A management team that books a saving from AI usually gets the same question back from a buyer: how do you know that, and will it stay that way. That is not distrust. A saving that has not been substantiated is treated at valuation as noise, not as profit. The question of how you substantiate it is therefore not a bookkeeping matter, but a question about evidence.
AI is taking over work, not all at once and not evenly everywhere. In some companies part of the customer communication or the initial review of documents already runs without human hands, with someone approving or rejecting the outcome. In other companies that is still entirely human work, not because it could not be done otherwise, but because no one has recorded which tasks lend themselves to it. The difference between those two situations rarely lies in the technology. It lies in whether someone has broken the tasks down and looked at each one: can AI take this over, can it be done with oversight, or does it remain human work. Without that breakdown, a saving is a feeling. With that breakdown, it is a claim that can be checked.
A statement such as "we have freed up one FTE" cannot be tested by a buyer. An FTE consists of dozens of tasks, and those tasks do not all shift equally hard or equally certainly. Substantiation begins with unbundling the role into tasks, with each task classified into the three categories that recur everywhere: AI can take it over, AI can do it with human oversight, or it remains human work. Only at that level can you say which part of the hours has been structurally freed up and which part depends on oversight that could be tightened or loosened tomorrow. How you record this per task, including the evidence a buyer expects with it, is covered in how you record what AI does in your company.
No scan, no sharp calculation method, no assessment can determine with certainty how much capacity an AI application will still free up in two years. The uncertainty sits in three places. First, the assumption about oversight: if a task is currently done with human oversight, the question is how much of that oversight will still be needed once the system has been running longer, and that is an estimate, not a fact. Second, the dependency on the supplier: a saving that rests entirely on the price, terms, or continued existence of one AI supplier is less solid than a saving that is anchored in the company's own process, and that distinction weighs more heavily with a buyer than the size of the amount. Third, the time horizon: what today runs with oversight may run more autonomously in a year, or may not, if the task turns out to be more complex than expected. A substantiation that names these three points is more credible than one that presents a single percentage without saying where it comes from.
A buyer accepts a range with reasoning behind it. Not: thirty percent of the role has been taken over. But rather: the largest part of the repetitive document review has been taken over, with an employee assessing the exceptions, and that ratio has remained stable over the past six months. That is a statement that can be verified using logs, error rates, and the time the supervising employee still spends on it. It makes the saving part of profit quality instead of an assumption that falls away during due diligence. Which dependencies push down the price in this respect, even when the saving itself is correct, is set out at which dependency on an AI supplier pushes down the price.
If the freed-up work relies mainly on knowledge that only the owner has about how the system is set up, the saving is not transferable. That does not only reduce the saving itself, but also the broader owner dependency index, and that often weighs more heavily in the valuation than the amount of the saving. A buyer therefore asks not only what has been saved, but who maintains it once the owner is gone, a question explored further in why a buyer asks who maintains the AI.
This is not personnel advice and not a substantiation for a dismissal decision. Whether and how an employer makes personnel decisions based on changing work falls under its own statutory requirements, which are not addressed here. What is described here is solely how a saving that has already been realized can be made demonstrable and transferable towards a buyer.
The underlying question, which work in this company can truly be taken over by AI and which part remains human work, is answered per task with the work scan from FTE TO AI. For those who first want to know where the substantiation is weakest in their own company, there is the free value check: eight short questions, one per value driver, giving a picture of which driver is pushing down the price the most today. The full value scan, with evidence per driver and a two-year calendar towards the exit moment, is under construction. What this specifically affects within the eight drivers is explained at what is reducing the value of my company, and those who also want to see the saving reflected in the reporting itself will find the approach at how you professionalize management information.