Every company with a technical product or process has someone who knows the answer before the question is finished. Often the founder, often the first technical specialist who was hired. Over the years, that person has made thousands of small decisions about how something works, why a solution does or doesn't function, which deviation is normal and which is not. That was never recorded because there was never time for it and because asking questions was faster than documenting.
That concentration did not arise from an unwillingness to share knowledge. It arose because it works faster: one person who has the answer ready is more efficient than a process in which three people have to consult a document. For day-to-day operations, that is an advantage. For a buyer, it is a risk, because a buyer does not pay for the people who are there, but for what keeps working when one of those people leaves.
In an acquisition, technical knowledge concentration becomes visible as soon as someone asks: what happens if this person is absent for six weeks. If the answer is that the quality of decisions then drops, or that customers with specific questions have to wait, then that person counts not as an asset but as a risk on the balance sheet. That risk is often translated into a lower price or into a longer period during which the seller has to stay on, which pushes the exit moment further away than the owner wanted.
This is not a personnel question and not about steering who does or doesn't stay: that is up to the owner, and if it touches on employment-law choices, those are governed by their own legal requirements. What matters here is something else: is the knowledge that currently sits with one person also accessible without consulting that person.
Technical knowledge consists of three kinds of work, and AI does not affect them all equally.
One part is pattern recognition: which fault belongs to which cause, which specification fits which application. With the right documentation, that is the work AI takes over best, because it can be traced back to comparable cases from the past.
A second part is judgment: an AI system can make a proposal, but someone with experience approves or rejects it, and can say why. That is the part requiring human oversight, and it remains necessary as long as the underlying situations differ enough from each other to make a fixed rule unusable.
A third part is explaining things to a customer who is not technical themselves, with a feel for what that customer already understands and what not. In most companies that remains human work, although here too the share shifts once the first two layers have been documented, because that leaves more time for the conversation itself.
Companies where this already works are the companies where someone once invested in recording the first layer: fault history, error codes, solutions, in a form a system can search. Companies where this does not yet work are not necessarily behind; they simply never had a reason to document something that was already readily available in one head.
Recording the knowledge is not a matter of writing one document and being done. It works better step by step: start with the questions that come back most often, record what the answer was and why, and let an AI system make those answers searchable for the rest of the team. The specialist remains necessary for the cases that deviate, but becomes less the sole point of entry for everything.
This affects more than just the owner-dependence index. Savings that arise because one person spends less time on repeat questions count differently for a buyer than savings that depend on a single external supplier: profit quality is about where a result comes from, not only about its size. Knowledge embedded in the process retains value even if the person leaves; knowledge that hangs on a single supplier or specialist disappears with that person.
The same logic applies to other parts of the company. Anyone who wants to know how customer relationships can become less dependent on one salesperson should read how the network in the market is made transferable; anyone wondering whether frontline customer contact can also run without the owner will find that in how customer contact becomes transferable with AI; and anyone who wants to see how consistent quality assessment becomes less dependent on a single assessor should read how quality control is made transferable. Anyone who wants to know how big this problem already is in their own company can measure how much runs through the owner, and anyone wondering from what point this becomes relevant for a sale should read when preparation for a sale should begin.
The question of which part of the technical knowledge in this specific company is genuinely transferable to a system, and which part remains human work, differs by process and by customer group. That question is answered per task with the work scan from FTE TO AI.
The free value check consists of eight short questions, one per value driver, and gives a first picture of which driver is putting the most pressure on your company's price today. The full value scan, with a maturity score per driver, evidence, an owner-dependence index, and a two-year calendar toward the exit moment, is under construction.