AI in the trades, 29 October 2026
Teaching AI your rate card, rules and margins
You teach AI your prices by writing them into one rate card file, with your labour rates, outputs, material prices, waste, margin and rules, and loading it into a project so every chat reads it. ChatGPT and Claude both have projects that hold instructions and files. Then test it on three jobs you've already priced and fix it until the answers match.
Why does AI get my prices wrong?
Because it doesn't know them. Ask a general AI tool to price a 40 m² flat roof and it estimates from what it has read online, which is somebody else's labour rate, somebody else's merchant and somebody else's margin. It will give you a confident number either way.
Chatting with it doesn't fix that. Telling it your day rate in one conversation doesn't mean it uses it next week. What it does use, every time, is the set of instructions and files attached to a project. That's where your rates belong. If you want to see the kind of mistakes that follow, read why ChatGPT quotes go wrong.
Where do you put your rates so AI uses them?
In a project. Both ChatGPT and Claude let you create a project with its own instructions and uploaded files, and every chat inside that project reads them.
| Tool | Where rates go | Limits (as of October 2026) |
|---|---|---|
| ChatGPT | Project settings: instructions plus project files | 5 files on Free, 25 on Go and Plus, 40 on Pro and Business |
| Claude | Project instructions plus project knowledge | Free users can create up to five projects |
ChatGPT's project instructions apply only inside that project and override your general custom instructions, so you can keep a pricing project separate from everything else. Check each tool's help pages for current limits: they change.
What goes in an AI rate card?
Everything you'd tell a new estimator on day one. Labour by the day and by the unit, how much you get done in a day, materials at your merchant price, waste, margin, minimums, and the things you never include.
RATE CARD: Business name, last updated date LABOUR Gang day rate (2 people): £650 per day. This already includes overheads. Single operative day rate: £340 per day. Minimum charge for any job: £450 + VAT. OUTPUT PER DAY (gang of 2) Single-ply flat roof: 30 m² per day Strip and re-tile, concrete tiles: 25 m² per day Fascia and soffit replacement: 12 linear metres per day MATERIALS (my merchant price, ex VAT) Single-ply membrane and adhesive: £9.00 per m² Concrete interlocking tiles: £x.xx per m² Skip (8 yard): £x each RULES 1. Labour per m² = gang day rate ÷ output per day. 2. Add 10% waste to all sheet and tile materials. 3. Price = (labour + materials with waste) ÷ (1 − margin). My margin is 15%. 4. Round the final price up to the nearest £10. 5. If a job is under 1 day, charge the minimum. 6. Scaffolding is always priced separately by scaffolder name. Never estimate it. 7. If you don't know a material price, say so and ask me. Never guess a price. 8. Show your working line by line so I can check it. ALWAYS EXCLUDED Asbestos removal, structural repairs, building control fees, making good of decorations.
Two rules do most of the work. Rule 7 stops the tool inventing a price when it doesn't have one. Rule 8 makes it show the sum so you can spot a wrong number in seconds. If you don't yet know your gang day rate, work it out first from overheads per man-day.
How does a day rate become a price per m²?
Divide the gang's day rate by what they get done in a day, add materials with waste, then gross up for margin. This is the sum you want the AI to do, so do it once by hand and check that it matches.
Flat roof, 40 m², from a £650 gang day rate
A two-person gang lays 30 m² of single-ply membrane a day.
| Labour per m²£650 ÷ 30 m² | £21.67 |
| Materials per m² with waste£9.00 × 1.10 | £9.90 |
| Cost per m²£21.67 + £9.90 | £31.57 |
| Price per m² at 15% margin£31.57 ÷ 0.85 | £37.14 |
| Price for 40 m²£37.14 × 40 | £1,485.60 |
Your numbers, not a promise. Change any figure to match your firm.
If the AI's answer differs from yours, ask it to show its working and find the line that's wrong. Usually it has applied margin as a markup (multiplying by 1.15 instead of dividing by 0.85) or forgotten the waste. Add a rule that fixes it. Our guide on markup and margin explains the difference.
How do you test that AI has learned your rates?
Run it on jobs you've already priced and won. If it matches your price within a few pounds on three jobs, start using it on new ones, still checking every quote.
- Pick three finished jobs.
One small, one typical, one awkward. You know what you charged and whether it made money.
- Describe each one as you would from site.
Measurements, materials, access. Don't give it the answer.
- Compare line by line.
Labour, materials, waste, margin, total. Note where it differs.
- Add a rule for each mistake.
"Always round up to the nearest £10." "Never price scaffold."
- Repeat until all three match.
Then use it on live quotes and keep checking.
A roofing and cladding firm in Ireland, eight on the tools, saw why the coverage figures matter. During their trial, Nexley flagged that board coverage had been entered as 0.75 m² per board when the correct figure for that board was 1.44 m², which changes the number of boards on every job. A wrong figure in the rate card gives a wrong price on every quote that uses it. Get your coverage rates from the manufacturer's data sheet, not memory.
What can't AI do with your rate card?
It can't see the job. It prices what you describe, so a missed chimney, rotten deck or awkward access stays missed. It also can't tell you whether your rates are right for your market. A roofer we spoke to had been pricing at about £60/m² where the work was going for nearer £36.50/m². An AI with his rate card would have priced every job at £60 too.
Use AI to do the arithmetic and wording quickly and the same way every time. The judgement on what the job needs and whether the price will win stays with you. See also how to price a job for costs people forget to put on the card.
Where Nexley fits
Nexley is the AI back office for UK trade firms: five AI assistants run your tenders, quotes, buying, scheduling and invoicing. It connects to the tools you already work in, you run it all from WhatsApp, and you approve everything. £799 a month.
Nexley keeps your rate card, outputs, margin and exclusions as part of your business setup, builds a price list from your merchant invoices, and applies it to every quote it drafts. You send a voice note or WhatsApp message from site; Nexley drafts the priced quote with its working and you approve it before it goes out.
Common questions
Does ChatGPT remember my prices between chats?
Not reliably from ordinary chats. Put your rates in a project's instructions or files, which are read in every chat in that project. Memory settings vary by plan and change, so don't rely on them for prices.
Should I upload my whole price list or type the key rates?
Both. Type the core rates and rules into the instructions, because the tool reads those every time. Upload longer price lists as a file and tell it in the instructions to use that file for materials.
Can AI work out my margin for me?
It can do the sum if you tell it your costs and target margin. It can't tell you what margin you need: that comes from your overheads and how much risk the job carries.
How often should I update an AI rate card?
Whenever a merchant price or wage changes, and at least monthly for materials. Put the date at the top so you and the AI can both see how old the figures are.
Sources
- OpenAI Help: Projects in ChatGPT
- OpenAI Help: Uploading files and audio to ChatGPT
- Claude Help Center: What are projects?
Every rule, rate and date above was checked against these sources on the date shown at the top.
Topic: AI in the trades
Keep reading
AI in the trades
ChatGPT or trade-specific AI? What each is good at for a building firm
An honest split of what a general AI chat tool does well for a small trade firm and where a tool built around your office takes over. With a side-by-side table, a monthly cost comparison and a quick check on which you need.
AI in the trades
Why ChatGPT gets your quotes wrong: hallucinations, old prices and no live data
Why an AI chat tool's price for a building job is a guess, what that guess costs on a real bathroom refit, and a five-minute test and checklist to run before any AI-drafted quote goes to a client.
AI in the trades
Is it safe to give AI your email, Xero and customer data? (UK GDPR)
What UK GDPR expects when a trade firm connects AI to its email, Xero and customer records, what the main AI vendors say about training on your data, and a quick check and supplier checklist to run before you click connect.