Chapter 1: The Machine That Couldn't Be Built

The crates arrived at a villa outside Bangkok on a Tuesday morning.

Inside them: a home elevator, sold six months earlier through a distributor. The homeowner had taken time off work to be there. The installation team unpacked the panels, checked the drawings, and started measuring the shaft. The distributor was on WhatsApp within the hour.

Left had become right somewhere between the quote and the factory. The travel height was off by forty millimetres. The door type didn't match the local building regulation.

The quote had been clean. The site wasn't.

Here's the thing about complex machinery that nobody in the software industry seems to understand: the product often doesn't exist when you sell it. A home elevator is a promise. It gets engineered after the signature, manufactured to order, shipped in pieces, and assembled months later in someone's house. As one manufacturer put it to me: "It's not until the product arrives on-site that the product is put together."

That sentence should terrify you more than any AI headline.

Because it means quoting errors in this business don't fail loudly. Nobody catches them at signing. They travel quietly through engineering, through the factory, through a shipping container, and reveal themselves at delivery, in front of the customer, months after anyone can cheaply fix them. A large share of on-site failures are quoting failures that took a long time to show up.

The quote is the digital blueprint for a physical promise. If the blueprint is wrong, everything downstream is expensive.

The daily version

Most days, the failure is less dramatic and more corrosive.

It's 9:02 on a Tuesday morning. A salesperson needs a "quick check" on a valve package. By 9:45, three senior engineers are gathered around a screen, debating a dependency that lives in a PDF last updated in 2017. The quote will go out, eventually. Correct, probably. And three of your most expensive people just spent their morning as a verification service.

I've watched versions of this scene for twenty-five years, in more than a hundred projects across machinery manufacturers. We take engineers, people hired to design products, and quietly promote them to Human Middleware: the connective tissue between what sales promised and what the factory can build. It feels like diligence. It's actually a system design failure with a salary.

And it produces a question every executive should be able to answer and almost none can: if your top expert left tomorrow, what part of your quoting logic would leave with them?

What the delay costs

Let's put numbers on the quiet version, because the loud version, the crates in Bangkok, at least gets noticed.

Deals decay. In my experience the win probability of a quoted deal drops by roughly half a percent per day it sits unanswered or unresolved. That sounds small. It isn't. A quote that takes twenty days to finalize has given away ten points of win probability before the customer has said a word about price. Quote age correlates with discount size, too. Delay drives desperation, and price becomes the apology for time.

Now run the arithmetic on a mid-sized manufacturer doing 300 complex quotes a year.

Two days of avoidable delay per quote is 600 days of pipeline drag. A ten percent drop in win rate on contested deals is roughly thirty lost orders. At €60,000 average margin, that's €1.8 million. Add rework: four hours of cleanup per quote is 1,200 engineering hours a year. Add discount drift: two points of unmanaged discounting on €40 million of bookings is another €800,000 that left the building without anyone deciding it should.

None of this appears on a P&L as a line item. It appears as "competitive pricing pressure" and "long sales cycles." The cost of inaction is already on your books. You're just not labeling it.

Anti-pattern: Human Middleware. When your engineers spend their days validating quotes instead of designing products, you don't have a quality process. You have a concierge service with an engineering payroll. Saves by heroes are a signal your structure is broken, not a medal for the heroes.

The wrong diagnosis

When executives see these symptoms (slow quotes, error rates, engineers drowning in validation), the reflex diagnosis is a knowledge problem. Our products are complicated. Our people need more training. Maybe the new AI can read all our documentation and answer the questions.

Wrong diagnosis. Your engineers already know everything needed to quote correctly. The knowledge exists. What's missing is structure. The knowledge lives in PDFs from 2017, in spreadsheets with hidden formulas, in the heads of the three people everyone calls, in an ERP that has become a museum of variants. Every hidden formula is a rule. Every extra spreadsheet tab is a policy. Every local tweak is a fork. And none of it can be checked, versioned, or guaranteed.

You don't have a knowledge problem. You have a structure problem.

That distinction is the whole game, and it's worth restating as plainly as I can. A quote is a contract. A quote is a promise your factory has to keep, at a price your CFO has to live with, on a date your customer is planning around. Anything that generates quotes without guaranteeing that promise is generating risk with good formatting.

Which brings us to the obvious move of the moment. Faced with all of the above, the tempting fix is to point a large language model at the problem. It has read everything. It writes beautifully. It never gets tired.

Half your sales engineers are probably already doing it. Quietly, in a personal ChatGPT tab, without telling you.

The next chapter is about why that makes everything worse, faster.

A quote is a promise your factory has to keep. If your structure can't guarantee that promise at signing, you will pay for it at delivery: in margin, in time, and in trust. And no AI can fix at the end what your organization hasn't structured at the start.