A construction company builds twice: once on site and once on paper. The paper side (estimates, RFIs, submittals, logs, schedules) is where AI gives a project manager hours back.
Where does AI help a construction company first?
On the documents that stack up on every job.
- Estimating and takeoff help. Reading plans to draft quantities and a first-pass estimate for your estimator to check. The estimator owns the number.
- Plan and spec questions. "What does the spec call for on the roof membrane?" A tool that answers from the project's own plans and specs and cites the sheet or section, so anyone can open the page and check.
- RFIs and submittals. First drafts of RFIs from a superintendent's notes and photos, and submittal packages pulled together from the spec section. A PM reviews before anything goes to the architect.
- Daily logs. A foreman talks for two minutes before leaving the site, and the tool turns it into a clean log entry: crew, weather, work done, deliveries, delays.
- Schedule updates. Drafting the weekly update for the owner from the logs and the schedule, in plain language.
Can AI find bid opportunities?
Yes. Permit filings, planning agendas, and public bid postings are public, and a tool can watch them and flag projects that fit your trade, size, and area. Someone on your team still decides what to chase. At Quartz Importers MN, where I am the fractional CTO, I built an AI agent platform and knowledge graph that tracks companies, people, projects, and permits for the sales team. The platform took in the whole market in about 30 days. The details are in the Quartz Importers case study.
Where should AI stay out?
Safety decisions stay with people. Fall protection, excavation, crane picks, confined spaces, and the call to stop work belong to your competent person and your safety lead. A tool can draft the toolbox talk or find the right section of your safety plan. It does not decide whether a site is safe. The same goes for anything you stamp, sign, or certify.
Can a tool read plans without making things up?
It can when it answers only from the project documents and shows the sheet or spec section for each answer. When the answer is not in the documents, it should say so and hand the question to a person. That is a build, set up and tested on a real project set, with a PM checking answers until the crew trusts it. A general chatbot answering from the open internet is the wrong tool for a spec question.
What does it cost to start?
- The AI you already pay for, a 90-minute on-site session, is $450. Most offices have AI in their email and office software that nobody has turned on.
- A half-day workshop with estimators, PMs, and supers is $1,500. You leave with a ranked list of what to fix first.
- Builds, like a plan-and-spec question tool, a daily log assistant, or a permit watcher, are scoped at a fixed price or run at $150 an hour. See AI agents and workflow automation.
Answered by Jared Lukes, Minneapolis. Twenty-five years in brand, UX, and technology, currently holding fractional AI and technology seats at three companies. About Jared.