Introduction: A Confession From 100 Projects

"Can someone sanity-check this configuration before I send the quote?"

I heard that sentence for the first time in 2000. I have heard it every year since, in different languages, in different industries, from salespeople selling cranes, compressors, elevators, trucks, and medical equipment. The tools changed. The sentence survived.

I have spent twenty-five years in configure-price-quote, now at cpq.se, across more than a hundred projects with manufacturers of complex machinery. In 2007 I was proud of a rollout that consumed seven hundred hours and shipped on time. The design binder was thick, the model was elegant, and the customer applauded. That was what good looked like.

Earlier this year, I built a working configurator from a product PDF in days.

Not a demo. A working tool: asking sensible questions, producing valid configurations, explaining its choices. The kind of thing that used to be the seven-hundred-hour binder.

A former customer called me around that time. He has led manufacturing businesses longer than some product managers have been alive, and he had seen the same demonstrations I had. He didn't ask how the technology worked. He asked one question: "How is that going to change your revenue model?"

That was the moment the penny dropped. He was not asking about features. He was asking about economics. When the cost of something collapses by ten times, you don't get the same work done faster. You get a different market, different winners, and different losers.

Here is what twenty-five years in the field has taught me about who those winners will be, and it is the entire argument of this small book: AI cannot sell what your organization hasn't structured. Everyone is bolting language models onto sales processes and wondering why they hallucinate machines that can't be built and prices that can't be defended. Meanwhile, the manufacturers quietly pulling ahead are not the ones with the best models. They are the ones whose products, prices, and knowledge are structured enough for machines to reason about.

The engine, in other words, already won. Constraint-based configuration has been selling complex machinery correctly for a quarter of a century. What's changing now is everything around the engine: who can use it, what it costs to build, and how much of your market it can reach. Most CPQ failures I have witnessed were never technical. They were behavioral and organizational. That's good news and bad news. Good, because the fix is available to everyone. Bad, because most companies won't do the work, and they will buy AI instead of doing it.

This is a field guide, not a textbook. Thirteen ideas, one per chapter, each built the same way: a scene from the field, the pattern behind it, one number that should bother you, one named anti-pattern to hunt down in your own company, and one takeaway you can repeat in the elevator. You can read it on a flight. You can act on it Monday.

The companies winning with AI are not the ones with the best models. They are the ones with the best structure.