Morsa publishes this guide. The plant, the people and the setup in the examples and the demo are a made-up sample. Workforce numbers come from the U.S. Bureau of Labor Statistics and from Deloitte with The Manufacturing Institute, linked where they are used.
Quick answer
Meaning: Unwritten know-how held by one person or a small group, learned by doing and passed on by talking.
Other names: Institutional knowledge, tacit knowledge, undocumented knowledge, know-how, the unwritten rules.
On a plant floor: Machine quirks, the insert that actually works, what an alarm really means, which supplier’s ship dates to trust.
Why it is urgent: About 1 in 4 people working in U.S. manufacturing is 55 or older (BLS, 2025).
How to capture it: Find the jobs only one person can run, record that person at the machine, write it into the setup sheet, and make it answerable on every shift and in every language.
The modern way: An AI system like Morsa answers the new hire’s question from the setup sheet and the expert’s own words, in the new hire’s language, and learns from every answer your people give.
What is tribal knowledge?
Tribal knowledge is information that people inside a group know and use every day, but that is not written down anywhere the rest of the organization can find it. It lives in memory, gets taught by example, and survives only as long as the people who hold it stay. Wikipedia describes it as institutional memory that lacks adequate documentation, kept alive only by mentoring, on-the-job training and continuity of staffing.
The word “tribal” comes from an analogy with cultural anthropology: a tribe shares knowledge that outsiders do not have. In companies, the “tribe” is the handful of people who have done a job long enough to know how it really works.
Two things make it different from ordinary documentation. First, it is usually right, because it was learned the hard way. Second, it is sometimes wrong, because nobody ever checked it with data. Both are reasons to capture it.
Tribal knowledge vs tacit, explicit and institutional knowledge
Term | What it means | Plant example |
|---|---|---|
Explicit knowledge | Written, findable, the same for everyone | The setup sheet: program O2001, a 2.0 in collet, coolant at 8 to 10 percent |
Tacit knowledge | Skill a person has but finds hard to put into words | Hearing that a cut is about to chatter |
Tribal knowledge | Know-how a group could write down but has not | Keep the roughing feed at 0.010 on 316 or the chips wrap on the bar feeder |
Institutional knowledge | Everything an organization knows, written or not; often used as a neutral synonym | Why the plant stopped buying from a supplier in 2019 |
Tribal knowledge overlaps with tacit knowledge, but most of it can be written down once someone asks the right question.
What tribal knowledge looks like on a plant floor
Take the sample plant in this guide: a 120-person machine shop that makes valve and pump parts. Lathe 6 runs the CVB-200 check valve body out of 316 stainless bar. The setup sheet has six steps. Dan, the senior setup operator, has run the job for years, and what he tells every new hire is not on it.
On the setup sheet | What Dan says (not on the sheet) |
|---|---|
Tools: T1 CNMG 432 roughing, T2 VNMG 331 finishing, T3 drill 0.750, T5 parting blade. | On 316 the chips wrap on the bar feeder if you run the roughing feed above 0.012. Keep it at 0.010 and use the chip breaker insert, the MR geometry, not the general one. |
First piece: check bore 0.750 +0.002/-0.000 with the bore gauge, face to shoulder 1.250 +/- 0.005, record on the first-piece sheet. Supervisor signs before the run. | Before the first piece, warm the spindle 10 minutes. Cold, the bore comes out 0.001 small. |
Step 3 lists T5, the parting blade, and nothing about how long it lasts. | Check the parting blade every 50 pieces. On 316 it chips around piece 60 and then the part drops into the chip conveyor. |
Nothing about alarms. | If alarm 1023 (bar feeder fault) comes up, it is almost always chips around the feeder push rod. Clear them, do not reset more than twice; the third time call maintenance. |
Nothing about spares. | Spare MR inserts are in the tool room, drawer 14. Kevin keeps them. |
The sample plant’s CVB-200 Op 10 setup sheet next to Dan’s notes. Every line of Dan’s is a scrap, downtime or delay event waiting for the day he is not there.
None of Dan’s notes are secret or hard to write down. They are tribal because nobody has asked him to, and because his way of passing them on is standing next to a new hire for a week.
Examples of tribal knowledge in manufacturing
Setup and changeover: the order that actually works, the shim a fixture needs, the offset that drifts after lunch.
Machine quirks: warm-up times, the axis that needs to be homed twice, the machine that only holds tolerance on one side of the table.
Tooling and materials: which insert grade lasts on which alloy, when a blade chips, which bar supplier’s stock runs hard.
Alarms and faults: what an alarm code usually means on this machine, what to clear, and when to stop resetting and call maintenance.
Quality: the dimension that drifts first, the gauge that reads high, the customer who checks a feature the drawing does not call out.
Maintenance: the sound a pump makes before it fails, the part that is always on the shelf of the shop down the road.
Suppliers and customers: who to call when a delivery slips, which buyer accepts a partial shipment, which ship date to trust.
Scheduling: which jobs can share a setup, which machine can take the overflow, which order really has to ship Friday.
Try it: ask the sample plant
This is what captured tribal knowledge looks like when a new operator can simply ask for it. You are Luis, who started last week on second shift and reads Spanish first. Ask how to run the CVB-200 on Lathe 6, in English or Spanish, or tap an example. The answer comes back in your language, and every sentence shows whether it came from the setup sheet or from what Dan said. Dan gets the same answer in English, so he knows what Luis was told.
Notice two things. When the sheet and Dan’s notes cover the question, the answer cites the exact line, and every number in it appears in that line. When they do not, as with the spindle speed example, the answer says so and names who to ask instead of guessing. That is the bar any system that answers shop-floor questions should meet.
Tribal knowledge synonyms, and is the term OK to use?
Common synonyms for tribal knowledge are institutional knowledge, tacit knowledge, undocumented knowledge, know-how, legacy knowledge, collective wisdom, the unwritten rules and, on the floor, “how we really do it”. Institutional knowledge is the closest neutral replacement. Tacit knowledge is narrower: it means skill that is hard to put into words at all.
Some people avoid “tribal knowledge” because of how the word “tribe” lands for Indigenous communities, and the question comes up often, for example in this Reddit thread. It is still the most established term for the idea, which is why this guide uses it. If your team prefers “institutional knowledge” or “undocumented know-how”, use that: what matters is that the know-how gets written down, not what it is called.
Why tribal knowledge is a risk in manufacturing
The people who hold it are getting older. In 2025, 3.82 million of the 15.13 million people employed in U.S. manufacturing were 55 or older, about 1 in 4, and the industry’s median age was 43.9 years against 42.1 for all industries (BLS, Current Population Survey table 18b, 2025 annual averages). Many of the people who know why a machine behaves the way it does are within a few years of retiring.
And they are hard to replace. Deloitte and The Manufacturing Institute estimate that U.S. manufacturing could need around 3.8 million new employees between 2024 and 2033, and that around 1.9 million of those jobs could go unfilled (2024 talent study). The new people who do arrive need to get productive faster than shadowing allows.
When tribal knowledge walks out, it does not show up as one big failure. It shows up as:
Slower setups while someone relearns what the expert knew.
More scrap and rework, usually on the jobs that always ran fine.
Downtime at night, when the one person who knows is asleep. See the machine downtime log to see where it lands.
Quality escapes, because the extra check the expert did by habit stops happening.
A single point of failure: software teams call it the bus factor, the number of people who would have to leave before work stops.
Wrong habits spreading, because unwritten shortcuts are never tested against data.
Self-check: how much walks out when Dan retires?
Think of the most experienced person on your floor. Tick each statement that is true at your plant today. The score runs from 0 to 13; 3 statements count double because they are the ones that stop a job outright. You get the first three things to capture, in order.
How much walks out when Dan retires?
Think of the most experienced person on your floor. Tick every statement that is true at your plant today.
Low: most of it stays. Most of what your experienced people know is written down or shared. Keep it that way: update the sheets every time a fix is found.
Runs in your browser. Nothing you tick is stored or sent.
A score of 0 to 3 is low, 4 to 7 is medium and 8 or more is high. Most plants with a few long-tenured setup people and no skills matrix land in medium or high.
How to capture tribal knowledge in manufacturing
Find the single-person jobs. Build a skills matrix: jobs or machines down the side, people across the top, a mark where someone can run it alone. Every row with one mark is tribal knowledge at risk.
Rank them by risk. Put first the jobs where the expert retires soonest and where a stop costs the most: a bottleneck machine, a big customer’s part, a safety-critical step.
Capture it at the machine, in their words. Record the expert talking through a real setup. Then ask the question that pulls out the tribal part: “What would you warn a new hire about on this job?” Dan’s answer is the five notes above.
Check it before you write it down. Test each tip against data you already have: first-piece records, scrap logs, downtime by alarm code. Most tips hold up. The ones that do not are the habits you want to stop spreading.
Write it where the work is. Add it to the setup sheet or work instruction for that job, with a short “what goes wrong” box. A work instruction template gives you the format; it doubles as a manufacturing SOP template.
Make it answerable on every shift and in every language. A binder in the supervisor’s office does not help second shift at 10 PM. Put the answers where people already are, in the languages your operators read first, and hand over open issues at shift change with a shift handover template.
Keep it alive. Close every fix by updating the sheet it came from. Use root cause analysis on repeat failures, and fold machine know-how into total productive maintenance routines so it is checked, not just remembered.
Ways to capture it, compared
Method | Good at | Falls short when |
|---|---|---|
Shadowing and mentoring | Tacit skill, judgment, trust | The mentor is out, retires, or teaches a different version to each person |
Written setup sheets and SOPs | One version, auditable, cheap | Nobody updates them, or the person on second shift cannot read the language they are in |
Video work instructions | Showing hand skills and sequence | Someone has to find the right clip at the machine, and it goes stale quietly |
Training platforms and connected-worker apps | Assigning, tracking and signing off training | The question is not in the course: a new alarm, a new material, a one-off problem |
An AI system that answers from your documents and your people | Answering the exact question, on any shift, in the asker’s language, with the source | It is not given the sheets and notes to answer from, or nobody checks what it learns |
Most plants need a mix: written sheets as the record, people to teach the hands-on part, and a way to get an answer at 10 PM.
Overall equipment effectiveness is a good way to see the payoff: faster setups and fewer stops show up directly in OEE.
Morsa is the person who has been there ten years
The old way to keep tribal knowledge is a binder that nobody updates and a phone number for Dan. The modern way is to let an AI system like Morsa do the remembering, the answering and the updating, while your people approve what it learns.
A new operator asks, in their own language, on the channel they already use. Luis asks in Spanish over WhatsApp, Teams, text or email how to set up Lathe 6 for the check valve body. There is no new app to learn.
The answer comes from the setup sheet and what Dan said. Morsa answers in Spanish from your setup sheets, work instructions and the messages of the people who know the job, and shows where each line came from.
Dan gets the English copy. Dan sees exactly what Luis was told, so he can correct it in one reply.
It learns from every answer people give. When Morsa does not know, it asks the right person, and that answer is kept for the next person who asks. Changes to what it knows are shown to your people first and can be undone.
It works with what you already run. Morsa reads your ERP, MES, shared drives, email and chat, so the alarm that stopped Lathe 6 overnight also becomes a work order for maintenance.
