Manufacturing Quoting Process: Nine Stations, Six Failures

The email arrived on a Wednesday from the managing director of a custom label and overlay maker in central Europe. Screen printing, digital printing, die cutting, laser, CNC milling, assembly: one customer part can cross six of those before it ships. He was rebuilding the company's quoting, and his description of the manufacturing quoting process he had inherited is the most honest one I have read from anyone running a plant: "I personally find our estimating process surprisingly difficult to get my head around. I don't know how to analyse it properly as a process." The quoting lived in Excel tools that his one experienced estimator had built over many years, so comprehensive that nobody else could maintain them, and the estimator was approaching retirement.

We were not the right software for him and I said so in my first reply. But his three questions have stayed with me, because they are the same three questions a four plant company with twenty five estimators is asking, with more zeros attached: what is actually broken, how do we get the knowledge out of the person before he leaves, and should we buy something, build something, or combine what we already have.

This page is my answer to all three, written for the people who own the problem rather than the people who do the work: the VP of sales who signs the quotes, the CIO who has been told to fix quoting and has only been shown configurator demos, and the CFO who reads the fixed price risk factor in the annual report and wonders what sits behind it. It is a long read. The short version is first.

What the manufacturing quoting process actually looks like

Ask a company to draw its quoting process and you get four boxes: request comes in, estimate, quote goes out, follow up. Ask the estimator what happened between the second and third box on the last job and you get a forty minute answer. The process has nine stations. The four box version leaves out the five where the money is decided.

Station by station, with where each one lives in a typical company:

1. The request arrives. Almost never as a clean request for your product. It is a bid package written by somebody else, a consulting engineer or an EPC contractor or an OEM's purchasing group, in which your product is a minority of the pages. It lands in a sales inbox.
2. Bid or no bid. In theory a decision. In practice a feeling, formed in an email thread, often after somebody has already spent a day reading.
3. Reading. Finding every sentence that changes the price. On a 668 page municipal treatment package we studied, 14 sections carried requirements that moved the equipment price and the other sections did not. On a fire apparatus request for proposal, roughly three hundred paragraphs each needed a yes or a no. This is the slowest station and the only one nobody times.
4. Scope and assumptions. What is included, what is excluded, what was assumed where the document was silent or contradicted itself. On a 179 page marine package we found seven places where the drawings and the specification disagreed. Every one of them is a decision, and the decision is usually made silently.
5. The cost build. Material, routing, hours, bought out items, freight, testing. This is the estimator's workbook, and it is the station everyone means when they say "estimating".
6. The price. Margin, risk, the customer, the market, the backlog. A number chosen by a person. The formula is usually one line and it is almost never written anywhere.
7. Review. Twenty minutes from someone senior, who looks at the total and the margin and not at the reading, because the reading is not visible to be looked at.
8. Submission. One number on a bid form, an exceptions list, a validity period that will be held longer than it says.
9. Feedback. Won or lost, then built, then the actual cost. In most companies none of this ever returns to the quote that started it. The loop is drawn with a dashed line because it is usually not there.

Where it breaks: six failure points

We have built a quoting engine on a real manufacturer's archive and kept a registry of every class of mistake it made along the way. It stands at 44 entries. Seventeen are document reading, seven are scope and counting, seven are pricing policy, five are product classification, and the rest are process. Zero are arithmetic. Human quoting fails in the same places, for the same reasons, and it fails there quietly.

1. Reading is unowned and unmeasured

Nobody in the company is responsible for reading as a step. It is folded into "estimating", so it has no owner, no time budget and no quality check. Yet it is where the expensive mistakes live. On an overhead crane specification we read for this site, a single leftover sentence required nuclear certification for a gate crane on a flood control dam. On the fire apparatus document, a foam system was plumbed, gauged and bracketed on three pages and never itself specified. On a heat exchanger preselection, a reissued section changed the sphere size on page 5 and left the old size on page 9. None of these is a pricing error. Each is a reading error that becomes a pricing error, and a review that looks only at the total will never catch it.

2. Scope is decided silently

Give the same package to two careful estimators and they will come back with two different numbers, not because one is careless but because the document forced a dozen small decisions and each made them differently. The public version of this is a Florida DOT letting we analysed where six bidders priced an identical floating dock scope and the spread was 2.7 times from lowest to highest. Inside a single company the spread is smaller, but we have seen double digit percentages between two people in the same room pricing the same drawing, and neither could tell you afterwards which assumption caused it, because the assumptions were never written down.

3. The rules live in heads and in formulas nobody reads

The managing director who wrote to me described his estimator's tools as so comprehensive that nobody else could understand, maintain or take them over. That is the normal condition of a quoting workbook after ten years. When we mine a company's costing archive we read the formulas, not the computed values, because the values hide what the formulas do. On one archive the single largest thing we found was a second structural member stacked inside what looked like a rate cell. It was invisible in every printed quote and it had been there for years. Nobody was hiding it. It was simply how the estimator had solved a problem one afternoon and never had reason to explain.

4. Margin is a feeling with a name

When we reverse engineered one manufacturer's pricing from their own filed quotes, the whole identity fitted on one line: price equals material plus labour, divided by one minus a margin. It reproduced their totals to the cent. The margin itself moved from job to job, chosen by the estimator and the general manager from the customer, the backlog and the mood of the month. That line had never been written anywhere in the company. The rule was real, the company ran on it, and nobody could have told a new hire what it was.

5. One person, one file

Stations 4, 5 and 6 usually sit with one estimator and one workbook. When that estimator is on holiday, quotes stop or get worse. When the estimator retires, the company discovers that the workbook cannot be operated without him. The email that started this page is the small company version. The large company version is a plant whose entire pricing memory is one senior estimator in his sixties, and a succession plan that consists of hoping.

6. There is no feedback loop

Software development learns within minutes when a change breaks something. Quoting learns months later, if ever, and usually without being told why. We wrote about this structural gap in why quoting is hard to automate. The actual cost of a job lives in the ERP system. The quote lives in a spreadsheet or a PDF. In most companies no one has ever joined the two tables, so the estimator has been calibrating for years against a signal he cannot see. Won jobs teach nothing about margin. Lost jobs teach nothing at all.

Where the days go: quote turnaround time in custom manufacturing

Owners measure quote turnaround time as one number, request in to quote out, and it is usually two to four weeks for an engineered job. Inside that number the time is not spent where people assume. Station 5, the cost build that everyone calls estimating, is typically a day or two. The rest is reading, waiting and queueing.

Reading is the largest block and the least visible. A 99 page fire apparatus request gave bidders three weeks in total, and roughly three hundred paragraphs each needed a yes or a no with a justification for every no. A federal crane section ran 27 pages, cited 43 standards and required 31 approved submittals; the reading was the work. Then come the waits: supplier quotes for bought out items, an engineer's opinion on a detail the document leaves open, a customer's answer to a question that should have been asked in the first hour and was asked in the second week. Then the review queue, where a finished estimate sits until the one person who signs has an afternoon.

The useful measure is not the total. It is time to first number: how long from arrival until someone can say, with a written list of assumptions, roughly what this job costs. Companies that measure it find that reading is most of it, and that the two week quote could have been a three day quote with a one day reading step that had an owner. Quote turnaround time is a symptom of station 3. Treat it there.

The same process with twenty five estimators and four plants

Scale does not change the six failures. It multiplies them by plant and then hides them behind an organisation chart.

Each plant has its own lineage of workbooks, descended from whoever set up estimating there, with its own conventions for scrap, for setup, for what counts as overhead and for what margin is normal. A customer who sends the same drawing to two plants gets two prices, and the difference is the difference between two people's habits. Every plant believes its own number. Head office sees a margin report that averages them and looks fine.

The CIO's instinct at this point is to buy a configure price quote system, because that is what the market sells and because it demonstrates beautifully. It fixes station 5 for products that can be described as options on a list. For a company whose quotes begin with a three hundred page document written by someone else, it fixes nothing at stations 3 and 4, which is where the money was lost. We have written a buyer's guide that says this against our own interest: for catalogue and formula products, a configurator is the better tool and you should buy one.

Consolidation projects fail for a simpler reason. You cannot put four plants on one quoting system until their rules are written down, and the rules were never written down, because they lived in four heads. The software project becomes a rule capture project in disguise, discovered in month six, after the licence is signed.

What your peers tell their investors

The risk is not hidden. It is in the annual report, in the language of fixed price contracts. Search the latest 10-K filings of Powell Industries, Graham, ESCO Technologies, CECO Environmental, SPX Technologies, Oshkosh, Ducommun, Babcock and Wilcox, Lindsay and JBT Marel and you will find the words cost estimates and fixed price in every one of them. Graham's, filed in June 2026, says it most plainly: "the original cost estimates in these or other contracts prove to be inaccurate, or the contracts do not permit us to pass increased costs on to our customers, our profitability may decrease or losses may be incurred". Its defence contracts, the same filing notes, can run more than five years from order to shipment.

Read that sentence as a description of a process rather than a legal disclaimer. The inaccurate estimate was made at stations 3 to 6, months or years before revenue was recognised, by a person reading a document and building a cost in a spreadsheet. Fixed price contract risk is quoting risk with an accounting label on it, and it is the only risk factor in the report that the sales and estimating organisation owns entirely.

How to audit a quoting process: five instruments

This is what I would do in the first two weeks at any custom manufacturer, from the label maker to the four plant group. None of it requires software. All of it requires access to the archive and a few hours of the estimators' time.

Instrument 1: the quote genealogy

Take one recent, typical quote. For every number in it, trace where it came from: a rate table, a supplier quote, a formula in the workbook, a previous job, or a person's judgement. Draw it as a tree. The output is a single percentage, the share of the final price that traces back to something written down, and a list of the numbers that trace back to a head. On the first genealogy we ever drew, the written share was under half, and the estimator was surprised, because it all felt like method to him.

A genealogy for one line of a real quote looks like this. The pattern is the point, not the figures, which are illustrative:

Six numbers, two and a half of them written. Now multiply by the eighty lines on a real quote and you have the ratio the company actually runs on.

Instrument 2: the two estimator test

Give one real package to two estimators, independently, with no discussion, and a day each. Compare the totals, then compare the scope lists and the assumptions. The spread between the totals is your quoting variance, the number your customers experience as inconsistency and your CFO experiences as margin noise. The differences in the assumption lists are the decisions your process currently makes silently. Run it quarterly and the spread becomes a metric.

Instrument 3: the rule capture ratio

List the pricing rules the estimators actually use, from the genealogy and from interviewing them. For each rule, ask whether it is written where a new hire could find it. Then apply the admission bar we use when mining an archive: a rule counts as captured only if it is stated with its mechanism, it fits at least three past jobs, and it still holds when any one of those jobs is removed. Rules that fail the bar are not wrong; they are judgement, and judgement needs a different treatment, which is a question to the estimator with a bracket on the answer rather than a silent default.

Instrument 4: the single point of failure register

People, files, and the formulas inside the files. For each: who else can operate it today, what happens next Tuesday if that person is unavailable, and what happens in three years if they retire. This is the register the managing director needed before his estimator left, and it is the one register almost no company keeps. It takes an afternoon.

Instrument 5: the feedback check

Pull ten jobs won in the last two years. For each, put the quoted cost beside the actual cost from the ERP. If that takes more than a day, you have no feedback loop, and everything the estimators believe about their own accuracy is untested. If it takes an hour, look at the spread, then look at whether anyone has ever acted on it.

Together the five instruments produce a picture no dashboard shows: how much of the price is method and how much is memory, how far two people diverge, which person or file the company cannot lose, and whether the process has ever learned from a result. We have turned the five into a twelve question quoting system health check you can score in five minutes. It is not a substitute for running the instruments. It tells you which one to run first.

The estimator is not the problem

Every version of this story tempts the owner to conclude that the estimator is the bottleneck, and every version of that conclusion is wrong. The estimator is the only part of the process that works. He reads the document nobody else reads, makes the decisions the process has no other way of making, and holds the rules the company never wrote. The managing director's complaint that his estimator's tools were too comprehensive is, read the other way, a description of a man who took the job seriously for twenty years while nobody asked him to explain it.

What the process has done is treat his judgement as free. It is not free. It is the company's most concentrated asset, held in a form that cannot be copied, audited or inherited. The audit is not an audit of him. It is the first time anyone has written down what he does, and in our experience estimators are relieved by it rather than threatened, because the alternative they can see coming is a software project that ignores them and a retirement that takes the company's memory with it.

So run the instruments with the estimator, not on him. The genealogy is his tree to draw. The rules that fail the admission bar are his to keep as judgement, with a bracket. And the two estimator test is a measurement of the process, which is why it should be run on the most senior person too, and why the result should never be attached to a name.

Manufacturing quoting best practices, in order of payoff

The internet is full of quoting best practices lists written by software vendors, and they all recommend buying software. This one is ordered by what fixes the most for the least, and software is last.

1. Write the rules down before you buy anything. If the rules cannot be written, no system will hold them. Rule capture is the whole project in the small company and most of the project in the large one. Start with the genealogy and the admission bar above.
2. Make reading a measured step with an output. A list of every requirement that changes the price, each with its page number, produced before any costing begins. Estimators call it a compliance matrix. It makes station 3 visible, so review can look at it, and it turns a silent scope decision into a line someone can disagree with.
3. Review the reading, not just the total. Twenty minutes on the requirements list finds the nuclear sentence and the phantom foam system. Twenty minutes on the total finds nothing, because the total looks reasonable either way.
4. One pricing identity per company, not per plant. If price equals cost divided by one minus margin, say so, and say which costs are in and which margin bands apply. Plants can keep their own rates. They cannot keep their own definitions.
5. Record the assumptions on the quote. Every place the document was silent or contradicted itself, write what you assumed, in the quote. It protects you commercially and it is the raw material for the next rule.
6. Sample the variance. The two estimator test, quarterly, on one live package. Publish the spread internally. Nothing else makes estimators converge faster.
7. Close the loop with actuals. Quoted against actual, per job, quarterly, in front of the people who quoted. This is the only calibration signal that exists, and most companies have never once looked at it.
8. Then, and only then, choose software. By the time the first seven are done, you know which lane you are in, and the choice is usually obvious.

Buy, build, or combine

The third question from the label maker is the one every CIO ends up asking, and the honest answer depends on what the quote starts from.

If the quote starts from a known product and a list of options, the customer chooses a model, a size, a finish and a quantity, you are in configurator territory and there are twenty mature products that will do it well. If the quote starts from a print job or a converting job with a routing across known machines, the print industry's management systems already price that layer by layer, and I told the label maker exactly which ones to call. If the quote is a repeat of a part you have made before, your ERP's quoting module is probably enough.

If the quote starts from a document somebody else wrote, hundreds of pages in which your product is a minority, and the thing being priced has never been built, none of those help at the station where the money is decided. That is the lane we build for, and it is also the lane where writing the rules down matters most, because the engine is only as good as the pricing logic it can be given. We wrote a longer build or buy piece for people who want to try the general purpose route first, and it sometimes answers build.

Combine is the usual honest answer for a group: a configurator for the standard lines, a quoting engine for the engineered lines, and one written pricing identity across both so that the margin report means something.

What we do, and where the audit fits

Mavlon builds quoting engines for custom manufacturers. The engine reads the whole package, cites every requirement to its page, puts the contradictions in front of a person, and prices in the company's own logic, mined from its own archive. We test it blind, against filed quotes the engine has never seen, with the pass bar written down in advance; the protocol is public.

Every engagement now starts with the five instruments above, run by us on your archive, because we learned the hard way that the engine cannot be better than the rules that can be written. We call it a quoting system audit. It ends in one of three answers: keep your spreadsheets and here are the rules we wrote down for you; buy a configurator, because your products are more standard than you believed; or build the engine with us, on the archive we have just mapped. We give the first two answers when they are true. We gave the second to the label maker, by email, before any call.

If you own a quoting process you cannot get your head around, that is the conversation to have. Bring the package you least want to read again.