An estimation team that can finally bid on everything it wants to.

Estimation is where the margin on a bid is decided. It is also, in most shops, where the senior people spend their week on work that is not estimating: pulling numbers across, filling the same cells, chasing the same details from one bid to the next.

What makes this one specific is how the problem arrived. Over the years the bar for precision was raised, repeatedly and for good reasons, and the process never changed to carry it. Output per estimator-hour drifted about twenty percent below where it had historically sat. Nobody made a bad decision along the way. The work got harder while the method stayed the same, which is how most of these situations actually happen.

Leadership wanted that productivity back at unchanged precision, and could see roughly what it would take. What it did not have was the work counted step by step, and nobody wanted to invest in tools against a hunch.

Count it before building anything

We mapped how the work actually flows today, step by step, with volumes, handoffs, tools and friction points. We evaluated the tools on the market against that real process rather than against a feature list. For each step we said whether it should stay human, be assisted, be automated, or be left alone. Then we quantified the hours at stake against the department's own volumes.

Closing that gap is a twenty-five percent efficiency gain, which is the same work in eighty percent of the time. Against a department of roughly twenty-five thousand hours a year, that is a little over five thousand hours. That was the target.

What the count found was three levers worth ten to fifteen thousand hours a year between them.

Which reframed the decision entirely. The target did not need the plan to work. It needed about half of the conservative case to materialize, and everything above that was upside. A department told it needed a twenty-five percent gain was looking at two to three times that, and the conversation stopped being about whether to do it.

The count also killed the shortcut. The measurement and quantity work is about sixty percent of an estimator's week and splits into three distinct processes, and none of the three is fully automatable while all three are partly automatable. Of the tools evaluated against the real process rather than against a feature list, none was a silver bullet. One was set aside as too young a company to carry a function this critical, which is a much easier call to make with the process mapped in front of you.

The target process was validated at a steering committee, and the decision to build was taken on a costed design rather than on assumptions.

Twelve skills, built with the estimators, on real bids

Every skill is built the same way, in one ninety-minute session with an estimator, on a file they have to deliver anyway. They stay at the keyboard throughout. The session produces two things: the deliverable for that file, finished and usable, and the first version of the skill. Nothing gets done twice, and no case is fabricated for the exercise. The estimator then runs it on their next real bid for a week, and the corrections they make become corrections to the skill.

The skills fill the cells of the spreadsheet the estimators already use. They do not touch the macros and they do not change how anything is measured: the measurement stays in the tool it was always done in, and it stays human. The assistance is in finding things, not in measuring them. Claude proposes, the estimator checks and signs, and every skill states its own boundary: what it covers and what it does not.

One piece of the work is not a build at all. The largest of those three processes is roughly forty percent of the quantity work, and none of the rules for it are written down anywhere, because the knowledge sits with the senior estimators. Those rules get captured in a workshop, written up, corrected by the seniors and signed off before anything is built against them. They stay the company's property whatever happens to the skills, which is arguably the more durable asset.

On a typical firm bid of eighty hours, twenty-seven to forty-three hours come back. The gain is in transcription and administration. It is not in the measuring and it is not in the judgment, and saying so precisely is what made the number believable to the people who would have to live with it.

None of this is about a smaller team. Every hour released goes back into bids the company does not submit today, which in a bidding business is the most direct line there is between released time and revenue. That framing was fixed before the build started, along with two commitments that were repeated to the estimators in writing: this is not a headcount exercise, and nobody is assessed on their use of Claude. That is part of why the estimators were willing to sit in the sessions.

A second diagnostic is now running on project management, the function immediately downstream, on exactly the same pattern: count it first, then decide what to build.