The assumption about how many people this work costs

Where the assumption comes from

Every budget, every organizational chart and every capacity plan rests on an estimate: this work costs so many people. That estimate was made at some point based on how the work actually proceeded — how much time a quote took, how many files an employee processed per week, how much consultation was needed to substantiate a decision. At the moment that estimate was established, it was correct. It was a description of reality, not a guess.

The problem is not that the assumption was wrong. The problem is that no one put a date on the assumption. In most organizations, a capacity estimate is not revised until something visibly goes wrong: a team that is structurally behind, a budget that must be adjusted every quarter. Until that visible disruption occurs, it is assumed that the old ratio between work and people still stands.

What is changing now

AI is taking over work, and this is not happening in an evenly distributed way. Some tasks an AI system can already do independently today — a first draft of a document, a classification, a routine calculation. Other tasks are partly taken over: the system does the work, a human approves or rejects and gives a reason for that. And a third category remains human work, because judgment, negotiation or responsibility cannot be automated.

Those three categories run through every department. Within a single team, one task may already be largely done by AI while the task next to it is still entirely human work. That is why the old assumption does not collapse all at once, but loses its validity task by task. The capacity estimate made at team level does not account for that — it counts hours, not tasks.

Why it is already visible at one company and not at another

The difference does not lie in the sector, but in whether an organization has broken down its work at task level. A company that knows which part of a process consists of repeatable, well-defined steps sees fairly quickly which steps an AI system can handle and which it cannot. A company that describes work only at function level — "two customer service employees" — has no point of reference to see that shift, and therefore keeps calculating with the old ratio until the backlog or the underutilization becomes visible.

This is why the question also touches on how customers and competitors behave. If customers who want telephone contact are being served differently, or if competitors complete quote processes faster, that changes what "normal capacity" means for your own team. In this way, for example, the assumption that customers want telephone contact is also undermined by the same mechanism: not because someone decides it, but because the behavior has already shifted.

How a management team notices that the assumption has expired

The signals are rarely dramatic. They usually look like this:

a team that structurally needs more or less time than the estimate predicts, without a clear one-off cause;
lead times that decrease in areas where no one has started a project;
capacity issues being resolved with temporary hires, while no one has checked whether the underlying work has changed;
budget round after budget round in which the same ratio between work and FTE is repeated, without anyone asking whether that ratio still matches how the work is done today.

None of these signals proves anything on its own. Together they form the pattern of an assumption that is no longer tested, but repeated. That is exactly how an outdated strategy usually presents itself: not as explicit contradiction, but as a figure that keeps checking out on paper while practice deviates from it.

Here the subject also touches on the personnel domain, and there a boundary applies: which consequences an organization attaches to a changed capacity estimate falls under its own legal requirements for personnel policy and employee representation. That is up to the organization and its advisors, not to a strategic check.

What you should measure

To know whether the assumption still stands, a percentage is not needed — a few concrete observations are:

Per task, not per function. Which subtasks within a process have already been demonstrated elsewhere in the recent period to be takeable over by AI, with or without oversight?
Last review date. When was the current capacity estimate last revised, and based on which observation?
Lead time versus staffing. Is the lead time of a process changing faster than the staffing deployed on it?
External reference. Are comparable functions at competitors or in the sector now doing the work differently, and can that difference be explained by a task or by a tool?

The underlying question — which work in your company can genuinely be taken over by AI — is answered per task, not per function or department, with the work scan.