A board member who says that AI delivers less in his domain than the rest of the board thinks doesn't usually hold a different opinion. He has different information, or older information, or information from a different part of the company where the shift is simply moving more slowly. AI takes over work in parts: in one department a task can already be handed over entirely to a system today, in another that same task is still human work because the data is messier, the risk greater, or oversight not yet in place. Two board members looking at the same company can therefore both be right about their own corner and still hold an opposing picture of the whole.
The problem is not that there is disagreement. The problem is if no one can point to where that disagreement comes from.
A self-plot does not ask for consensus, but for a claim per person: which part of the work in your domain is, in your estimate, ready to be taken over by AI today, which part partly with oversight, and which part remains human work. Lay the board members' plots side by side and the difference takes shape. Sometimes it is a matter of definition: one person already counts "partly with oversight" as taken over, the other does not. Sometimes it is a matter of timing: one formed his picture three months ago, the other last week. And sometimes it is a genuinely factual difference, because the departments simply are not moving at the same pace.
This calls for an assumptions list. Not because the list settles the dispute, but because it shifts the question from "who is right" to "when was this last checked". An assumption that hasn't been confirmed for two years carries different weight than an assumption from last month, regardless of who put it forward. You can read exactly what that means in how often an assumption about AI needs to be reconfirmed.
The stress test is an estimate, not a pronouncement from an oracle. In a few places, it is important to hold on to that distinction.
A self-plot measures perception, not capacity. If a board member assesses his domain optimistically because he has seen a pilot that isn't yet running in production, the plot measures that conviction, not the actual freed-up hours. That is why a single pilot should never count as proof for the entire domain; see why an AI pilot is not yet a strategy.
The stress test also says nothing about who is right when external data is missing. Sector data and regulatory clocks are thin or outdated in some domains. Where the outside world offers little to hold on to, the board's self-plot remains the only source, and then the outcome is a difference in estimation, not a measured fact. That must be stated explicitly alongside the outcome, otherwise the method suggests a precision that isn't there.
And the stress test does not settle a strategic choice. If two board members, after seeing the same plots, still assign a different weight to the risk of deploying AI too early versus too late, that is not a measurement error. That is a choice that remains with the board, with or without a plot.
Part of the disagreement on a board is not about estimation but about information that simply isn't distributed evenly. Whoever first sees that AI is taking over work is usually not the board itself but someone closer to the work: a team leader who notices that a reporting task is now done in a fraction of the time, or an employee who adds an oversight step to something that used to be done entirely by hand. Those signals reach the boardroom late and unevenly, which explains why one board member already knows something the other does not; see who in the organization is first to see that AI is taking over work.
The underlying question — which work in this specific company, in these specific departments, can genuinely be taken over by AI — cannot be answered with an opinion. That question is answered by the FTE TO AI work scan, task by task, which turns the difference between board members from a disagreement of opinion into a checkable fact.
Once the plots have been laid side by side and the definitions aligned, a remaining difference sometimes persists that is not a measurement issue but a priority issue: which domain gets attention first, which risk is acceptable, which assumption may go untested for another quarter. That is a board conversation, not a method conversation, and the approach for it is described in our board disagrees on priorities, what now. Matters touching on personnel decisions fall explicitly outside this: those are governed by their own legal requirements, and the stress test provides no basis for such decisions.
There is no need to first reach agreement before starting. A good first step is the free assumptions check: a short round in which you name the key assumptions underlying your strategy and see, per assumption, when it was last confirmed, and by whom. This often already reveals whether the difference on the board comes down to different definitions, different information, or a genuinely factual difference between domains. You can read more about what a strategic assumption is exactly and how you check whether it still holds at what a strategic assumption is and how you check whether it still holds. The complete stress test, with sector data, self-plot and assumptions list side by side, is under construction.
Stel uw vraag. Vaak zit de echte vraag een laag dieper — daar mag ik naar vragen.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.