We build software for other companies, so it’s a little embarrassing to admit that the most useful thing that happened during the build of our own system was a stretch of nearly a month where nothing happened at all.
Some background. One of us is a partner in LawnJunkie, a landscaping company here in Niagara, and with its owners’ blessing we started replacing its office — the spreadsheets, the whiteboard, the pile of carbon-copy estimate pads one of us still refuses to throw out — with software we wrote ourselves. The centrepiece was estimating. If you’ve never priced landscaping work, the short version is that a quote is a stack of guesses: how many hours to demo the old patio, how much aggregate goes under the new one, what a crew actually costs you per hour once you count the truck, the insurance, the downtime.
The build went fast. That part of the sales pitch is true, and we say it with mixed feelings, because we remember paying for the other kind of project. Records, screens, the crew app, the plumbing between them. Weeks, not quarters. For a while the system was growing faster than we could agree on what to point it at next.
Then it stopped.
Not the code. The code was fine, sitting there finished like a truck with no keys in it. What stopped everything was a list of eleven numbers the software needed before it could price a single job. Production rates, mostly. How many square feet of interlock the two-man crew actually lays in an hour. Not the number in the industry manual. Ours. Real material costs, not the supplier’s list price before the discount nobody writes down. And the big one: the loaded cost of a crew-hour, which it turned out none of the three of us could state, and all three of us could argue about.
We argued about it for weeks.
I want to be precise about what those weeks were, because this is the part every software vendor’s website leaves out. Including, until this post, ours. Nothing was blocked in any technical sense. There was no bug. No integration was fighting us. The machine was waiting on the business to decide what was true, and the business discovered it didn’t know. Deciding meant timing crews and feeling slightly ridiculous doing it, pulling a year of supplier invoices, and one genuinely uncomfortable meeting about whether some of the company’s work had been priced on hope rather than arithmetic.
No AI could have shortened those weeks. We’d go further: no AI should. The entire value of an estimating engine hangs on those eleven numbers being true, and the truth was in our trucks and our invoices and our stubbornness. It was not in any model’s training data. A tool that guessed them for us would have produced something worse than slow software: fast, confident, wrong quotes, at scale.
This is the thing we now say out loud in sales conversations, and it lands strangely, because people expect a firm like ours to claim the opposite. AI compresses the software side of a build. The coordination, the typing, the eighteen months of meetings where a spec quietly goes stale. It compresses that part dramatically, and we’ve stopped being surprised by it. What it cannot compress, even slightly, is the part where your business has to decide what’s true. Your production rates. Your real costs. Which customers you actually make money on. If a vendor tells you their AI handles that part too, they’re describing a system that will lie to you quickly.
Here’s the ending the story deserves, though. The system came back to life the day the last number went in, and the numbers turned out to matter more than the software. The software just made them unavoidable. We reprice against them every season now, in an afternoon, and the work that was quietly losing money stopped, just as quietly, being sold. Those stalled weeks were the only part of the whole project we’d refuse to compress, looking back. That’s where the company learned something, instead of just the computer.
So when we tell a prospective client the first day is free and gets spent inside their systems instead of on a call, this story is most of the reason. Somewhere in your business there’s your version of the eleven numbers. The fastest build in the world will stand around waiting for them, exactly like ours did. Better to find out what they are on day one, while finding them is still free.