beat a license chart.
Can we send one person instead of the whole team?
You can, and it answers a different question. One seat on a program builds one internal reference point. A session for the team produces a shared standard and decisions the group has already agreed to.
What about leaders who refuse to engage?
Treat it as a management question, not a training question. A session can remove the reasons people give for refusing, usually fear of being wrong in public and uncertainty about what is permitted. It cannot supply the will. If both reasons are gone and the refusal remains, the next conversation belongs to someone’s manager.
Key takeaways
- A leadership session is carried by the questions, not the curriculum. Hand over the first forty minutes, put the questions on a board, and answer nothing until they are all up there.
- Nine questions arrive in nearly every room. Six resolve in minutes. Three take real time: the board answer, the owner, and whether each leader has to learn the tools personally.
- The board question is two contradicting headlines. MIT’s 95% finding used a strict six-month profit test; Wharton’s 74% counts only firms that measure, at companies larger than most in the room. Both are true and they answer different questions.
- The tenth question never gets asked: what are we going to stop doing. Without one subtraction per function, the plan has no hours.
- Ownership without fluency is the failure mode. 72% of CEOs now own the AI decision, while in BCG’s most advanced group of companies the chief executive spends eight hours a week with the tools personally.
- Take two baselines before the room empties: cycle time on one named task per leader, and the owner’s weekly hours.
If your leadership team is being asked to show a return on the AI licenses you already bought, book a free AI readiness call. We will go through the nine questions with you and tell you which three your team is going to fight about.
Arthur Wandzel