A buyer who sees a higher margin asks one question: will this recur next year, even under a different owner and a different team. With a margin that comes from AI use, that question is harder to answer than with a price increase or a purchased saving. Not because the margin is not real, but because its source behaves differently from the sources a buyer is used to weighing.
AI is already taking over work today, but in parts and not everywhere at the same pace. At one company a tool independently checks invoices from start to finish; at another company an employee does exactly the same work by hand, even though the technology exists to do it. The difference usually lies not in the sector, but in how the work has been broken down. Work that follows a fixed pattern with a clear right-or-wrong lends itself to being taken over. Work that requires judgment remains human work, or at best gets supervision: someone who approves or rejects the outcome with a reason. A margin that arises because a task has shifted into the first category has a different origin than a margin that arises because someone has started working harder with an AI tool alongside them. A buyer wants to see exactly that distinction, because it determines whether the saving is structural or hangs on one person and their habits.
The usual way to prove a margin improvement is a before-and-after comparison in the figures. That works with a price increase or a supplier deal, because the cause is stated on an invoice. With AI, the cause appears on no invoice at all. There is no contract stating which part of the profit comes from which task. What does exist: hours that are no longer needed in a place where they used to be needed. That freed-up capacity is the unit in which the substantiation must run, not the margin percentage itself. The margin percentage is the result; the freed-up hours are the evidence.
A substantiation that holds up with a buyer places three things side by side. First: exactly which task has been taken over or placed under supervision, and whether that is a task that can safely remain with AI or a task that could just as easily become human work again once circumstances change. Second: whether the freed-up capacity has actually been deployed elsewhere, reduced, or left unused — because only the first two translate into margin. Third: how dependent the saving is on one person, one supplier or one tool subscription. A saving that is embedded in the process and carried by multiple people counts more heavily than a saving that disappears the moment one key figure leaves or a license expires. That last point connects directly to the question of whether a buyer ultimately buys your people or your way of working, and those two produce a very different margin in a buyer's eyes.
An honest substantiation also states where it has no ground to stand on. Freed-up hours are easy to count at the moment a task has been taken over, but harder to predict for work that is currently still mixed — partly AI, partly supervision, partly manual work. If that mix shifts further toward AI over the next twelve months, the margin grows; if supervision remains heavily needed because the margin for error is too large, the saving stays where it is. A buyer who knows this does not ask for a guarantee but for a range with a reason attached. An outcome without that reason says nothing: a saving percentage without knowing which task lies behind it is a number that can just as easily disappear again at the next audit as it appeared.
The shift does not only affect profit quality. It also affects how heavily owner dependency weighs, because work that used to sit with the owner can now partly sit with a system — provided that system is documented and does not exist only in the owner's head. It affects management information, because a buyer wants to see whether the management information is set up professionally enough to demonstrate the saving per quarter rather than reconstruct it afterwards. And it affects the due diligence itself, because a buyer who sees AI use in place will naturally also look at the licenses and subscriptions behind it — precisely why a due diligence scrutinizes your software licenses closely. Anyone wanting part of the margin to be counted only after a transition period will also arrive at the question of what an earn-out is and why a buyer proposes one — often exactly the instrument with which a buyer makes an AI margin that is not yet proven structural financeable after all.
The underlying question — which work in this company can genuinely be taken over by AI, and which work remains human work — is answered task by task with the work scan from FTE TO AI.
This page is not about personnel decisions. What an employer does with freed-up capacity is up to the employer, and any employment-law steps are subject to their own statutory requirements, on which this page makes no statement.
Before a margin in the figures is explained to a buyer, it helps to know which of the eight value drivers is weighing most heavily on your price today. The free value check consists of eight short questions, one per driver, and immediately gives a picture of where that weight currently lies. The full value scan, with the owner-dependency index and a two-year calendar toward the exit moment, is under construction.