A board makes decisions in a sequence. Investments, job profiles, capacity planning, location choices: they usually rest on an estimate of how much work there is and who does that work. That estimate is in motion right now. Not everywhere at the same pace, and not always visible at the moment the decision is made.
AI does not take over industries. It takes over tasks, in parts, with varying degrees of oversight. Part of the work can be done entirely by a system. Another part remains human work with AI as a tool, where a person approves or rejects and substantiates that with a reason. A third part does not change for the time being.
The problem for decision-making is not that this shift exists. The problem is that a decision is often made at the moment the split between those three categories is still shifting, and that no one in the meeting names that explicitly. The decision proceeds based on how the work was divided last year.
A few categories in which a decision made too early typically occurs:
Capacity planning for a team or department. If part of the tasks within that team is a candidate for takeover by AI, an fte estimate that is fixed now says little about the need in twelve to eighteen months. The bandwidth of the estimate depends on how quickly oversight requirements in that specific task change, and this differs sharply per role.
Investments in fixed capacity or premises based on a growth forecast. That forecast often assumes a productivity per employee that can shift due to AI support, in either direction. An investment based on the old ratio can turn out to be either too generous or too tight in hindsight.
Reorganization of job profiles. When a role largely consists of tasks that fall into the second category — partial takeover, with human oversight — the content of that role changes faster than a job description is normally revised. For decisions that directly affect personnel, separate legal requirements apply, incidentally; these are not addressed here and are not substantiated either.
Strategic choices resting on a market share or cost position. If competitors have already reallocated part of their work toward AI support, their cost structure shifts. How you can detect this before it becomes visible in a competitor's figures is described in an explanation of how to recognize competitors shifting faster than you.
Postponement is not advice to wait until everything is clear. That moment does not come, because the shift keeps progressing. Postponement means something specific here: waiting on the decision for a concrete answer to one thing — which part of the underlying work in this company, in this department, is right now a candidate for takeover, and which part is not. Once that answer exists, a decision no longer needs to rest on an assumption that no one has checked recently.
The cost of that waiting is not zero. Shifting a year too late has a price, and that price is not the same everywhere; an elaboration on what a year's delay in the AI shift can cost shows what those costs depend on. Postponing a decision is therefore not a non-committal choice, but a trade-off against the risk of a decision made on outdated ground.
An assumption is not outdated by definition simply because time has passed. It is outdated when the circumstance it rested on has changed without the assumption being adjusted. In the case of AI takeover of work, that circumstance is the split between full takeover, oversight with approval, and human work. Signals that a plan rests on outdated figures are addressed in a description of how a plan comes to rest on outdated figures. An annual plan is, moreover, usually revised on a fixed cycle, while the AI shift does not adhere to that cycle; why an annual plan fails to keep up with that shift is worked out in an explanation of the mismatch between annual planning and AI pace.
The underlying question — which work in this company can truly be taken over by AI, which part remains under oversight, and which part remains human work — is answered per task with the work scan from FTE TO AI.
This way of looking at things has limitations worth mentioning. A task-by-task assessment says something about the technical and organizational feasibility of takeover at the moment of measurement; it says nothing about the speed with which a supplier, regulator, or competitor activates that feasibility. Two companies with identical task portfolios can therefore arrive at different outcomes, simply because one company gets access to an application sooner than the other.
The outcome is also not a prediction with a fixed validity period. It holds until the moment one of the underlying assumptions changes — a new oversight requirement, a supplier expanding a feature, a competitor visibly reallocating a task. How often a strategy therefore needs recalibration cannot be captured in one fixed interval; a discussion of the frequency with which a strategy tolerates recalibration addresses what determines that frequency. And when the market visibly changes faster than the company's own plan, the question is not only when to recalibrate but also how; that situation is addressed in an explanation of what to do when the market overtakes the plan.
Anyone with a decision on the agenda that rests on an assumption about work and AI can name that assumption first before the decision is made. A free assumption check offers a short round for this: you name the key assumptions underlying your current plan, and for each assumption it becomes clear when it was last confirmed. The full strategic pressure test, with sector data, a self-plot for the executive team and an ongoing assumptions list, 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.