A buyer does not settle on the number of employees there are now. He settles on what keeps running when the owner leaves, when a key figure steps down, or when a vendor changes the terms. A saving that appears on today's profit and loss account because AI has taken over part of the work is not automatically value in that calculation. It is value if the process itself carries the saving. It is risk if one vendor carries the saving.
That distinction may seem technical, but it directly determines the price. Profit that depends on an external party is discounted more heavily by a buyer than profit that is embedded in the company's own systems and procedures. Not because the saving is not real, but because its continuity does not lie in the company's hands.
Work that is now done or supported by AI falls in practice into three categories, and these categories cut across every department:
The first category delivers the largest saving, but also the largest question: who maintains that system, and what happens if that party stops, raises the price, or does not renew the contract. That question is exactly what why a buyer asks who maintains the ai addresses. The second category is often the most solid, because the human in the process safeguards quality and does not make the organisation fully dependent on an external party. The third category generally does not change the valuation, but does change the costs associated with it.
In one company, a large part of the administrative processing has already been taken over by a system that runs on a contract with a single party. In another company, comparable work is done with tooling that is replaceable, whose logic has been documented, and whose operation is understood by several employees. The difference does not lie in the sector and not in the size. It lies in whether the company has documented what the system does, why, and what happens if it falls away. Companies that started doing this early can show at a sale that the saving is structural. Companies that did not have to explain it at the negotiating table, with less evidence and less time.
Dependence on a single AI vendor is, first and foremost, a question about risk, and that question affects several drivers at once. It affects profit quality, because a saving that hangs on a cancellable contract is valued differently than a saving embedded in the process, as described in how do you substantiate profit quality. It affects recurring revenue, when customers take services that are themselves built on AI processing, explained in what is recurring revenue worth at a sale. And it affects working capital, because some vendor contracts involve prepayment or long-term obligations, a subject addressed in what is working capital and why does it count at a sale. The extent to which a single vendor depresses the price further depends on how replaceable that vendor is, as described in which dependence on an ai vendor depresses the price.
An estimate of this dependence is an estimate, not a measurement with a fixed outcome. It is based on what is documented today about contracts, systems and processes, and on how replaceable those components are. Where that documentation is missing, the outcome is less sharp: the evidence simply is not there, and that is itself a signal, not a flaw in the method. A score also says little when the company has just implemented a major change whose effects are not yet visible in the figures, or when a sector changes so quickly that today's estimate will be outdated within six months. In those cases, the outcome is a snapshot, not a forecast for the coming years.
The method also does not predict which saving will persist and which will disappear. It shows where the risk lies, not whether it will materialise. And it makes no statement about staffing: if a shift in work has consequences for staffing levels, its own statutory requirements apply, separate from this scan.
The first step to reduce this uncertainty is to document what is already happening now: which work a system does, which work proceeds with oversight, and which work remains human work. This is described in how do you document what ai does in your company. Without that documentation, any valuation of the saving is an assumption, with or without AI in the story.
The underlying question — which work in this company can genuinely be taken over by AI, and which part of that depends on an external party — is mapped per task using the work scan from FTE TO AI.
Anyone who first wants to know which of the eight value drivers is depressing the price the most today can fill in the free value check: eight short questions, one per driver, with an initial picture of where the greatest pressure lies. The full value scan, with evidence per driver and a two-year calendar towards the exit moment, is under construction.