Annual planning is the moment when last year is converted into figures for next year. Staffing, cost per FTE, turnaround times: all based on how the work went last year. That is where a problem creeps in that plays less of a role at other planning moments. A reorganization or an investment decision forces you to look explicitly at assumptions, because the question itself demands it. Annual planning does not force that. You can simply carry last year's figures forward without ever asking whether the assumption behind them still holds.
And that is precisely what makes annual planning different from before, now that AI is in the picture. AI taking over work is not a separate project with its own decision moment. It happens at the margins: a team that speeds up a task, a process to which oversight has been added, a role of which part is already no longer done by people before anyone has explicitly recorded that. Annual planning counts hours and costs, but rarely checks whether the underlying assumption about who does the work is still the same as last year.
The question "how do you test your AI assumptions during annual planning" does not arise out of curiosity. It surfaces because annual planning is the one moment in the year when every department is required to weigh capacity and work against each other again. That is an exceptional opportunity. At no other moment is it asked so broadly and so simultaneously: how much work is there, and who or what does it.
The shift lies in the answer to that second part. Three years ago, the answer was almost everywhere "people, with some software alongside." Now, for parts of the work, the answer splits into three categories: work AI can handle independently, work where AI makes a proposal and a person approves or rejects it with a reason, and work that remains human work. That division differs per company, per department, and sometimes per task within a department. A customer service department that was fully staffed by people last year may this year have part of its inquiries in the first category. A legal department may largely remain in the third category, simply because the nature of the work leaves less room for it.
The difference between companies that already test this and companies that do not yet do so rarely lies in the technology. It lies in whether someone has asked the question. Companies where annual planning contains a standing question — which assumptions underlie these figures, and when was each assumption last confirmed — see the shift sooner, simply because they look for it. Companies where annual planning is a sum of last year plus a growth percentage only see the shift once the figures no longer add up. By then it is already a disappointing quarter, and the question is more urgent than if it had been asked during the planning itself.
There is also a difference in scale. Larger organizations with multiple departments see the pattern sooner, because one resembles another and differences stand out. Smaller organizations with less material for comparison have to look outward more actively — at sector data, at what comparable companies are doing — to see whether their assumption is still common practice or has already been overtaken.
During annual planning you can establish which assumptions underlie your figures at all. That seems simple, but in practice nowhere is it recorded who introduced which assumption and when. Naming them alone is already a step.
You can also establish when each assumption was last tested. Not whether it was correct at the time, but whether it has been reviewed since. An assumption that has not been touched in three years is not proof that it is wrong, but it is a signal that no one has checked it recently.
What you cannot establish during annual planning is the definitive division between what AI takes over, what happens under oversight, and what remains human work. That division is still changing, per task and per period, and a determination made today is no guarantee for next quarter. Nor is annual planning the place to base personnel decisions on; decisions about jobs are subject to their own legal requirements, separate from what an assumption test yields.
The underlying question — which work in this company can genuinely be taken over by AI — is answered per task by FTE TO AI's work scan, independent of when you put that on the agenda within annual planning.
Annual planning does not stand apart from other moments when assumptions come under pressure. Those who want to know how the same question arises after a disappointing quarter can read how you test AI assumptions after a disappointing quarter. Those preparing an investment decision in which AI capacity plays a role will find the approach for that in how you test AI assumptions before an investment decision. And for those who, after annual planning, want to record why a choice made sense at the time, there is how you record why you made a strategic choice.
During this annual planning, you can make a list of the assumptions underlying your figures, and note per assumption when it was last confirmed. That is not a major exercise; it is half an hour per department, provided someone asks the question.
Those who prefer an outside view can use the free assumption check: a short round in which you name your key assumptions and see, per assumption, when it was last confirmed. The full proof, with sector data and a self-plot of the management team 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.