Where AI Actually Pays Off in a Trades Business (And Where It Doesn't)

There are two conversations about AI happening in the trades right now. One is vendors selling 'AI-powered' versions of software you already didn't need. The other is a small group of operators quietly compressing their back office by 30–50% on specific workflows. We run our own construction company on an AI-forward stack — Claude, MCP integrations, and modern tooling wired into the systems we already owned — so this is a field report, not a forecast.

Where it pays today

Intake and lead handling. Every call, text, and form fill summarized, logged to the CRM, and routed with next steps — the same day. The trades lose more revenue to slow follow-up than to lost bids. This is the highest-ROI workflow we've automated, and it's not close.

Reporting and analysis. Weekly job-cost summaries, AR aging narratives, and exception flags ('this job's labor is tracking 22% over estimate') generated from data sitting in QBO and your CRM. What used to be a half-day of someone's week becomes a standing morning report.

Document-heavy workflows. Insurance scopes, supplements, subcontractor agreements, warranty letters, collections sequences. Anywhere the job is 'read this pile, produce that document,' current AI is already faster and more consistent than the person who used to dread doing it.

Dispatch and scheduling support. Not full automation — a copilot that drafts the schedule against crew capacity and flags the conflicts a human should decide.

Where it doesn't (yet)

Estimating judgment, customer trust, and field quality. AI can draft an estimate from photos; it cannot tell you the roof has two layers under the architectural shingles, that this customer will be a referral machine if you take care of her, or that your best crew lead is about to quit. Vendors promising to automate judgment are selling the one thing the technology doesn't have.

How to sequence it

Don't buy tools first. Map where the hours go, pick the one workflow that costs you the most (it's usually intake or reporting), automate it inside the systems you already own, and run it for a month before touching the next one. The wins compound; the failures stay small. That sequencing discipline is the whole game — and it's exactly what an AI roadmap session is for.

The right question isn't 'what can AI do?' It's 'which of my workflows is a pile of reading and writing that a smart system could own?'