AI for tradies: where to start when you've only used ChatGPT

If you run a trade business, you've probably heard some version of "you need to be using AI" from a supplier rep, a mate, a podcast or your accountant. Maybe you've opened ChatGPT, asked it to write an email, thought "that's handy", and then gone back to work. Fair enough. Nobody explained what the next step was.

This guide is for owners and office managers of subcontracting businesses (sparkies, plumbers, HVAC, carpenters, fire, data and comms) who want a straight answer: where does AI actually save time or money in a trade business, and where should you start?

First, what "AI" means in 2026

When people say AI now, they almost always mean large language models: software like ChatGPT, Claude, Gemini and Copilot that can read and write text, understand photos, and transcribe voice. That's it. It's not a robot, it's not going to swing a hammer, and it doesn't "know" your business unless you give it the information.

What these tools are genuinely good at:

Turning messy input into tidy output. A rambling voice note becomes a clear job description. A photo of a switchboard becomes a list of what's in it. A long email thread becomes three dot points.
Reading long documents quickly. A 60-page subcontract, a scope of works, a set of specs. It can find the clause about variations or notice periods in seconds.
Drafting. Emails, quotes, safe work method statements, toolbox talk notes, job ads, website copy. It gets you to a first draft fast.
Following rules you give it, consistently. "Only use items from this price list." "Always mention the clause number." This is where the real business value is.

What they're bad at:

Knowing your numbers. Ask ChatGPT what it costs to relocate a switchboard and it will confidently give you a number. That number has nothing to do with your labour rate, your supplier pricing or your margin.
Being right every time. They can misread things and they can make things up. Anything that goes to a customer or a builder needs a human to check it.
Remembering. A chat window doesn't know what happened on site yesterday unless someone tells it.

Keep those two lists in mind and most of the hype sorts itself out. AI is a fast, tireless office assistant that needs clear instructions and someone checking its work.

The three levels of using AI in a trade business

It helps to think about AI in three levels. Most businesses are stuck on level one, and that's fine. The point is to know what level two and three look like so you can tell real tools from hype.

Level 1: You use a chat tool yourself

You open ChatGPT (or Claude, Gemini, Copilot), type or paste something in, and copy the answer out. It's cheap, it's flexible, and it's where everyone should start. Good uses:

Rewriting a blunt email to a builder so it's firm but polite.
Summarising a long subcontract so you know what you're signing.
Drafting a SWMS or a toolbox talk to edit and check.
Writing a job ad for an apprentice.

The limit is that it only works when someone sits down and does it. It doesn't happen on site, it doesn't happen at 7am when the builder asks for a change, and it relies on whoever's doing it to give it the right information each time. We've written a separate guide with practical ways to use ChatGPT in a trade business, including a few things not to use it for.

Level 2: AI built into the tools you already use

Your job management software, accounting software, email and phone are adding AI features. Think call summaries, auto-filled timesheets from voice notes, invoice data pulled out of supplier PDFs, or suggested replies in your inbox.

These are worth switching on when they appear, because you're not changing how you work. Check the settings, try them for a couple of weeks, and keep the ones that save time.

Level 3: AI that does a whole job for you

This is where the bigger savings are, and it's what people mean by AI agents. Instead of you prompting a chatbot, a tool is set up for one specific job in your business, with your information loaded in. It runs every time that job comes up and hands the result to a person to check.

For a subcontractor, good candidates have three things in common:

It happens often, every week or every day.
It follows rules you could write down: a price list, a contract, a checklist.
It slips when people are busy, and that costs money.

Variations are the classic example. A builder asks for a change on site. Someone needs to record who asked, check whether it's in your scope, price it from your rates, send written notice before the deadline, and chase the approval. It's rules-based paperwork, it happens constantly, and when the office is flat out, small variations quietly don't get claimed. (We wrote about this in why verbal instructions cost subbies money.)

Other level-three candidates in a trade business: turning site diaries or voice notes into daily reports, reading supplier invoices and matching them to jobs, and pre-filling compliance paperwork like test sheets from photos.

Where to start: a sensible order

You don't need an "AI strategy". You need to try one thing, see if it saves time, and then try the next. Here's the order we'd suggest for most subcontractors.

1. Get one person comfortable with a chat tool (week 1)

Pick whoever does most of the office work. Get them a paid account on ChatGPT, Claude or Copilot (roughly the price of a couple of coffees a week), and give them one task: use it for every email to a builder or client for a week. They'll learn quickly what it's good at and what needs fixing.

2. Sort out the privacy basics (week 1)

Before anyone pastes in contracts or customer details, read our guide on whether it's safe to put contracts and customer details into ChatGPT. The short version: use a business account, turn off training on your data, and don't paste in anything you wouldn't email to a stranger unless you've checked the settings.

3. Use it to read your contracts (week 2)

Upload a subcontract for a live job and ask it: "What does this say about variations, written notice and the time limits for notice? Quote the clauses." Then check what it says against the document. This is one of the highest-value things you can do with a chat tool, because most subbies sign contracts they've never fully read. It's not legal advice, but it tells you where to look.

4. Find your most expensive leak (week 3)

Ask yourself: where does money or time slip through the cracks every month? For most subcontractors it's one of these:

Variations done on a verbal instruction and never claimed, or claimed too late.
Quotes that take days to get out, so the job goes to someone else.
Timesheets and job costs that don't get matched up until month end.

Pick the biggest one. That's where a level-three tool is worth paying for.

5. Try a tool built for that one job (month 2)

Look for something that works the way your crew already works (a photo, a text, a voice note) rather than a new app they have to learn. Ask the provider three questions:

Does it use my prices, contracts and rules, or does the AI make them up?
Does a person in my office approve everything before it goes to a customer or builder?
Is my data used to train their AI? (The answer should be no.)

What not to do

Don't let AI price your work from thin air. If a tool can't show you which line on your rate card each price came from, don't send it.
Don't give your crew another app. Adoption on site is everything. If it's harder than sending a text, it won't get used.
Don't buy a platform to "do AI". Buy something that fixes one specific, expensive problem.
Don't remove the human check. The good tools draft; people approve.

The bottom line

AI isn't going to run your business, and you don't need to understand how it works to get value from it. Start with a chat tool for emails and contracts, get your privacy settings right, then find the one piece of paperwork that costs you the most when it slips, and look for a tool that does exactly that job using your own rules and prices.

For a lot of subcontractors, that piece of paperwork is variations. The Variation Capture demo shows what level three looks like: a photo and voice note from site turned into a priced, contract-checked variation that the office approves and the builder signs off from a link.