There is a rule of thumb I have used for years: a CPQ license should cost no more than about 0.1% of your revenue. It is a useful sanity check, and it exposes an uncomfortable truth about my own industry.
Run the numbers for a €15M manufacturer of, say, awnings or garage doors. By the rule, they should spend roughly €15,000 a year on CPQ licenses. Now look at what traditional CPQ actually costs once the implementation project, the integrations and the consultants are counted. The market never built anything for that budget. So an entire tier of manufacturers, configure-to-order businesses with real complexity and real dealer networks, has been quoting from Excel sheets and PDF price lists for twenty years. Not because they do not need a configurator. Because a glass floor kept them out.
I spent two decades on the enterprise side of that floor, running large CPQ implementations, and I will confess: for most of that time I barely thought about the companies below it. They were simply not the market. This post is about why that floor existed and why it just broke.
Picture the companies. Sunshading and awning manufacturers. Garage door makers. Veranda and outdoor-living builders. Machine builders and vehicle upfitters. Typically somewhere between €5M and €25M in revenue, with products that are genuinely configurable: dimensions, fabrics, motors, controls, mounting options. Sold largely through dealers and installers who need correct quotes fast.
These companies have every CPQ problem the enterprise has. There is one senior person in the office who really knows what can be combined, and every quote queues behind that person. Dealers wait days for answers. The website shows a brochure and a contact form, while buyers increasingly want to explore, configure and price things themselves. Often on a phone, often outside office hours, asking the questions they do not dare ask a salesperson. I find that last part quietly fascinating: the first thirty minutes of a buying journey now happen in private, and most manufacturers are not even present for them.
What these companies never had was a CPQ project that made economic sense. Six-to-eighteen-month implementations, consultant day-rates, license models designed for enterprise procurement. The vendors were not wrong to build for the enterprise. But the result is that the segment with perhaps ten times as many companies was simply never served.
Two things, and they arrived together.
AI removed the implementation mountain. The slow, expensive part of CPQ was never the software. It was translating product knowledge into a configuration model: workshops, spreadsheets, rule-writing, testing. I have personally spent months of my life in those workshops. Large language models have collapsed that work. Product documentation, price lists and option catalogues can now be turned into a working, solver-backed configuration model in days. In the AI-native projects I am involved in, a working product takes roughly 40 hours and goes live on a website in about two weeks. That is around ten times faster than a traditional CPQ project, and speed is not a convenience here. It is the thing that makes the economics work at all.
Lean modeling became a philosophy instead of a compromise. The enterprise instinct is to model everything. But everything you put into a CPQ is a tax on future development, because every rule must be maintained forever. The lean approach keeps the model to roughly 500 variants and applies what I call the 90%-of-price principle: find the minimum set of modules that determines about 90% of the price, and cover the rest with a budget line. A dealer does not need the last screw specified to move a deal forward. They need a correct price, a professional proposal and a buildable bill of materials, today.
Put those together and the 0.1% rule finally has products on the right side of it: from around €1,000 per month, live in weeks, no mega-project.
One customer I work with, a lift manufacturer, went from roughly two quotes a week to twenty after putting an AI-guided configurator on their website.
I will be honest, as always: a strong ad campaign helped drive the traffic. But the traffic converted because visitors could configure and price a real machine instead of filling in a contact form.
The shape of the change matters more than the number. Quoting stopped being a bottleneck around one expert and became a self-service channel. Customers and dealers guide themselves to a valid product, and every proposal arrives as an interactive page rather than a static PDF: configured price, a product summary, a ready-to-go bill of materials, still adjustable, with every adjustment re-validated. The senior expert's knowledge did not disappear. It got cloned into the assistant, which doubles as training for every new rep and dealer.
And because the guidance is an LLM working inside a constraint solver's guardrails, the quotes are valid by construction. Self-service without correctness would be a liability, not a channel. That architecture deserves its own article, and I have written it: Can You Trust an AI-Generated Quote?
Here is the part I find most exciting, and it is the part that gets lost when people fixate on speed. When implementation stops eating the whole budget, a CPQ project can finally tackle the problems we never got to in the old days.
Think about what was always cut from scope in a traditional project. Pricing analytics: which segments are price-sensitive, where margins leak, which options are quoted but never bought. Customer self-service as a real sales channel, not a brochure. Dealer onboarding that takes an afternoon instead of a training week. The long tail of products that never made it into the configurator because modeling them did not pay. Proposal tracking, so you know which quotes are being read and which are dying in an inbox. Every one of these was a "phase two" that never came, because phase one consumed everything.
Now the math flips. Go live in the first month and the configurator starts earning or saving money immediately: quotes that used to queue behind your expert now go out the same day. That is not the finish line, it is the funding for everything after it. From there you build step by step, in the order the live data tells you matters: sharpen the pricing, open self-service to dealers, extend the model, connect the analytics.
In our own projects we are typically live within a month and then work together with the customer for about a year. Not because go-live is slow, but because go-live is where the interesting work starts.
The goal is not a configurator.
The goal is a lean selling machine and the quick start is simply how you pay for the journey while making it.
Skip the eighteen-month evaluation. That is an enterprise ritual from the other side of the glass floor.
Take your best-selling configurable product line, apply the 90%-of-price principle, and put a guided configurator for it on your website within a month. Then let the data speak: which options customers actually explore, where quotes stall, which dealers self-serve and which still call. You will learn more about your pricing and your product in eight weeks of live quoting than in a year of workshops. I have written before that waiting for perfect pricing before starting CPQ is like waiting to be fit before starting to exercise. It applies double here.
The glass floor was never about your company being too small for CPQ. It was about CPQ being too heavy for the economics. That excuse is gone. The real question is no longer whether you can afford to configure and quote digitally. It is which of the problems you have been living with for twenty years you want to solve in year one.
What should CPQ cost for a mid-sized manufacturer? A useful rule of thumb: license costs at no more than about 0.1% of revenue. For a €5-25M manufacturer that means roughly €5,000-25,000 per year. AI-native platforms start around €1,000 per month; traditional enterprise CPQ rarely fits under the rule once implementation is counted.
How long does it take to implement a configurator? Traditional CPQ projects run six months to two years. An AI-native, lean-scoped configurator can be live on a website in about two weeks, with a working product in roughly 40 hours of effort, provided you resist modeling every variant on day one.
Is go-live the end of the project? No, and it should not be. Go-live within the first month is what funds the rest: the configurator starts earning or saving money immediately, and you then expand step by step guided by live data. Pricing analytics, customer and dealer self-service, the long tail of products. A typical engagement runs about a year, but it is live and paying for itself from month one.
Do configurators work for dealer networks? Yes, dealer sales is one of the strongest use cases. Dealers get correct prices and professional proposals without waiting for the factory's product expert, and the manufacturer keeps control of validity and pricing through the constraint model. Sunshading, garage doors, verandas and machinery are typical examples.
Do I need to clean up my pricing before starting? No. Waiting for perfect pricing before starting CPQ is like waiting to be fit before starting to exercise. Start with the products that drive revenue, and let live quoting data show you where the pricing logic actually leaks.
What is the 90%-of-price principle? Model the minimum set of modules that determines about 90% of a product's price, and handle the remainder as a budget quote. It keeps the configuration model lean, fast to build and cheap to maintain, because every additional rule is a tax on future development.