Most construction AI writing is written for people who want to sell you something, or for people who have never had to explain a cost overrun to a client. This guide is written for the person who has to decide: is any of this worth doing on my jobs, and where do I start?
It is short on hype and long on the parts that decide whether AI works in a construction company — which decisions it can improve, what your data has to look like before any of it works, what goes wrong, and how to run a pilot you can actually judge.
Updated September 2026 · roughly a 45 minute read, or a 90-day plan if you do the work.
If you read nothing else, read this:
- Most of the value today is arithmetic over data you already own — cost at completion, schedule risk, cash — not magic. Prefer the forecast you can audit.
- The label covers three different technologies with three different failure modes: computer vision, language models and predictive statistics. Name which one you are looking at before you buy it.
- Your data decides. Coded, current, reachable. Fixing those three pays for itself before any AI arrives.
- One decision, one owner, one job, twelve weeks. That is the pilot, and it beats a company-wide rollout every time.
- If the output leaves the company, a person reviews it. If you cannot see the working, you cannot defend the number.
| If you are… | Start with | The question you actually have |
|---|---|---|
| Owner / GM | A 90-day plan | "What should I spend on this, and what do I get?" |
| Estimator / chief estimator | Estimating and takeoff | "Will this bid more work with the same team?" |
| Project manager / superintendent | Progress tracking, Scheduling | "Does this tell me something I don't already know?" |
| Controller / finance | Job cost and cash forecasting | "Which jobs are going over, while they're still running?" |
| IT / ops lead | Your data decides, Risks and controls | "What has to be true before we plug a tool in?" |
| Anyone being sold to | How to buy and pilot | "How do I tell a real product from a demo?" |
-
What "AI in construction" actually means Three different technologies, three different failure modes, and how to tell which one you're being sold.
-
The use cases that work today — what each does, what it needs from you, where it goes wrong, and how to judge it:
-
Your data decides what AI can do Coded, current, reachable. The three properties that separate a tool that works from one that produces plausible nonsense — and why fixing them pays for itself before any AI arrives.
-
Risks and controls Confidently wrong answers, data leaving your control, alert fatigue, black-box forecasts, write access, and what your contracts and insurers expect from you.
-
A 90-day plan to start Pick one decision, check the data, pilot on jobs where you know the answer, then run it live with a named owner. Includes a worked example and kill criteria.
-
How to buy and pilot The questions that separate a product from a demo, what a good answer sounds like, and how to structure a paid pilot.
| Template | Use it for |
|---|---|
| AI use-case scorecard | Comparing candidate use cases on hours, cost and risk before you buy |
| Pilot scorecard | Scoring a pilot on evidence instead of enthusiasm |
| AI acceptable use policy | A one-page rule set for what staff may paste into general AI tools |
| Vendor questions | The checklist to take into a sales call |
- It does not name a winner. Tool capabilities change quarterly; the questions that expose a weak product do not. Where tools are named at all, they are named as categories.
- It does not quote statistics it cannot source. Where a number would help, it tells you how to measure your own instead. Your hours are the only statistic that matters to your decision.
- It does not promise that AI replaces anyone on site. Nothing in here does that, and nothing you buy today will either.
- It is not legal, tax or insurance advice. Chapter 4 covers the questions to take to your lawyer, broker and accountant — not the answers they should give.
Read it once, then do chapter 5. The failure mode of AI in construction is not bad technology — it is a company buying six tools, running none of them on a real decision, and concluding that AI doesn't work in construction. One decision, one owner, one job, twelve weeks.
Each use-case chapter repeats the vendor questions and the back-test method on purpose, so that any one of them can be handed to the person who owns that decision and read alone. Start with the 90-day plan if you would rather read one chapter and get moving.
Worked examples in this guide use invented numbers and are labelled as such. Formulas and process steps are standard construction practice.
Open construction resources from the same team, all maintained alongside this one:
| Repository | What it is |
|---|---|
| Construction data migration | A guide and toolkit for moving a contractor between systems, and proving nothing was lost. |
| Construction project records | Open schemas, templates and a checker for RFIs, submittals, change events, daily reports and punch lists. |
| Construction reference data | Cost codes, units, waste factors, pay units, trade sequence, glossary and metric formulas in CSV. |
| Construction reference MCP server | An offline MCP server that gives AI assistants construction reference data and calculators. |
| Construction prompts | 28 prompts for ChatGPT, Claude and Gemini, from bid go/no-go to notice letters. |
| Construction agent skills | 28 installable agent skills for Claude Code and any agent that reads SKILL.md. |
| Open source construction tools | Open source software for BIM, CAD, scheduling and site work, verified against the GitHub API. |
Maintained by Constructelligence — building the AI infrastructure for construction.