The Most Valuable System in Your Plant Is About to Retire

The Asset That Never Appears on the Balance Sheet

There is a person in your plant who knows that the number two line runs hot in August and needs the feed rate backed off before anyone sees a defect. Who can tell from the sound of a spindle that a bearing has about two weeks left. Who remembers why that one customer's parts get inspected differently, and which supplier substitution caused the problem in 2014 that nobody wants to repeat.

None of that is written down. It is not in the ERP, not in the work instructions, not in the maintenance system. It exists in one head, it took twenty-five years to build, and it is walking out the door on a Friday afternoon with a cake in the breakroom.

Every manufacturer knows this is happening, and almost none of them treat it as the uninsured loss of an asset.

Every manufacturer knows this is happening, and almost none of them treat it as the uninsured loss of an asset. That is exactly what it is: the most valuable operating knowledge in the building, walking out on a schedule you can predict years in advance and almost never act on in time. The knowledge is real, the departure date is knowable, and the window to do something about it closes quietly.

The Numbers Behind the Cliff

This is not a soft problem, and the scale of it is well documented.

Deloitte and The Manufacturing Institute project that United States manufacturing could need as many as 3.8 million new employees by 2033, and that as many as 1.9 million of those roles — roughly five in ten of the skilled openings — could go unfilled if the skills and applicant gaps are not closed. In the same research, 65 percent of manufacturers named attracting and retaining talent as their primary business challenge, ahead of every other concern on the list.

The maintenance function is where the cliff is steepest. Industry reporting in 2026 puts 42 percent of maintenance technicians over the age of 55, with retirement rates accelerating. These are the people whose knowledge is least documented and most operationally load-bearing, because diagnostic skill is exactly the kind of expertise that resists being written into a procedure.

And the replacement math is worse than a headcount gap suggests. Reported figures across the industry put the average lag at roughly eighteen months between a senior technician retiring and a replacement reaching equivalent productivity. That is not eighteen months of a vacant seat. It is eighteen months of a filled seat producing at a fraction of the output, while the failures that the departed technician would have caught early become unplanned downtime instead.

Deloitte's 2026 outlook adds a further wrinkle worth naming plainly: immigrant workers filled nearly one in four United States manufacturing production jobs in 2024. Whatever your view of the policy environment, the labor supply assumptions that many workforce plans were built on are less stable than they were.

Why Documentation Projects Always Fail

Almost every manufacturer has tried to solve this, and the attempt usually looks the same. Someone is assigned to document the tribal knowledge. Templates go out. A binder or a wiki gets created. Six months later it is stale, and eighteen months later nobody opens it.

The failure is structural, not a matter of effort. Asking an expert to write down what they know requires them to first become aware of what they know, which is the hardest part. Genuine expertise is compiled into intuition. The technician does not experience themselves as running a diagnostic decision tree; they experience themselves as knowing. Ask them to document it and you get the procedure they were taught, not the judgment they actually use.

It also competes directly with the day job. The person with the most knowledge is by definition the person most in demand on the floor, and documentation loses that competition every single time.

And the output degrades on contact with reality. A written procedure captures one state of one machine at one moment. The plant changes, the procedure does not, and the first time someone follows it and it is wrong, the document loses its authority permanently.

What Changed: Capture Became Passive

The reason this is worth revisiting in 2026 rather than filing under permanent problems is that the capture mechanism changed. It stopped requiring the expert to do the work of writing.

Recorded conversation is the clearest example. Experienced technicians will readily explain a failure signature, an operating nuance, or a diagnostic shortcut in ordinary conversation while looking at the machine — the same content they cannot produce in a structured form. Run those transcripts through a language model and you get the beginnings of a troubleshooting guide, built from how the expert actually thinks rather than how a template asked them to think.

The second shift is that the record of expertise has been accumulating in your systems all along without anyone treating it as knowledge. Historical work orders, technician notes, failure codes, repair histories, scrap and rework records, quality dispositions, and production parameters are a decades-long log of what went wrong and what fixed it. That corpus is now readable at scale. Patterns that no individual could hold across ten years of records — this failure follows that symptom, this part number correlates with that defect, this line drifts under these conditions — can be surfaced and turned into procedures, checklists, and maintenance intervals.

The third shift is delivery. Captured knowledge that lives in a document nobody opens has not been captured in any useful sense. Knowledge that surfaces at the moment of the task, in the work order, on a tablet at the machine, answering the question the technician actually has, gets used. Vendors in this space report new-technician time to productivity falling to roughly six to eight months against the eighteen-month traditional path. Treat those figures as directional rather than audited, since they come from the companies selling the tooling, but the direction matches what well-documented plants have always achieved.

The Knowledge Is Already in Your ERP

This is where the workforce problem and the systems problem turn out to be the same problem.

If your ERP, MES, or maintenance system has been running for a decade, you are sitting on the raw material. Every work order closed with a note, every nonconformance with a disposition, every unplanned stop with a cause code, every routing that got revised and why. Most manufacturers regard this as transactional exhaust — data that was captured to close a record, not to teach anyone anything.

Treating it as a knowledge base rather than a filing cabinet is a genuinely different posture, and it does not require replacing anything. It requires getting at the data, which for most manufacturers is the actual obstacle. The information is there and it is effectively unreachable, spread across tables designed for transactions, not for questions.

There is an honest caveat, and it is the same one that applies to every AI initiative on top of an operational system. If your cause codes are used inconsistently, your work order notes are blank half the time, and the real reason a line went down gets recorded as other, then the corpus will teach you very little. Data quality is not a prerequisite you can skip. It is worth an honest assessment before anyone builds anything, because the assessment usually reveals both more usable history than expected and specific habits worth fixing now so that the next five years of records are better than the last five.

What This Looks Like in Practice

The version of this that works is narrow, and it starts with a person, not a platform.

Identify who is retiring in the next twenty-four months and what specifically they know that nobody else does. Not a general knowledge audit — a named list of people and the failure modes, machines, customers, and processes that only they can currently handle. That list is usually shorter than feared and more concentrated than anyone expects, and it makes the problem finite.

Then capture in the format that fits the knowledge. Walk the floor with them and record it. Interview them about the last twenty things that went wrong. Pair them with their successor on live problems and record those sessions. The goal is not a document. It is a corpus.

In parallel, mine what the systems already hold for the same equipment and processes, so that the interview fills the gaps in the data rather than duplicating what is already recorded.

Then put the result where the work happens. Attached to the asset. Surfaced in the work order. Answerable in plain language by someone standing in front of the machine with a problem and no time to search a wiki.

The Part That Is Not a Technology Problem

One thing will sink this faster than any technical obstacle, and it deserves to be said directly.

If the people whose knowledge you are capturing believe the purpose is to make them replaceable, you will get compliance and not candor. They will answer the questions asked and volunteer nothing, and the volunteered material is the whole value.

The framing that works is the one that happens to be true: this is about the expert's legacy and their leverage, not their redundancy. The technician who has spent thirty years learning a plant generally wants that plant to run well after they leave, and wants to stop being the only person who can be called at 2 a.m. Position the effort as building the thing that lets them retire without guilt, and give the senior people visible ownership of the output — their name on it, their corrections, their authority over what is right.

This is ordinary change management, and it is the difference between a knowledge program that produces something useful and one that produces a folder of transcripts nobody trusts. Trust determines what gets said out loud, and what gets said out loud is the entire asset.

The Bottom Line

The workforce cliff facing manufacturing is usually discussed as a hiring problem, which makes it feel like someone else's department and a decade away. The more urgent framing is that a specific, knowable set of people in your plant hold operating knowledge that has never been recorded, and their departure dates are already on a calendar.

The tooling to capture that knowledge without asking experts to become technical writers now exists and is accessible to small and mid-market manufacturers, not just to companies with research budgets. The history that supports it is already sitting in systems you own. What the moment requires is a decision to treat it as an asset with an expiration date, and to start with the people closest to leaving.

At Cherry Street, this is the work we do. We help manufacturers find the knowledge that is genuinely at risk, get at the operational history already buried in their ERP and maintenance systems, build the capture and delivery that puts answers where the work happens, and run it in a way that the people whose expertise it is will actually support. If your succession plan for the plant currently depends on one person not retiring, that is worth addressing while they are still here to help.