You can’t automate what was never written down

More than 90 percent of all job openings in Europe are driven by replacement needs. Not new roles, not growth, not expansion. Simply filling the gap left by someone who walked out the door. That figure comes from Cedefop, and it counts both the people who retire and the people who move on to something else.

I read that as a labor market number the first time. Lately I have started reading it as a supply chain number, because in every plant I visit there is someone who knows things that are not written down anywhere. And in the next five years, a lot of those people are leaving.

Every plant runs two systems

The first one is the one you paid for. ERP, item master, approved vendor list, contract database. It tells you what a part costs, who supplies it, and what the lead time was the last time anyone checked. That information transfers. It is often a mess to clean up, but once it is clean it can be handed to anyone.

The second system has no license fee and no implementation project. It lives in the head of the person who has been buying your non-strategic materials for twenty years, supported by a personal spreadsheet and a phone list that exists nowhere in your IT landscape. It knows which supplier says yes to everything and delivers late anyway. It knows that the quoted lead time on a particular item is fiction in Q3, because that supplier always has holiday shutdowns. It knows that another item looks like a commodity in the system but is single-sourced in practice, because the alternative was qualified once in 2019 and never worked properly.

That is not information. It is judgment, and judgment is the part you cannot export. It builds up over years, and it transfers through conversation, through overlap, through someone sitting next to someone else for long enough.

When that person retires or takes another job, the first system stays exactly where it is. The second one leaves the building.

And nobody notices for a while. Tail supply has a way of looking solved right up until it isn’t. The orders keep flowing, the lines keep running, and the gap only becomes visible when something goes wrong and there is no longer anyone who knows why it used to go so smoothly.

Meanwhile, the overlap that transfer depends on is getting thinner. Cedefop’s latest forecast projects the labor force that is aged 65 and over, will be growing by more than 7 percent per year through 2035, while the 30 to 54 cohort, the group you would normally promote from, shrinks slightly.

What automates and what does not

To be clear, this is not an argument against technology.

Agentic AI is being positioned as the natural successor to the transactional buyer, and for part of the work that is correct. Where the knowledge has already been made explicit, in item data, contracts, agreed pricing and order history, the sourcing side automates well. That is not a theoretical claim for us. It is what we do: make the knowledge explicit first, then standardize, then automate.

The trouble starts with everything that sits deeper in the chain. Aligning stock positions with the real production schedule. Understanding when materials are actually consumed on the floor rather than when they were booked. Quality issues that only surface at assembly. Those are not transactions you codify. They are processes you improve, with people who know the factory. Low value does not mean simple, it only means nobody has looked closely.

And there is a reason so little of it is on paper. Among people over 55 in technical roles, only 30 percent hold a theoretical education degree, against 46 percent of those under 35. Fewer of those over 55 went through formal schooling, more of them learned the job by doing it. And that distinction matters, because practice does not get written down the way a curriculum does. Nobody hands you a textbook about the years spent finding out which supplier actually delivers on time.

Software codifies what has already been captured. If the knowledge lives in someone’s head, automating the process around it gives you a faster version of an empty process.

Every organization deals with staff turnover, and that is the fair objection to all of this. What makes this moment different is the timing. We are being asked to automate the tail at exactly the point where we are losing the people who could tell us what the tail actually does. If you digitize a process before you understand it, you have not preserved the knowledge. You have made its absence permanent.

What this asks of you

Treat tail supply as a spend category with its own process, not as the residue left over once the strategic categories have been handled. A category with no owner tends to produce no documentation either.

Assume the handover has not happened. In my experience the formal handover document, when it exists, captures the first system and misses the second one entirely. The useful version of this exercise is not asking people to write down what they know. It is sitting with them for a month and letting them ask you why.

And be honest about whether this will ever win internally. Tail supply competes with the components that stop the line when they fail, and it loses every time, in every company I have seen. Which is why for many manufacturers the realistic answer is to place it with someone who treats it as a core activity rather than an overflow task.

The window is open now

The people who know how your tail supply really works are probably still in the building, and willing to explain it. In my experience they are slightly relieved when someone finally asks.

That window closes quietly. There is no alert, no dashboard, no exception report that tells you institutional knowledge just left at 17:00 on a Friday.

So the question I would put to any manufacturer reading this is not whether your systems are ready for automation. It is a simpler one: if the person who manages your tail supply resigned tomorrow, how much of what they know would you still have on Monday?