← All guides

Field-to-estimate loop: capturing what the job taught you

Jobs teach the company. Then the lesson dies in a closeout folder. The next bid runs on tribal memory again. A field-to-estimate loop is how the company keeps what the job paid for in blood, time, and change orders.

The job of this page

Define a simple loop: what to capture from the field, who owns it, and when it has to land back in estimating. Use AI to summarize and tag only when the notes exist. Make the next bid better without depending on who still remembers the last job.

Why closeout folders kill the lesson

Someone wrote a punch list. Someone dumped photos. Someone archived the final cost report. None of that is a loop into the next estimate. Estimating does not open that folder when a similar package hits the bid board. Ops does not have time to translate “what hurt” into units and assumptions estimating can reuse.

So the company pays for the same lesson twice. Or three times. Tribal memory is not a knowledge system. It is a single point of failure with a vacation schedule.

Define the field-to-estimate loop

Keep it boring and owned:

• What to capture. Productivity that was wrong. Scope that blew out. Vendor reality vs quote. Access and sequencing that broke the plan. A few lines on what you would price differently next time. Not every photo in the folder.
• Who owns it. A named role on the job (PM, superintendent, or coordinator) and a named intake on the estimating side. Two names. Not “the team.”
• When. A fixed window after substantial completion or after a major phase, before closeout amnesia. Same trigger every job.

If you cannot say what, who, and when in one breath, you do not have a loop. You have a hope that someone will remember.

Where AI helps

AI helps summarize and tag when the notes exist: field write-ups, cost actuals, change themes, short “what we would do differently” bullets. It can turn that into something estimating can find before the next similar bid.

AI cannot invent lessons from an empty closeout. If nobody wrote what the job taught you, the model has nothing honest to summarize.

One concrete picture

Mid-market GC. Three similar packages in eighteen months. Each one overruns on the same access assumption. Each closeout sits in a folder. Each next bid prices access like the first one. The people who lived the pain have moved on or are too busy to brief estimating.

Same company, after they define the loop: after each major phase, a short capture is required. What was wrong, who owns the write-up, when it is due. Estimating has a named intake. AI summarizes and tags the notes so the next bid can pull them. The fourth package still has risk. It does not have the same blind spot by default.

Scene is an illustrative pattern, not a named client case study.

What “done” looks like

Every job has a named capture into estimating on a fixed trigger. AI may summarize and tag. The next bid gets better without tribal memory as the only bridge.