Capturing tribal knowledge before a veteran leaves is a ninety-day job, and the first move is to stop asking them to write anything down. Sit beside the work and write down the judgment calls only they can make, record the conditions that triggered each one, and store the record on the machine or line it belongs to, where the next person will look. Knowledge management in manufacturing goes wrong when it is run as a documentation project. It works when it is run as a capture project with a date on it. The man who ran our fastest line had been there longer than I had. He could stand twenty feet off and hear something in the drive that nobody else heard, and tell you the line would fault inside the hour unless somebody backed the speed off a couple of percent. He was right most of the time. We gave him a lunch and a card. Two weeks later that line started losing time every day and nobody could say why. He had given us three months of notice. We spent it scheduling his vacation payout and walking his replacement through the standard work, which was the one part of the job already written down. Our systems logged every fault on that line to the second and still could not tell us what the last three faults had in common. That part he carried in his head. Tribal knowledge is the earned, undocumented know-how that lives in experienced people: the early signs before a fault, the workaround that saves a run, the supplier lot to watch. That drive noise was never in the historian. It lived in one man’s ears for a couple of decades and then it left the building with him. Every plant I have walked has a version of him, and every plant is one retirement party away from losing what he knows. What knowledge management in manufacturing usually gets wrong I have done every one of these myself. We ask people to write down what they know. Ask a veteran what he knows and you get the general version of the job: the sequence, the checks, the parts of it a procedure already covers. Ask him what he did at three in the morning last Tuesday and you get the part worth writing down. We capture away from the work. A conference room and a template produce general answers. The knowledge is bound to specific conditions, and those conditions are only in the room when the machine is running. We store it where nobody looks. A binder, a shared drive, a wiki with one contributor. The next person does not go looking for a document they do not know exists. They go to the machine. Who to capture tribal knowledge from first Not everyone with grey hair, and not the org chart. The test is simple: whose absence changes what the shift does? Who gets called at home. Who shows up in the escalation chain and in no written procedure. Who is the only person who can bring a particular asset back up after a particular failure. Who the new hires ask instead of reading. Pick three to five people per plant. Rank them by the cost of losing them, not by their retirement date, and start with the one at the top. A capture program aimed at everybody captures nobody. How to capture tribal knowledge in the last ninety days This is roughly three hours a week of somebody’s time. It is the cheapest insurance in the building. Put the last day on the schedule. Ninety days, named person, named owner of the capture. Without a date it becomes a someday project and someday arrives after the retirement lunch. Work beside them, an hour at a time, three times a week. Their work, your notebook. You are not there to be trained. You are there to watch for the moments they decide something. Capture at the moment they make a call. Every time they make a call, ask three questions in this order. What did you notice? What did you think it was? What did you do? Write the answers down before the next thing happens. Capture the exceptions, not the routine. Skip anything the standard work already covers. You want the startups that go badly, the odd lot, the weekend, the machine that behaves differently in August. Swap roles at week eight. The replacement runs the job, the veteran watches and says nothing unless equipment or a person is about to get hurt. Every correction the veteran makes is a capture you missed. Ask the two exit questions in week twelve. What will break after you go, and who will they call? And: what do you do that nobody has ever asked you about? The second one always produces something. Put one captured item to use on the floor inside the first week. If the first thing you capture never comes back as a change anybody notices, capture stops. People give knowledge to a system that visibly uses it. What to write down for each judgment call Four fields, every time. Conditions. Early sign. Action. Outcome. Line 2. Conditions: room over thirty degrees and running the heavy grade. Early sign: drive whine climbs about half an hour before the fault. Action: back line speed down two percent until the room cools. Outcome: no stop. That is one sentence of writing and it is worth more than a forty-page transfer document, because every part of it names something the plant already tracks. Attach every capture to the thing it belongs to: the machine, the line, the part number, the supplier, the lot. Not to the person’s name and not to the date of the interview. The next operator will not search for a story. They will be standing at an asset with a problem, and the capture has to be sitting on that asset when they get there. Keep it inside the system the work is already recorded in. A separate manufacturing knowledge management system is a second place to look, and the second place is the one nobody checks. Every plant has a retirement date on the calendar that nobody is planning around. Somebody’s last day is already booked, and the standard work will get handed over cleanly, and the line will keep running. It will just run a little worse every month, and no one will be able to point at the week it started. The ninety days before that date are the only chance you get, and they are already counting down. This week: Pick the one person whose retirement would cost you most, and spend an hour beside them this week writing down every judgment call they make and what they noticed first.