How to Use AI to Write Customer Quotes

In most small service businesses the quote is the bottleneck, and everybody knows it. The site visit takes an hour. Writing it up takes another hour, gets pushed to the evening, then to the weekend, and the customer who asked on Tuesday hears back on Sunday. By then two competitors have replied.

AI fixes the write-up, not the visit. Done properly it turns sixty minutes of typing into about ten, and the quality goes up rather than down because you stop reusing a template from 2023 with the wrong client name in paragraph four.

There is one rule that makes the difference between a system that helps and one that eventually costs you money: AI writes the words, your rate card produces the numbers. Everything below follows from that.

Why the numbers stay out of it

Language models are good at prose and unreliable at arithmetic under pressure. Ask one to price a job and it will produce a confident total that is plausible, internally inconsistent, and occasionally wrong by a factor that ends an argument with a customer badly.

You do not need it to price anything. You already know your rates. What takes an hour is describing the work, scoping the exclusions, and formatting it so it does not look like it was typed on a phone.

So the split is:

The four-part setup

This takes an afternoon once, then it runs.

1. Write the rate card. One document, plain text, containing every unit you charge for and the price. Per hour, per square metre, per unit, per callout, per mile. Include your minimum charge and your typical markup on materials. This is the file the AI reads from and never invents against.

2. Collect three quotes you were proud of. Real ones, anonymised. These become the examples that set tone and structure. Three is enough. Ten does not help and starts producing averaged, characterless output.

3. Write your standard exclusions. The things that are never included and always cause disputes: making good after removal, disposal of waste, out-of-hours access, parking, discovery of asbestos or damp, anything behind a wall you have not opened. Most trades can list twelve from memory once they start.

4. Write the instruction once, reuse it forever. Something close to this:

You draft quotes for [business], a [trade] working in [area].

I will give you rough notes from a site visit.
Produce a quote with these sections:
1. What the customer asked for, in one short paragraph
2. Scope of work, as numbered items
3. Assumptions
4. Exclusions, drawn from the standard list below
5. Timescale

Rules:
- Never state prices, totals or hours. I add those.
- Use only information in my notes. If something is missing and it matters,
  list it under "Need to confirm" instead of guessing.
- Plain English. No sales language. The customer is a homeowner, not a buyer
  of professional services.

Standard exclusions: [your list]
Tone examples: [paste your three quotes]

That instruction is the asset. Store it somewhere you can reuse it rather than retyping it, which is the same discipline as building any repeatable business context for AI.

The workflow on a live job

1. Dictate on site or in the van. Voice notes into your phone. Do not tidy them. "Kitchen, 4 by 3.5, existing units out, plaster looks sound on the north wall, socket needs moving, customer wants it done before the 20th."

2. Paste the notes in. With the standing instruction above, you get a structured draft in under a minute.

3. Read the "Need to confirm" list. This is the part that pays for the whole system. It reliably catches the thing you forgot to ask, while you can still ring the customer about it.

4. Add the numbers yourself. From the rate card, in your spreadsheet or quoting tool.

5. Read it once, out loud if it is a large job. You are checking for anything the AI asserted that you did not say. This is not optional and it does not take long.

6. Send.

Same day, usually within the hour if you do step two from the van.

A worked example

Notes dictated in a van, verbatim:

bathroom refit, upstairs, 2.1 by 2.4. rip out bath basin toilet.
tiles floor to ceiling on two walls, half height other two.
existing soil pipe boxed in, not opening it. customer supplying
sanitaryware. wants it before her mum visits mid october.
no parking on the street, permit needed.

What comes back, after the standing instruction:

What you asked for: A full refit of the upstairs bathroom, roughly 2.1m by 2.4m, removing the existing bath, basin and toilet and retiling, with sanitaryware supplied by you.

Scope of work: six numbered items, each traceable to a phrase in the notes.

Assumptions: Existing soil pipe boxing remains closed and in serviceable condition. Water supply and waste in current positions. Access available during standard working hours.

Exclusions: Supply of sanitaryware. Making good to ceilings beyond the tiled area. Removal of waste beyond one skip load. Any remedial work discovered behind the existing boxing.

Need to confirm: Tile choice and supplier. Whether the parking permit is arranged by you or by the customer. Exact date required in October.

Timescale: Five working days once materials are on site.

That "need to confirm" block is three phone-call questions you would otherwise have discovered in week two of the job. The parking permit alone is the kind of detail that costs half a day and a difficult conversation when it surfaces late.

What goes wrong

Invented specifics. The most common failure. You said "existing units out" and the draft says "including removal and disposal of existing units to a licensed facility". Disposal was never agreed. This is why exclusions are a fixed list rather than generated, and why step five exists.

Quiet scope inflation. Drafts tend toward thoroughness, which reads as professional and prices as expensive. If your quote is consistently coming back longer than the job, tighten the instruction to say scope items must map one to one with the notes.

Sounding like a corporation. A sole trader whose quotes suddenly read like a consultancy proposal loses the thing that won the work. Your three examples exist to prevent this. If drafts still drift, add "match the voice of the examples closely, including their plainness" to the instruction.

Customer data. Site notes contain names, addresses and sometimes access details. Know where that goes and whether the tool retains it. Worth ten minutes with how to check whether a tool trains on your data before you paste in your first real job.

What it is actually worth

The obvious gain is time: roughly forty-five minutes per quote, which for a business doing eight quotes a week is most of a working day recovered.

The real gain is response speed. In trades and local services the first credible quote wins a large share of jobs, not because it is cheapest but because it arrived while the customer was still thinking about the problem. Cutting turnaround from four days to four hours changes your win rate more than any pricing adjustment will.

The second-order gain is consistency. Quotes written at 10pm on a Sunday are worse than quotes written by a system, and the gap shows up as disputes months later over what was included.

If it works, keep going

Quoting is usually the first bottleneck in a service business, not the only one. Once the quote flow is running, the same pattern of fixed rules plus generated wording applies to follow-ups, job sheets and invoicing. Writing the process down as you go turns it into something you can hand to a new hire, which is exactly what a good SOP is for.

One caution on scaling it up. The system works because a person who understands the job reads every draft before it goes out. A quoting process that sends automatically, with nobody checking the "need to confirm" list, becomes a machine for issuing quotes with invented scope. Keep the human step even when volume makes it tempting to drop.

And if you are wondering whether to sell this setup to other businesses rather than just run it: that is a real business, and pricing an automation project is where it starts going wrong for most people. The wider map of what small businesses are automating first is in our guide to AI for small business.