Tribal Knowledge in Manufacturing: The Quoting Desk Version

The managing director of a custom label and overlay maker in central Europe wrote to me this month about his estimator. Not to complain. The man had spent years building the Excel tools the company quotes with, and they worked. The problem was the sentence that followed: the tools had become so comprehensive that nobody else could understand, maintain or eventually take them over, and the estimator was approaching retirement. That is tribal knowledge in manufacturing in its purest form, and it is worth noticing where it was sitting. Not on the shop floor, where every article on the subject looks for it. On the quoting desk.

On the floor, tribal knowledge is the gap between the work instruction and what the good operator actually does. On the quoting desk there is no work instruction. The knowledge is not a gap in the documentation; it is the documentation, and it is held in one person and one file. This piece is about that version: what it looks like, real examples from quoting desks, what it costs, and a checklist for the month before the person who holds it walks out. It is written for the owner, the VP of sales and the CIO, not for the estimator, who already knows all of this and has been waiting for someone to ask.

In brief

• Tribal knowledge in manufacturing concentrates in three places: maintenance, setup and quoting. Quoting is the most concentrated and the least written down, because the floor has work instructions and the quoting desk has a spreadsheet one person built.
• The occupation is shrinking and turning over at the same time. The U.S. Bureau of Labor Statistics projects cost estimator employment to fall 3 percent between 2025 and 2035 while about 17,600 positions open every year, all of them, in its words, to replace people who leave or retire.
• Real examples are small and specific: a percentage added "for now" to a supplier price, a second structural member stacked inside a rate cell, a margin rule that runs the company and was never written, a scrap rate estimated by feel.
• Interviews recover a fraction of it. The archive of past quotes and workbooks recovers most of it, because it records what the person did rather than what they remember doing.
• Below: an inventory template and a twelve point knowledge transfer checklist for a retiring estimator. Neither needs software.

What tribal knowledge in manufacturing is, and where it concentrates

Tribal knowledge is what the company knows only because particular people know it. It is not written anywhere a new hire could find it, it is not in the ERP system, and it usually cannot be explained by the person who holds it without a specific job in front of them. Every manufacturer has it. The question is where it has pooled, and the answer is different from the one most articles give.

There are three pools. Maintenance: the technician who knows that the press on line 2 needs its hydraulic pressure nudged before a cold start, which is in no manual. Setup: the operator who knows the offset that makes the first part good rather than the third. And quoting: the estimator who knows what this customer's drawings usually leave out, what the shop actually spends on a job like this, and what the number has to be to win without losing money.

The first two pools have been shrinking for twenty years, because lean programmes, standard work, work instructions and now video capture were built to drain them. The third has not shrunk at all. There is no standard work for quoting an engineered job. The quoting desk was never part of the lean programme, because it is not on the floor, and it was never part of the IT programme, because the spreadsheet worked. So the knowledge stayed where it was, in a person, and the company built a growing business on top of it.

The turnover numbers make this a timed problem rather than a general worry. The U.S. Bureau of Labor Statistics counted 226,400 cost estimators in 2025 and projects the occupation to decline 3 percent by 2035, mainly because software raises the productivity of the people who remain. Yet it expects about 17,600 openings a year over the same decade, and it states plainly that all of those openings are expected to result from the need to replace workers who transfer to other occupations or exit the labor force, such as to retire. Fewer estimators, replaced continuously, each replacement starting from a spreadsheet they did not build. The wider industry sits behind the same wall: the Manufacturing Institute and Deloitte projected in 2024 that manufacturing could need as many as 3.8 million additional workers by 2033 and that 1.9 million of those positions could go unfilled.

How it forms: one afternoon at a time

Nobody decides to build tribal knowledge. It accrues. A customer's package leaves out a dimension; the estimator works out a way to infer it and puts the inference in a cell. A supplier's price goes stale; he adds a percentage. The general manager says "we should be at twenty two on this kind of job" in a corridor; the estimator remembers. Each of these is a small, correct solution to a problem that arrived that afternoon, and each is folded into the tool or the habit rather than written beside it, because writing it down would take longer than solving it and there is another quote due.

Ten years of afternoons produce a workbook with forty tabs and a person who can operate it. The company's revenue grows on top of that workbook, which is what makes the arrangement invisible: nothing is broken, so nothing is examined. The label maker's estimator was not hoarding. He was doing the job well, alone, in a company that never had a reason to look inside the tool until the day he mentioned retirement. The first time most owners see their quoting process is the day it announces that it is leaving.

This is why tribal knowledge in manufacturing quoting is different in kind from the floor version. A missing work instruction is a gap in a system that exists. A quoting workbook nobody else can operate is the system, and its author is the only documentation. You cannot fix that with a training video. You have to extract the rules from the artefact and write the system that should have existed, and the artefact is the archive.

Tribal knowledge vs institutional knowledge vs tacit knowledge

People search for the difference, so here it is briefly. Institutional knowledge is what the organisation knows in a form that survives any individual: procedures, drawings, the rate table, the ERP history. Tacit knowledge is what an individual knows but cannot easily state, the feel for a weld or for a customer. Tribal knowledge sits between them: it is stateable, it just was never stated, and it is held by a group or a person rather than by the institution. The important practical point is that most quoting tribal knowledge is not tacit. It is rules, and the rules could be written. Nobody asked.

Tribal knowledge examples from real quoting desks

These are the kinds of things we find when we take apart a custom manufacturer's quoting archive to build a pricing engine on it. None is dramatic. All of them decide money.

The percentage added "for now"

A supplier's price per pound sits in the workbook from an email fourteen months old. The estimator knows it is stale, so he adds a percentage in his head, or in a cell with no label, before using it. The email is filed and the percentage is not. Ask him what it is and he will say "about eight, depending", and the depending is the tribal part.

The second member inside a rate cell

The single largest thing we found in one archive was a structural member that appeared nowhere in the bill of materials. It was stacked inside what looked like a rate per foot: the cell multiplied by two members' worth of steel because, years earlier, the estimator had decided the design always needed the second one and had folded it into the rate rather than adding a line. Every computed value in every quote was correct. Nothing on the printed page showed why. We only found it because we read the formulas rather than the numbers they produced.

The margin rule that runs the company

When we reverse engineered one manufacturer's pricing from its own filed quotes, the whole identity fitted on one line: material plus labour, divided by one minus a margin. It reproduced their totals to the cent. That line had never been written anywhere in the company. The margin itself moved job by job, set in a conversation between the estimator and the general manager. Nobody could have told a new hire what the rule was, and the company had been running on it for a decade.

Scrap rate by feel

On the label maker's milled panels, the tolerances and the cosmetic standard drive the scrap rate, and the scrap rate is the price. There is no table for it. The estimator looks at the drawing, remembers the last three jobs that looked like it, and writes a number. That number is probably good. It is also uninheritable.

The datum everyone knows

Specifications for floating docks state freeboard to the top of the walking deck. Our own engine once measured it to the top of the float, one frame depth lower, and bought more flotation than the job needed on every dock. Every estimator in that industry knows the datum. None of them had written it down, because why would you write down something everyone knows. We wrote it down after it cost us, and it is now a gate that runs before pricing.

The clause the estimator always ignores

Government crane specifications routinely carry leftover sentences from a master template, including on one job we read a requirement for nuclear certification on a flood control dam. An experienced estimator reads that sentence, recognises it as a leftover and prices past it. A new one either prices it, doubling the number, or asks, costing a week. Knowing which requirements are real is one of the most valuable things the senior person knows, and it exists nowhere except in his reading.

The number that is constant but not derivable

In most archives there are values that are the same on every job and that nobody can trace. A setup allowance, a freight figure, a testing line. The estimator inherited it from his predecessor, who may have calculated it once in 2011. It is not wrong, necessarily. It is tribal, and when the estimator leaves it becomes a number with no parent, which the next person will either keep out of superstition or delete out of confidence.

Which rate era a quote lives in

The shop rate changed in 2023. Quotes before that date are in old money. When a manager pulls up a similar job from 2022 to sanity check a new quote, the senior estimator knows to adjust; the junior does not know there is anything to adjust. The eras are in one head.

What it costs, in the owner's terms

The cost does not show up as a line item, which is why it survives. It shows up in four places.

Variance. Two estimators price the same package and come back double digit percentages apart, not from carelessness but because each carries a different set of unwritten rules. Customers experience this as inconsistency. The margin report averages it away.

Speed. Every quote waits for the person who holds the rules. When he is on holiday the queue stops or the quality drops, and nobody measures either, because quote turnaround time is reported as one number with no breakdown.

A single point of failure with a date on it. The estimator will retire, and the date is known. What the company has instead of a plan is the hope that the replacement will absorb it in the overlap period. In our experience the overlap period is spent quoting, because the work does not stop for the handover.

The retirement itself, worked through. The shape is the same in every company that has been through it. Three months of overlap in which the retiring estimator keeps quoting because the backlog does not pause, and the successor watches. A farewell. Then six to twelve months in which the successor quotes slower, asks the retiree by phone until the phone stops being answered, and either inherits the old workbook without understanding it or rebuilds a simpler one that drops rules nobody told him existed. The variance in that first year is the cost, and it is paid twice: in jobs won too cheaply because a rule was dropped, and in jobs lost because a rule was over applied. The company sees neither line, because there is no feedback loop between the quote and the built job, which we wrote about in why quoting is hard to automate. It sees a margin that drifted and a customer who went quiet.

Risk you already disclose. If the company is public, the annual report says that inaccurate cost estimates on fixed price contracts can reduce profitability or produce losses. We quoted one such filing in the manufacturing quoting process piece. Read that sentence knowing the estimates were built from rules that exist in one person, and it stops being boilerplate.

The four plant version

Everything above scales, and it scales by plant. A group with four sites has four quoting lineages, each descended from whoever set up estimating there, each with its own leftovers, its own margin corridor and its own stale percentages. The corporate view is a consolidated margin report that averages the four and looks reasonable. Run the inventory at each plant and put the four side by side, and the contradictions appear in an afternoon: the same customer's packages read as leftover heavy at one site and taken literally at another; scrap allowed at eight percent here and by feel there; a rate era boundary that one plant marked in 2023 and another never marked at all.

The CIO who has been told to put all four on one quoting system discovers at that point that the system cannot be configured until the four sets of rules are written, and that writing them is the project. Done in the right order, the inventory becomes the specification for the software. Done in the wrong order, the software becomes a very expensive way of finding out that the rules were never written, usually in month six, after the licence is signed. The inventory costs an afternoon per plant. Start there.

Why interviews fail and the archive does not

The instinct when an estimator announces retirement is to sit him down and ask him to explain how he quotes. It produces a document that is true and mostly useless. People describe the method they believe they follow, which is tidier than the one they actually follow, and they leave out everything that feels like common sense, which is precisely the tribal part. The datum, the leftover clause, the percentage for now: none of those would come up in an interview, because none of them feels like knowledge to the person who holds it.

The archive has no such modesty. Twenty to fifty past quotes with the packages that triggered them and the workbooks behind them record what the estimator did, job after job, including the things he does not know he does. Read the formulas rather than the values, compare the same rule across jobs, and the tribal knowledge comes out as a list of candidate rules, each with the jobs it was seen on. Then you interview the estimator about the list, which is a different conversation: not "how do you quote" but "on these three jobs you added eleven percent to the plate price and on this one you added nothing, what is the rule". He will tell you. He always knows once the job is in front of him.

We described that method in full in how to capture estimator knowledge before it walks out, including the bar a rule has to pass before it counts as captured: three past jobs, a stated mechanism, and survival when any one job is removed. This piece will not repeat it. What follows is the two artefacts that piece did not give you: the inventory and the checklist.

The tribal knowledge inventory

An inventory is the register the label maker needed before his estimator left and that almost no company keeps. It takes an afternoon to start and it is never finished, which is the point. One row per item of knowledge. The columns are the questions an owner needs answered.

Three things about using it. The "who holds it" column is the single point of failure register in disguise; sort by it and you see what one retirement takes with it. The "three job test" column is what stops the inventory becoming a wish list: an item that does not fit three past jobs with a stated mechanism is judgement, not a rule, and the honest entry for judgement is a question with a bracket, not a made up number. And the "written?" column, once the inventory is a few dozen rows long, gives you a rule capture ratio, which is the one number that tells you how much of your quoting is method and how much is memory.

One warning about the table. It is tempting to fill it in a single sitting from what the owner and the estimator can think of, and to feel finished. That inventory will be short and flattering, because memory produces the rules that feel like rules. The rows that matter come from the genealogy and the formulas, where the knowledge nobody thinks of as knowledge lives. Fill it from the archive first and from memory second, and expect the second pass to argue with the first.

Knowledge transfer checklist for a retiring estimator

This is what I would do in the last ninety days, in the order I would do it. It assumes the work does not stop, because it never does. Print it.

1. Freeze the archive before the farewell. Copy every quote, package and workbook from the last five years into one place, read only, before anyone tidies up. Retiring people delete things they think are embarrassing, and the embarrassing things are the useful ones.
2. Read the formulas, not the values. Open every workbook with formulas showing and list what each tab actually computes. The stacked member and the unlabelled percentage live here. Two days for one person; more for a workbook with forty tabs.
3. Draw a quote genealogy for three typical jobs. Trace every number on each quote to its source: rate table, supplier quote, formula, previous job, or head. The share that traces to a head is the size of your problem.
4. Build the inventory from the genealogy, not from memory. Every "head" source becomes a row. Then walk the rows with the estimator, job in hand, one at a time.
5. Apply the three job test to every row. Rules that pass go into a written pricing identity. Rows that fail are recorded as judgement, with the question you will have to ask on each future job and a plausible bracket for the answer.
6. Run the two estimator test once before he leaves. The successor and the retiring estimator price the same live package independently. The gap is the knowledge not yet transferred, shown as a number and as a list of differing assumptions. Do it again a month later.
7. Write the customer notes. For each repeat customer: what their packages usually leave out, which of their clauses are leftovers, who to call, what they have rejected before. This is the fastest loss when a person leaves and the easiest to prevent.
8. Mark the rate eras. A dated table of every shop rate, material basis and margin band change, so that old quotes can be read in old money by someone who was not there.
9. Record the assumptions on the last twenty quotes retroactively. Where the document was silent or contradictory, what did he assume. He remembers now. In six months nobody will.
10. Move the successor into the review seat early. For the last month the successor reviews the retiring estimator's quotes rather than the other way round. Reviewing forces the questions that shadowing never does.
11. Agree what stays with him. Some judgement will not transfer and should not be faked. Write down which decisions the successor will escalate, to whom, and for how long a paid consulting arrangement with the retiree covers them. Most retirees are glad to be asked.
12. Put the pricing identity where a stranger could find it. One document, in the shared drive, linked from the quote template, reviewed once a quarter against actual job costs. If it lives in the successor's head, you have moved the problem five years down the road.

Is "tribal knowledge" offensive, and what to say instead

People search this, and it deserves a straight answer. Some organisations have moved away from the phrase because it borrows the word "tribal" for something unrelated to any actual tribe, and prefer "institutional knowledge", "undocumented know-how" or "experience based knowledge". Any of those is fine and this article uses them interchangeably in places. I have kept the common term in the title because it is the one people type when they are looking for help, and an article they cannot find helps nobody. Use whichever phrase your company is comfortable with. The inventory works the same under any name.

Where software helps, and where it does not

No software captures tribal knowledge from an interview, whatever the demo says. What software can do is read the archive faster than a person, find the candidate rules and the places where they disagree, and put a question list in front of the estimator instead of a blank page. That is how we build quoting engines for custom manufacturers: the archive first, the rules second, the estimator's judgement kept as questions with brackets rather than silent defaults, and the whole thing tested blind against filed quotes under a protocol we publish.

It is also why every engagement starts with the inventory above rather than with a demo. We run it as a quoting system audit, and one of its honest outcomes is that the rules are now written, the spreadsheet is fine, and you do not need us. If you want to know before any call how much of your quoting is memory rather than method, the five minute health check will tell you which of the instruments to run first. The label maker's estimator has not retired yet. The inventory is the thing to do this month, not the month after.