In June 2026 we hired our own estimator team. Not to run a service, and not to check customer work. To sit inside the product team and tell us when we were wrong.
Software people make confident guesses about drawings
A drawing set looks structured, and it is tempting to treat the whole thing as a database with an awkward interface.
It is not. A drawing set is a document produced under deadline by a person, for other people who share a lot of unwritten context. Two engineering firms will tag the same equipment differently. A revision will change a schedule without changing the plan. A detail sheet will contradict a floor plan, and someone on the bid desk will know which one to trust.
You cannot infer that from the file. You have to have done the job.
What changed once estimators were in the room
The most useful thing they did was reject features. Several ideas we thought were valuable turned out to be solving problems distributors do not have, usually because we had borrowed the idea from contractor estimating.
They also changed how we think about being wrong. An engineer's instinct is to push accuracy up. An estimator's instinct is to ask what happens on the items the model is unsure about, because that is the part that costs them time. That is why ambiguous items are flagged for review rather than quietly guessed, and why we publish 85% first pass accuracy instead of a number that sounds better.
The hiring argument
If you are building for a trade, the domain expert is not a consultant you call at the end. They are the person who stops you shipping something that looks right and is not.
We would make the same call again, and earlier.
