AI-drafted quotes from site-visit notes and photos
Short answer: Capture the visit in a fixed template (measurements, notes, a voice memo and labelled photos), then give an AI model that material plus your own price list and ask it to draft the scope and line items using only items from that list. It should flag anything it can't price instead of guessing. You check the draft against a short review list, adjust, and send it from your usual quoting tool. The AI saves typing; the measurements, judgement and prices stay yours.
Most tradespeople don't lose jobs on price. They lose them because the quote went out four days after the visit, by which time the customer had signed with whoever replied first. Writing up a clear scope at 9pm is the bottleneck. AI is good at turning rough notes into tidy, well-ordered text, which makes it a reasonable helper here, provided it never makes up a measurement or a rate.
Below is a worked, fictional example.
The visit (hypothetical)
Dan runs a two-person kitchen fitting business in Leeds. On Tuesday he visits a semi-detached house: the customer wants an existing kitchen stripped out and a new flat-pack range fitted, with a new worktop and tiled splashback. She's buying the units herself.
Step 1: Capture on site with a template
Dan uses the same note template on his phone for every visit, so nothing depends on memory.
Site visit template
- Customer, address, date, access notes (parking, stairs, pets)
- Job type and what the customer said they want, in their words
- Measurements, each labelled (wall A, wall B, ceiling height, window position)
- Existing condition (what's being removed, any damp, uneven walls, old services)
- Who supplies what (customer-supplied units? worktop?)
- Exclusions discussed
- Photos, each named: "wall-A-full", "under-sink-pipework", "socket-by-window"
- Voice memo: a two-minute walk-round summary
The labels matter. OpenAI's vision guide notes (as of November 2026) that models don't process original file names or metadata, struggle with precise spatial tasks and may give only approximate counts of objects. That's why measurements are typed, not "read" from photos, and why each photo is described in the notes.
Step 2: Transcribe and tidy
Back in the van, Dan transcribes the voice memo with his phone's dictation or an AI transcription tool and pastes it under his notes. He reads it once to fix misheard words ("worktop", not "work top") and wrong numbers.
Step 3: Give the AI your price list, not permission to guess
Dan keeps a simple price list as a spreadsheet: item code, description, unit, his rate. Things like "strip out existing kitchen (per linear metre)", "fit base unit (each)", "cut and fit laminate worktop (per metre)", "tile splashback (per m²)", "waste removal (per load)". The rates are his own and don't appear here.
He pastes three things into the AI tool: the notes, the photos, and the price list. Then this prompt:
You are drafting a quote for a UK kitchen fitting business.
Use ONLY items and rates from the price list provided.
If something in the notes needs work that isn't in the price list,
list it under "Needs pricing" and do not invent a rate.
Use only measurements written in the notes; never estimate from photos.
Output:
1. A plain-English scope of works (what we will do, in order).
2. Line items: item code, description, quantity, unit, rate, total.
3. Exclusions and assumptions.
4. Questions to confirm with the customer.
If you connect this to your own system through an API, a structured output feature keeps the line items in a fixed format. OpenAI's documentation describes Structured Outputs as making responses follow a JSON Schema you supply, which is useful if the draft is going straight into a quoting tool.
Step 4: Read the draft
The AI's draft for Dan includes a sensible scope, about a dozen line items, and three useful flags:
- Needs pricing: "Reposition socket by window." There's no electrical item on Dan's list, correctly flagged.
- Assumption: "Worktop supplied by customer?" The notes said the units were customer-supplied but didn't mention the worktop.
- Question: "Is the existing tiling to be removed or tiled over?"
It also got one thing wrong: it multiplied the worktop length using wall A plus wall B without allowing for the corner. Dan catches it on review.
Step 5: Review checklist
Dan runs every draft through the same checks before it goes anywhere.
| Check | What to look for |
|---|---|
| Every quantity traces to a written measurement | No numbers that came from "the photo shows" |
| Every rate matches your price list | No new rates, no rounded figures |
| "Needs pricing" items are priced by you or removed | Including specialist work you'd subcontract |
| Scope matches what the customer asked for | In their words, not upsold or trimmed |
| Exclusions are explicit | Electrics, plumbing, plastering, making good, waste removal |
| Customer-supplied items are named | So nobody assumes you're buying the units |
| Arithmetic re-checked in the quoting tool | Let the software total it, not the AI |
| Tone and spelling | UK spelling, no American "estimate" phrasing if you call it a quote |
Here, Dan prices the socket move using his electrician's figure, corrects the worktop length, confirms by text that the customer is supplying the worktop, and adds "existing tiles to be removed" after a quick call.
Step 6: Send through your quoting tool
Dan doesn't send the AI's text as an email. He copies the scope into his quote template and enters line items from his saved items, so totals, numbering and branding come from his accounting software. In Xero, for instance, quotes are created from the Sales menu, and Xero's help says you can email the customer a link to an online quote they can accept or decline. Dedicated trade quoting apps have their own version of this step.
Sent the same evening, with photos of the areas discussed attached.
Where AI fits, and where it doesn't
Good at: turning notes into an ordered scope; spotting gaps between notes and price list; drafting exclusions and customer questions; keeping wording consistent.
Not good at: measuring from photos; knowing your costs; judging site conditions; counting fixtures accurately; being accountable for a price.
On data: only share customer details your AI tool's terms allow, and strip anything you don't need.
FAQ
Will customers know the quote was drafted with AI?
They'll see your scope, your prices and your template. What matters is that it's accurate and specific to their job.
Can AI estimate materials quantities?
Only from measurements you've written down, and you should re-check the arithmetic. Don't rely on it to measure from photos.
What if my prices change?
Update the price list file. Because the AI only uses that list, the next draft uses the new rates.
Does this work for small jobs?
For a single-item repair it's probably faster to type the quote. The workflow pays off on multi-trade jobs with long scopes.
Quoting faster without losing accuracy is mostly about a good template and a firm review habit, and we can help set up both for your business.