A management team plots its assumptions and sees something uncomfortable: two assumptions, both stated in the strategy document, contradict each other. One says a department is growing because demand is increasing. The other says the same department is shrinking because part of the work is now being done by AI. Both assumptions were once written down for good reason. The problem is not that one of the two is foolish. The problem is that no one has recorded at what point which assumption should give way.
A contradiction between assumptions is usually not a calculation error. It is the visible remainder of a shift that has come through differently in two places within the company. One assumption was written by someone who sees the department growing in number of requests. The other by someone who sees that part of those same requests are now being handled without human hands, with oversight that approves or rejects. Both observations can be correct. They simply have not been reconciled since AI took over part of the work.
That is where the method becomes useful, and also where it has its limits. The stress test exposes a contradiction; it does not explain who is wrong. That requires a conversation, not a calculation.
A management team's self-plot is a snapshot of what people think they know, not of what is true. If two board members both mark an assumption as "valid," that means they agree — not that the assumption is correct. Conversely: if an assumption is marked "doubtful" because one person has seen a signal the others are not aware of, that is a reason to verify it, not confirmation that the assumption is void.
The external data against which the self-plot is compared carries its own uncertainty. Sector data on the extent to which AI is taking over work lags behind practice. Regulation shifts the playing field before the figures show it. A competitor communicating about automation does not always say what is actually happening internally. The stress test combines these sources to make a pattern visible, not to guarantee an outcome. Where the picture is thin — one source, one quarter, one region — the stress test indicates that it is thin, rather than hiding it behind a seemingly precise figure.
An outcome, then, says nothing on its own. It says something in combination with the question: how recently was this confirmed, and by whom.
In one company, an assumption about AI is adjusted within a quarter as soon as the first task is actually taken over. In another company, the same assumption still stands unchanged in the document after two years, while the task it rests on is already largely handled by AI. The difference rarely lies in the quality of the people. It lies in whether there is a fixed moment at which assumptions are retested, and whether that moment is independent of who is shouting loudest at that time that something needs to change. Companies where the loudest voice does not automatically determine which AI application takes priority let assumptions stand or fall based on what can be observed, not on who dominated the last meeting.
There is a difference between a risk you consciously accept and an assumption you have lost track of without noticing; the distinction between an AI risk and an AI assumption explains why two assumptions sometimes clash not because they are wrong, but because one describes a risk and the other a fact that has since changed. Once that distinction is made, it also becomes clearer why the contradiction does not necessarily call for a decision, but for an update of what is assumed to be true.
The value of the stress test does not lie in the moment of contradiction itself, but in what happens with that contradiction afterward. If it is recorded which assumption was adjusted and on the basis of which signal, that can be traced back later. If it remains just a conversation without a note, the same contradiction will resurface in a year, with the same confusion. Anyone who wants to be able to explain afterward why a strategic choice was made at the time also needs to have recorded why an assumption about AI has become void — not as an archival record, but because otherwise the next decision-maker starts again from zero.
Important: all of this concerns recognizing and tracking a shift, not what an organization subsequently does regarding its staff. Where an outcome touches on a decision about people, separate legal requirements apply to that; the stress test is not a basis for that decision.
A management team that has once found a contradiction often wonders how often something like this repeats itself. The answer depends on how quickly AI in the sector concerned actually takes over tasks and how visibly that comes to light. That is also the question behind how often a strategy deserves recalibration: not on a fixed calendar date, but at the moment the assumptions and reality begin to diverge again.
The underlying question — which work in this company can genuinely be taken over by AI, which part with oversight, and which part remains human work — is answered per task with the work scan from FTE TO AI. If you would first like to see where your own assumptions stand before figuring that out, there is the free assumption check: a short round in which you name your key assumptions and see, for each one, when it was last confirmed. The full stress test, with the external comparison and the self-plot 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.