An annual plan is written based on an assumption about how work is divided within the company: what people do, what systems do, and where the line between those two lies. That assumption is correct at the moment of writing. The problem is not that the assumption was wrong. The problem is that the line between human work and system work does not stand still while the plan remains on paper.
AI taking over work is not a separate development alongside the strategy in this respect. It is the way an assumption underneath the strategy expires without anyone having withdrawn it. There is no moment at which the executive board makes a decision that the assumption no longer holds. The work itself shifts first, usually within a team, usually gradually, while the annual plan in the meantime keeps calculating with the old division.
In practice, the shift runs along three lines. A task is fully taken over by AI. A task is partially taken over, with a human approving or rejecting and giving a reason for that. Or a task remains human work, because judgment, responsibility or context cannot be transferred.
This breakdown changes per quarter, per team and per task. An annual plan is written at the level of departments or functions, not at the level of tasks. As a result, it misses exactly the place where the shift takes place. Two companies in the same sector can score completely differently on this breakdown at the same moment: not because one is ahead of the other, but because one has already broken down its tasks enough to see where the line lies, while the other still works with the function as a whole.
The difference is rarely in the technology itself. It lies in whether a company has already split its own work into tasks that can be assessed separately. A company that continues to view invoicing, customer contact or reporting as a single function cannot see that part of it has already been taken over and another part has not. A company that does look at the task level often sees the difference emerging within its own team, even before a supplier or competitor says anything about it.
That also explains why discovering for yourself whether an assumption still holds often delivers less than expected: the assumption was never formulated specifically enough to be tested. "Customer service remains human work" is not an assumption you can confirm or withdraw. "The first answer to a standard question remains human work" is.
A picture of the shift is not a prediction. It does not say when a task will be taken over, and it does not guarantee that a task considered transferable today will still be so tomorrow. Regulation, sector agreements and the quality of the underlying data can shift that picture at any moment.
The outcome also says nothing if the input is not correct. A self-assessment by the executive team that is more optimistic than what is actually happening on the work floor produces a distorted picture, not a cautious one. It is therefore better to mark an assumption as uncertain than to confirm it silently: those two look different in a report, but lead to very different behaviour at the next check. And where two assumptions contradict each other — for example an assumption about the speed of adoption alongside an assumption about available capacity for oversight — the outcome is not an average. It is a signal that one of the two must be revised.
The stress test does not replace the annual plan and does not decide what happens to people. If a shift in the work touches on personnel decisions, its own legal requirements apply; that is explicitly not part of what an assumptions list or a horizon scan answers.
An annual plan has an owner for the whole, but rarely an owner per assumption. As a result, an assumption can remain unchanged in a strategic document for years, while no one in the organisation can say when it was last tested. That is precisely the question that often remains unanswered when no one is linked by name to an assumption: not because no one knows, but because no one has to know according to the structure of the plan.
This also requires a fixed outward view, on regulation and sector movement, instead of an annual catch-up. How that gets a structural place on the executive agenda instead of being treated as an incident is a question that deserves separate attention when regulation gets a fixed place on the agenda instead of a one-off exercise.
The underlying question — which work in this company can genuinely be taken over by AI, partly with oversight, or not at all — is answered by the FTE TO AI work scan per task, not per function or department.
An assumptions list does not prevent a strategy from changing. It prevents the change from going unnoticed. Anyone wanting to see how a strategic document becomes something other than reference material for the drawer will find a starting point on this in what keeps a plan usable after the first few months.
The free assumptions check is a short session in which you name the key assumptions underlying your strategy and see, per assumption, when it was last confirmed. This is not a judgment on the strategy, but an overview of where confirmation is missing. The full stress test, with sector data, regulatory clocks and the executive team's self-assessment side by side, is under development.
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.