AI Price Estimator for a Construction Company: Built From Your Past Estimates

A construction company's estimating manager was building every estimate by hand, one job at a time. The estimates already written held the pricing history a tool could learn from.

The manager handles estimating and operations, and the same pattern shows up at many home service businesses that quote custom work. Here is the estimator we built from past work, the process work around it, and a side problem we found by looking at a shared screen.

The situation

Estimates at the company were produced manually. The company had two things worth learning from: its recent estimates and its pricing sheet.

We had two aims on the call. The first was to start an AI price estimator that helps draft estimates. The second was to put the estimating-to-bid process on paper before automating any of it.

What was broken

Two problems, one obvious and one not.

The obvious one was manual estimating. The less visible one was a set of scheduled AI triage tasks that had been left running. They were creating email drafts nobody wanted, even after people had been told they were off. We found them by sharing a screen and looking at what was actually scheduled.

What we built

These are the pieces that came out of the call.

• One folder: the estimates consolidated into a single shared Google Drive folder, with a subfolder for each job. A tool can only learn from estimates it can find.
• Process map: the current estimating-to-bid process written down first as an "ugly 1.0", with the gaps labeled, before anything gets automated.
• Estimator: an AI price estimator that drafts estimates, built from about 20 of the company's recent estimates plus its pricing sheet.
• Cleanup: the active scheduled triage runs were paused and the root cause investigated.

What changed

We cannot give an accuracy figure or a time saving from this call. What changed is that estimating now has a draft tool behind it. The estimator, the shared folder and the written process give the next review something concrete to inspect.

The stray triage runs were paused, and the root cause was still under investigation.

What to take from this

This section is advice for your own business, not a fact about the client. A few habits carried the most value here.

• Write the ugly 1.0 first. Describe how an estimate gets from request to bid as it works today, flaws included, and mark every gap. Automating a process nobody has written down only makes the wrong steps faster.
• Audit what is scheduled. After you tell a team a scheduled AI task is off, open the schedule itself and confirm it. A task that keeps running quietly can create work for weeks.
• Keep a person in the loop. Treat the first AI estimates as drafts, and have someone who knows the work check scope, quantities and the final number before anything goes to a customer.
• Start with finished jobs that have clean paperwork. Compare what the estimator produces against what those jobs actually involved, and fix the pricing sheet where the two disagree.

Frequently asked questions

Can AI estimate a job from past estimates?

It can draft a first number by learning how a company priced earlier work, using past estimates and the pricing sheet as its guide. A person should review every estimate. We have no accuracy results yet, so treat a tool like this as a draft maker until it earns trust.

How many past estimates do I need to build an estimator?

We started from about 20 recent estimates plus the pricing sheet. That was a starting set, not a rule, and we did not test a smaller or larger set on this call. Keep adding finished jobs so the estimator keeps learning from current pricing.

What is an ugly 1.0 process document?

It is the first honest write-up of how a process works today, with every gap labeled and nothing polished. We wrote the company's estimating-to-bid steps this way before automating, so the build matches the real process and not a tidy guess.

How do I find scheduled AI tasks that are still running?

Open the schedule list for each tool and read it with someone who uses the account. We found ours during a screen share. Pause anything unexpected, then work out why it kept running, so the cause is fixed and not just hidden.

If estimating eats your week, book a 30-minute AI follow-up and reporting audit. We map your estimate-to-bid steps with you and show where an AI draft could save the most time.