AI for quotes and proposals: from the site visit or the brief to a priced document the same day.

Input: whatever the firm captures today, voice notes from the van, photos, a measurement sheet, the client’s brief, a call transcript. The system extracts what the quote needs (scope, quantities, options, constraints), matches items against the price book and past quotes for similar jobs, computes the numbers with ordinary arithmetic, and drafts the document in the firm’s template and tone with the sections a customer expects: what is included, what is not, assumptions, timescale, terms. The person who did the visit reads it on their phone, corrects the scope, approves, and it is sent from the firm’s usual system with the CRM updated. Same day rather than next week.

Trades and construction, where the estimator is on site all day and quotes are written at night: a voice note and six photos become a structured, priced estimate for review. Agencies and consultancies, where proposals are assembled from past ones and a brief: the draft arrives with the right sections and case studies, and the senior person spends their hour on the thinking rather than the formatting. Brokers and advisers, where the comparison and the recommendation follow a pattern: the document is drafted from the case file, and the adviser edits the judgement. Lettings and property, where valuations and proposals are structured and frequent. In each case the win is the same: the firm’s expertise, delivered while the customer is still deciding.

A language model must never set a price. The rules are simple and they are the design. Prices and rates come from the firm’s price book, which the system reads and cannot change. Quantities extracted from notes or photos are shown to the approver next to their source. Totals, VAT and margins are computed in code, not written by the model. Anything not in the price book is flagged as “to be priced” rather than estimated. The draft carries a visible list of assumptions. And nothing is sent until a named person approves it. With those in place, the failure mode is a wrong quantity a person catches, not a wrong price a customer catches.

A first build, one quote type, one price book, one sending system, an approval step on the phone, is an implementation at £4,000 to £15,000 fixed at UK rates, with the running cost per quote stated before go-live, which on a small model is pence. Off-the-shelf quoting tools exist for standard trades and are worth checking first; custom is right when the firm’s pricing logic, document format or systems are its own. Several quote types, or proposals assembled from a library with a knowledge assistant behind them, is a custom system at £8,000 to £25,000.

Baseline two weeks: quotes requested, hours from request to sent, hours spent writing, win rate. Go live. Compare at thirty and ninety days. Time to quote should fall from days to hours; the writing hours should fall by more than half; and the win rate, which is the number that matters, usually rises because the quote arrives first. Keep the counts and the win rate in the same spreadsheet as the enquiry-response numbers, because they are one funnel.