Anyone asking what being a year late on AI costs usually expects an amount. That amount does not exist without assumptions you must supply yourself: how many hours in your organisation can already be described as a task, what share of that can be handed over to AI under human oversight, and what those freed-up hours are worth in your business. Change one of those three and the outcome changes with it. An estimate that pretends to be exact is worth less than an estimate that shows what it depends on.
What can be described: the delay itself. Not as an amount, but as capacity that is not being freed up this year while it already is elsewhere in the sector.
AI is taking over work. Not everywhere, not all at once, and not as a finished project with an end date. It happens task by task, and every task falls into one of three categories: AI can take it over, AI can do it under human oversight that approves or rejects with reason, or it remains human work. That classification shifts. A task that was entirely human work last year may fall into the middle category this year. This is the kind of change under which a strategy quietly ages: not because the plan was wrong, but because an assumption beneath the plan has lapsed without anyone withdrawing it.
The assumption often runs something like: this work costs us so many FTE, so we compete this way. If AI partly takes over that work for a competitor and not for you, the comparison shifts while the plan remains unchanged on paper. That is precisely why how do you recognise that your plan rests on outdated figures is a question that cannot be answered once a year. An annual plan is fixed at a single moment; the shift continues after that moment, and why an annual plan does not keep up with the AI shift describes exactly where that gap arises.
One company already works with AI that prepares reports or performs initial analyses, with an employee checking and deciding. The other company still does the exact same task entirely by hand. That difference rarely stems from ambition or budget. It stems from who in the organisation owns the question of which tasks are describable and transferable. Where nobody asks that question, the task stays as it always was, even if the technology has since moved on.
That is a governance question, not a technology question. Somewhere, someone must maintain the assumptions list and record, per assumption, when it was last checked. Without that owner, the list dilutes into a document nobody opens anymore, and that is the situation addressed by who guards the assumptions if nobody owns them.
"We are falling behind on AI" is not an assumption you can test. "This process currently costs us twelve FTE and will keep costing that for the next two years" is, because that sentence can be refuted or confirmed once it is known what share of that process has since become transferable under oversight. The translation from feeling to testable statement is exactly where most strategy processes get stuck: people sense that something is off, but nobody turns that into a sentence that can be contradicted by data. how do you reduce an assumption to something you can measure describes that step.
This is also where the limit of what a delay estimate can say lies. It can show that a task of ten FTE has largely become transferable under oversight, and that this transfer did not take place this year. It cannot say what should have happened with those ten FTE. Whether freed-up capacity is redeployed, phased out, or used for other work is a decision for the employer, with its own statutory requirements where that decision affects personnel. The estimate delivers facts about work, not personnel advice.
A calculated delay is only useful if the underlying figures are current. A sector benchmark that is two years old, an assumption that has never been reconfirmed, or a board's own projection that nobody has updated recently: each of these three makes the outcome unreliable, even if the calculation method itself is correct. A number derived from outdated input looks precise but says nothing. This is also why disagreement within a board about what takes priority is often not a difference of opinion but a difference in data: each board member calculates using their own, unshared picture of the assumptions, and our board disagrees on priorities what now shows how that difference can be made visible before it blocks a decision.
The stress test only works if the right people are at the table: not only the board that wrote the plan, but also those who see daily which tasks are already shifting. who should take part in a strategy review covers that composition, because an assumptions list filled in only by the top misses exactly the signals coming from the shop floor.
The underlying question — which work in your business can genuinely be taken over by AI, where it must remain under oversight, and where it stays human work — is answered task by task with the work scan from FTE TO AI.
A year's delay is not a fixed amount, but it is a question you can already start answering: which assumptions underlying your current plan have never been tested against what AI can already do this year. The free assumptions check is a short round in which you name your key assumptions and see, per assumption, when it was last confirmed. The full stress test, with sector data, regulatory clocks and your board's own projection set 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.