How to Use AI for Quotes and Estimates: A Practical Guide for Small Businesses

Ask any small business owner what eats their evenings and "writing quotes" comes up within the first three answers. A plumber pricing a bathroom refit, an agency scoping a three-month retainer, a caterer costing a 120-person wedding — the job is the same shape every time. Pull the details out of a messy enquiry, work out what it will take, attach numbers, write it up politely, send it, chase it. Most owners do this after hours, for work they have not yet won and may never be paid for.

AI will not price your work for you. It should not. But it can absorb most of the hours around the pricing decision, and that is where the win is. This guide covers a concrete workflow for using AI in quoting and estimating, the prompts that produce usable output, and the guardrails that stop a language model quietly destroying your margin.

Why quoting costs more than you think

Run the arithmetic on your own business for a moment. If a quote takes 45 minutes to produce and you send 12 a month, that is nine hours — more than a full working day, every month, on unpaid administration. At a modest internal cost of €40 an hour that is roughly €4,300 a year. If you win one quote in three, two thirds of that money bought you nothing except a polite "we've gone another way."

The cost is not only time. Slow quotes lose work: the first credible quote to land has a disproportionate chance of winning the job, simply because the buyer stops looking. And tired quotes written at 10pm are where pricing mistakes live — the forgotten disposal fee, the travel time nobody costed, the "we'll sort that out later" line item that turns into three unpaid days.

What AI can and cannot do in a quote

Be precise about the division of labour, because this is where most people go wrong.

AI is good at: reading a rambling enquiry email or call transcript and pulling out the scope; spotting what the customer has not told you and listing the questions you need answered; assembling line items from a rate card you provide; drafting the written quote in your tone; producing three pricing options from one set of costs; comparing a new job against similar past jobs you have done; and writing the follow-up sequence.

AI is bad at: knowing what your labour actually costs, knowing local material prices this week, judging how much a particular client will tolerate, and knowing which jobs are quietly a nightmare. A language model asked to "estimate a price" will produce a confident number drawn from nothing. That number is worse than useless because it looks authoritative.

The rule that keeps you safe is simple: AI structures the quote, your rate card prices it. Every figure in the output should be traceable to a number you supplied — a day rate, a unit cost, a markup percentage, a historical job. If the model invented it, it does not belong in the document.

The four-step AI quoting workflow

This workflow takes about a fortnight to set up and then saves most of those 45 minutes, permanently.

Step 1: Build a rate card the AI can read

Before anything else, write down what you charge, in a plain document or spreadsheet. Labour rates by role and by hour or day. Standard materials with current unit costs. Call-out and travel charges. Your markup on materials. Minimum job value. Typical durations for your ten most common jobs. Discount rules and their limits.

Most owners carry this in their head, which is precisely why their quotes are inconsistent and why nobody else in the business can produce one. Keep it in one file, date it, and update it when your costs move.

Step 2: Extract the scope from the enquiry

Paste the customer's email, your call notes, or a transcript into the assistant and ask it to produce a structured brief: what the customer wants, where, by when, any constraints, and — critically — a list of the unknowns that would materially change the price. This takes about 30 seconds and routinely catches the thing you would have discovered on site.

Step 3: Assemble the costed quote

Give the model the brief, the rate card, and the format you want. Ask it to build the line items, apply your rates, show the arithmetic, and flag any assumption it had to make. Then — and this is not optional — check the maths yourself. Models are much better at arithmetic than they were, but a transposed digit in a quote is your problem, not theirs.

Step 4: Write it up and set the follow-up

Once the numbers are right, have the assistant produce the client-facing document in your voice, plus a covering email and two follow-ups scheduled for day four and day ten. The follow-ups are where the quietly significant money is: jobs get won because somebody chased and the competitor did not.

Prompts that actually work

Generic prompts produce generic quotes. These three are the backbone of the workflow — adapt the specifics to your trade.

Scope extraction:

"Here is an enquiry from a prospective customer. Produce: (1) a structured scope of work in bullet form; (2) everything the customer has specified about timing, location, and access; (3) a list of every unknown that could change the price by more than 10%, written as questions I can send back. Do not estimate any prices. If something is ambiguous, say so rather than assuming."

Quote assembly:

"Using only the rates in the attached rate card, build a line-item quote for the scope below. For each line show quantity, unit, unit cost, and total. Apply a 15% materials markup and add the standard call-out charge. Show the subtotal, VAT at 21%, and the total. List separately any assumption you made, and flag anything in the scope that my rate card does not cover. Do not invent any price that is not in the rate card."

Three-option pricing:

"Turn this costed quote into three options — essential, recommended, and complete — keeping the same underlying rates. The recommended option should be the one I would advise. Explain in one sentence per option what the customer gets and does not get."

That last one is worth dwelling on. Presenting a single price invites a yes-or-no decision. Presenting three changes the question to "which one," and the middle option is chosen far more often than the cheapest. The tiered structure is an old consulting trick; AI makes it cheap enough to do on every job rather than only the big ones.

Pricing guardrails: keep AI away from your margin

Four rules, learned from the businesses that got this wrong first.

Set a floor and enforce it outside the model. Put your minimum acceptable margin in a spreadsheet formula, not in a prompt. A prompt is a request; a formula is a constraint. If a quote drops below the floor it should fail visibly before it reaches the customer.

Never let AI apply discounts. Ask for a discount and a language model will happily give one, because agreeable text is what it is optimised to produce. Discounting is a commercial decision with information the model does not have.

Always show the workings. Require quantity, unit, and unit cost on every line. Quotes that arrive as a single number cannot be checked, and uncheckable quotes are where errors hide.

Keep a human signature on every quote that leaves the building. Not a review in principle — an actual person looking at the total and deciding it is right. On anything unusual, read it properly.

Common mistakes with AI-generated quotes

Pasting customer data into a free consumer account. Quotes contain names, addresses, and sometimes commercially sensitive detail. Use a business or team tier where your inputs are not used for training — it matters more still if you handle anything counting as personal data under the GDPR.

Letting the quote sound like a robot wrote it. Customers notice. Feed the model three of your own past quotes as style examples and tell it to match them — tone, length, level of detail, how you describe your work.

Automating the enquiry-to-quote path end to end. It is tempting to connect the contact form straight to an automated quote. Resist it for anything beyond a genuinely standardised product. The failure mode is silent, fast, and expensive.

Forgetting to update the rate card. Material costs move. An AI quoting system built on last year's prices produces consistent, well-formatted, loss-making quotes at speed.

How to measure whether it worked

Track four numbers for 90 days and you will know whether this is working, rather than feeling like it is.

Time per quote. Time five quotes before you start and five after. The honest target is 45 minutes down to 10 or 15, not zero.

Time to first response. Measure from enquiry received to quote sent. This is the number most likely to move your win rate.

Win rate. Quotes won divided by quotes sent, measured the same way before and after. Watch for the trap: if quoting gets cheap you will quote for more marginal work, and your win rate can fall while your revenue rises. Track both.

Realised margin versus quoted margin. The one most owners skip. If jobs are consistently coming in below the margin you quoted, your rate card is wrong — and AI has simply been reproducing that error faster and more neatly.

If you want a broader framework for tracking this kind of change, our guide to AI adoption metrics for small businesses covers which numbers are worth watching and which are vanity. The pricing decision itself sits upstream of all of this — how to price services with AI deals with setting the rates that your quoting workflow then applies, and using AI for proposal writing covers the longer, narrative documents that follow a quote on larger deals.

The bottom line

Quoting is the highest-volume writing task in most small businesses and the one with the clearest return on automation — provided you automate the right half of it. Write down your rates, let the assistant handle extraction, assembly, and drafting, and keep every pricing judgement firmly in human hands. You reclaim most of a working day each month, you reply the same afternoon instead of the following week, and your quotes stop varying with how tired you were. That consistency is worth as much as the time saved.

Start this week with one thing: open a document and write down your rates. Everything else in this guide depends on it, and it is the step people skip.