A pilot answers a narrow question: can this specific piece of AI, in this specific environment, take over this specific task. That is a useful question, and the answer is often usable. But a pilot does not test whether the assumption underlying your strategy still holds. That assumption was broader: that a certain part of the work in your organisation remains human work, that a certain function remains scarce, that a competitor cannot quickly build up a certain capacity. A successful pilot in one corner of the company says little about that. A failed pilot even less, because it may be down to the data, the tooling, or the people who carried it out, and not to the question itself.
The difference between a pilot and a strategic test is the scope of the question. A pilot asks: does this work here. A strategic test asks: on which assumption does our plan build, and is that assumption still true today. Those are different questions, with different answers, and a management team that answers the first question and thinks it has dealt with the second is working with a blind spot that no one has noticed.
AI taking over work does not happen as an announcement. It happens task by task, team by team, and usually without a decision preceding it that the whole organisation sees. An assumption in a strategy plan from two years ago — that a certain analytical function remains scarce and time-consuming, that a competitor cannot quickly scale up a certain service — may in the meantime have lapsed without anyone withdrawing it. Not because the assumption was wrong when it was made, but because the world beneath it has shifted while the document has stayed in place.
Three categories run through every piece of work: tasks that AI can already take over, tasks that partly shift over with human oversight that approves or rejects, and tasks that remain human work. Which task falls into which category differs per company. It depends on the quality of the underlying data, on how much context a task requires, on whether regulation requires human approval, and on how the work is currently organised. Two companies in the same sector can therefore have completely different outcomes, and both can be right for their own situation.
A pilot is local by design: one team, one process, one period. That is precisely why it does not show whether the assumption underlying the strategy as a whole still holds. A pilot can succeed while the broader assumption has long since ceased to be true, because the pilot happens to sit in a corner the shift has not yet reached. A pilot can also fail while the assumption elsewhere in the company is already outdated, because the failure says nothing about other departments or other tasks.
Anyone who wants to know whether a strategy still holds should therefore not ask whether one AI application works, but which assumptions carry the strategy and when they were last tested. That is a different exercise than running a pilot, and it is the exercise that explains why how you distinguish an AI hype from a shift that stays cannot be answered with a single experiment.
The companies where this is already well arranged generally have one thing in common: someone in the organisation structurally signals when a task shifts category, and that signal reaches whoever determines strategy. In other companies that signal stays stuck with a loose team, or it is not noticed at all because no one owns that task. Who in the organisation is first to see that AI is taking over work is therefore not always the person with the most mandate, and that gap between who sees it and who decides on it is where strategies remain stalled the longest without anyone noticing.
The underlying question — which work in this company can genuinely be taken over by AI, and which work cannot — is answered per task by the work scan from FTE TO AI, with an estimate of what AI can take over, what requires oversight and what remains human work.
This approach estimates, it does not measure. The category into which a task falls — transferable, with oversight, or human work — is an estimate based on what is currently known about the task, the sector and the available technology. That estimate may change as soon as the technology changes or as soon as the task itself is organised differently. An outcome established three months ago says nothing more about today if the underlying assumption has since lapsed. That is also why a one-off test is never sufficient and why the question how you keep a strategy current without revising it every quarter requires a different approach than an annual strategy day.
Where an outcome touches on personnel decisions, a different framework with its own statutory requirements applies; this method does not provide grounds for dismissal and does not offer personnel advice. It provides a picture of which work is likely to shift, so that management knows which assumption in the plan still stands and which does not, and which decisions can therefore wait — a question set out in which decisions you would do better to postpone until the AI question is answered.
A pilot shows whether something works. A strategic test shows whether the plan still holds. Anyone who wants to start with the latter can take 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 sector data, regulatory clocks and the management 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.