A management team asking how it records what AI does is actually asking a different question: what remains if certain people were to leave tomorrow. A buyer asks that same question too, only about the entire company. He is not buying the number of employees on the payroll, he is buying what continues to run without them. Recording what AI does is therefore not a technical overview of tools and licenses, it is a description of which work hangs on a system and which work hangs on a person.
The mistake most overviews make is dividing work into two bins: automated or not. That misses the middle section where most of the movement is happening right now. There are tasks that AI delivers independently, tasks where AI makes a proposal and a human approves or rejects it with a reason, and tasks that remain human work because judgment, relationship or responsibility cannot be delegated. That middle category is unstable: the oversight still needed today may be lighter in a year, or may just as well remain just as heavy. Anyone who only registers what already runs fully automatically today misses where the movement is.
A job title says little about what AI can take over. A controller does reporting work that is largely automatable and judgment work that is not, often within the same working week. Recording what AI does therefore means descending to the level of tasks: which action, with which input, with which decision at the end. At role level a distorted picture emerges that a buyer will see through within a day during due diligence.
Every estimate of what AI can take over rests on assumptions about the quality of the data, the stability of the process and an organization's willingness to let a system make the decision. For tasks with many exceptions, varying input or sensitive judgments, the range is wide and the outcome is less a fact than a direction. An overview that leaves no room for that, and presents everything in fixed percentages, deserves less trust than an overview that indicates where the margin is large. That also applies to the effect on price: which dependency on an AI supplier lowers the price cannot be captured in a single figure, because it depends on how much of the taken-over work hangs on one external party versus being anchored in the company's own process.
There are situations in which an overview of what AI does has little value for a buyer. If the takeover of tasks rests entirely on the knowledge of one person who set up the system and no one else can explain why it works, the label "automated" says nothing about risk. That is the same question as with human work: why a buyer asks who maintains the AI touches exactly this point. A task that is handled today by a system but can only be restored by the owner in case of a malfunction counts differently for a buyer than a task that is understood and maintained by a team.
Some companies have already recorded this, others not at all, and the difference rarely lies in the sector or the size. It lies in who wrote down the tasks before AI came into view. A company with documented processes can relatively quickly point out which part of that can be transferred to a system. A company where the work sits in people's heads must first take that step before AI can take anything over there. That is immediately also a value question: how you measure how much runs through the owner explains why companies with documented knowledge can demonstrate faster what AI does than companies where that knowledge was never written down.
A saving resulting from AI does not automatically count for a buyer as profit of good quality. A saving tied to one contract with one supplier is more fragile than a saving anchored in the process itself that keeps working without that supplier. The effect on working capital is also not automatically favorable: faster processing of invoices or orders can shorten turnaround times, but what working capital is and why it counts in a sale shows that the relationship between automation and cash position is not straightforward.
What a company subsequently does with the overview of taken-over, partly oversight-requiring or remaining human work is a decision for the employer. Decisions affecting personnel are subject to their own legal requirements, and this overview is not a basis for that. The goal is a factual picture: which work hangs on whom or what, and what does that mean for what remains if a person leaves. That is also where spreading of knowledge plays a role: how you prevent your value from sitting in a handful of heads touches the same question from the people rather than the systems.
The moment when this overview yields the most is not just before a sale but well before it, when there is still room to do something about the dependencies before a buyer encounters them. When you should start preparing for a sale partly depends on how much of the current work still sits with people who are not staying.
The underlying question, which work in this specific company can genuinely be taken over by AI, is answered per task in the work scan of FTE TO AI. For those who first want to know where the price is under the most pressure today, there is the free value check: eight short questions, one per value driver, with a picture of which driver weighs most heavily today. The full value scan, with maturity scores, evidence per driver and a two-year calendar toward the exit moment, is under construction.