The estimator at this construction company built every job estimate by hand, working line by line through a Google Sheets calculator. A single estimate took more than 20 minutes. Multiply that across a year of estimates and it adds up to 325-plus hours of pure calculation work, more than eight 40-hour work weeks spent producing numbers rather than doing anything else the business needed from that person. The spreadsheet itself was accurate. The problem was that the volume of estimates going out was growing faster than one person working by hand could absorb.
Mapping the process before touching it
Before writing a line of automation, we spent a week embedded with the team, watching how an estimate actually got built: which numbers came from a straightforward lookup, which ones needed a judgment call only the estimator could make, and where the process was really just repetition dressed up as expertise. That distinction, process versus judgment, is what separates a tool the estimator actually trusts from a generic calculator bolted on top of the old spreadsheet.
What we built
We built the estimating tool as a web application running on the company’s own pricing rules, shaped around the process we had just mapped rather than a generic template. Price is resolved by lookup: base rate first, then location, thickness, diameter and area band, with tiered volume pricing on top. The estimator sees every step, so what comes out is a structured estimate ready to review, not a black-box number with no way to check it. Rates live in an admin panel the company updates itself, so a price change lands everywhere at once without waiting on a developer.
On top of that we built an MCP layer, so an AI assistant could drive the calculator directly: read the job inputs, run the same rules, and draft the estimate for a person to sign. It was built for two jobs. Inside the company it gave staff a way to ask for an estimate in plain language instead of filling the form out themselves. On the outside it plugged into the sales bot on the company’s own site, so a visitor asking what a job would cost got a real number off the same pricing rules instead of a promise that someone would call back. We also built an MCP server for Service Fusion, the field-service platform the company already ran its jobs and invoicing on, so the assistant could reach job data where it already lived instead of asking anyone to re-enter it.
The result
Estimates now take under 3 minutes instead of 20-plus, on a system that was live within 2 weeks of the mapping week ending. The mobilization and minimum-charge mistakes that used to slip through manual calculation are gone, because those rules now apply themselves. The estimator handles a growing volume of jobs without the manual workload scaling right alongside it.
This was one of three systems we built for the same company. The others: connecting the five separate tools they ran the business on, so an invoice mismatch now surfaces in hours instead of months, and giving the team plain-language access to their job and customer data, so nobody has to dig through the field-service platform screen by screen for an answer.
If your operation runs on estimation, quoting, or project scoping, our automation work for construction follows this same embed-first approach.