How to use AI to draft a quote from your past quotes
To draft a quote with AI, you give it a small library of your own past quotes, then describe the new job and ask for a draft built from the closest matches. What comes back is your structure, your section order, and your usual wording, already filled in around the new job. What does not come back is the price. You set that yourself, every time, and you read every line before it reaches a client.
The reason this is worth doing is that the slow part of quoting is rarely the thinking. It is starting from an empty document, hunting for the last similar job, and trying to remember how you worded the exclusions. That part is copying, and it is the part a tool can do in seconds while you keep the judgment.
Your past quotes are the input that matters
A generic model asked to write a quote will produce something generic: plausible headings, filler terms, wording that belongs to no particular business. Give it five to fifteen of your own quotes and the output changes character completely, because it now has your section order, your level of detail, your phrasing for scope and exclusions, and your standard terms. You are not teaching it to quote. You are giving it your house style to copy.
Pick the library carefully. Choose quotes that represent the work you actually want more of, covering a range of sizes. Strip out client names and anything confidential, because these files will sit in a tool. If you know which ones were accepted, say so, and lean the library toward those. A folder of every quote you have ever sent is worse than a curated handful, since half of it will be work you no longer do in a format you no longer use.
Exhibit 1
History supplies the shape of the quote. You supply the price and the scope.
What the tool needs to be
Nothing exotic. The tool class is a general assistant that accepts documents and can hold a decent amount of text at once. Most of the widely used ones can do this today. You upload or paste the library, keep that conversation, and start each new quote from it. There is no system to build and no integration to commission.
If quoting is a daily job rather than a weekly one, the natural next step is a form that collects the job details and returns a draft into your document folder. That is a small piece of everyday automation, and it is worth doing only once the manual version has proven the library is good. Build it earlier and you will be automating a draft nobody trusts.
The prompt approach
Describe the new job the way you would to a colleague: what the client wants, the rough size, the constraints, anything unusual, and the deadline. Then ask for four things. Draft this quote using the structure and wording of my past quotes. Use the two or three closest examples and tell me which ones you used. Leave every price and every rate blank. List separately anything you had to assume about the job.
That last request is the one people skip, and it is the one that earns its keep. The assumptions list is where the model tells you what it did not know, which is usually the same set of questions you should have asked the client before quoting at all. Reading it is often more useful than reading the draft. Asking which past quotes it drew from is the second most useful line, because it lets you sanity-check the match in a couple of seconds rather than reading the whole thing suspiciously.
Pricing is not a drafting job
Keep the numbers out. A model asked to fill in a price will produce one that looks reasonable and is anchored on whatever it saw in the library, which means last year's rate on a job that is not really the same job. It has no view on your capacity this month, what the relationship is worth, or how much of this work you want. Blank fields are the safe default, and they are also faster, because a blank field gets filled in deliberately while a plausible wrong number gets skimmed past.
Exhibit 2
Some parts of a quote are safe to reuse. Others have to be decided again every time.
What to check by hand
Read the scope first and read it as a client would, looking for anything that could be understood more broadly than you mean it. Then the exclusions, because a draft assembled from other quotes tends to inherit exclusions that fit the old job and miss the one that matters here. Then the timeline, which the model will have borrowed from somewhere. Then the client name and the job description in the opening, which is where a copied detail from the wrong project is most likely to survive. Finally, add your prices and check the arithmetic yourself.
None of this takes long, but it is not optional, because a quote is an offer. What you send is what you have agreed to do. The parts worth getting right go well beyond accuracy: what a quote says, how it is ordered, and how easy it is to say yes to all affect whether it comes back signed, which is the subject of writing quotes and proposals that get answered. A draft assembled from history is a head start on the typing, not a substitute for that thinking.
When it will not help
A genuinely new kind of job has no closest match, so the draft will be an assembly of things that only look relevant. That is more dangerous than a blank page, because a confident-looking document invites you to skim. When the work is unfamiliar, write the first one yourself, then add it to the library so the next one has something honest to copy from. The same applies after you change your pricing model or your terms: update the library before you draft again, or the tool will keep faithfully reproducing the old ones.
Used within those limits, this is one of the plainest wins available to a small business, and it sits alongside a handful of similar jobs on the list of tasks AI can genuinely take over. The pattern is always the same. The tool does the first pass from your own material, and a person makes the decisions that carry a commitment.