Automate your sales proposals
without putting words in your clients’ mouths.

01The task, as it is done today

Take a small business somewhere between Bastogne and Bertrange. The first meeting is on a Tuesday, at the prospect’s premises. You come away with two pages of notes, a photo of a whiteboard and a promise to send a proposal by the end of the week. It goes out the following Monday, written on Sunday evening, reworked from the most recent proposal sent to another client.

It is complete. It was not written for the client who receives it.

And the person writing it is not always the one who ran the meeting. Sometimes it is the managing director, working from notes someone else took. The previous client’s name has been replaced everywhere except once, in a paragraph halfway through. The priorities follow the order of the brochure, not the order the client gave them in. The client knows what they said. They spot it on the first page.

02Can AI write my sales proposals?

AI drafts the proposal from your notes and links each point to one of your services, but it is not allowed to put words in your client’s mouth, it does not set the price, and it sends nothing.

A general-purpose assistant will already turn three lines of notes into four tidy pages. That is exactly where the trouble starts: whatever it adds to reach four pages, it makes up.

The full circuit, as it is built in a company of three to fifty people, has twenty-four stations. At five places it hands control back to a person, and only five of the twenty-four stations involve a model.

03From the first meeting to the proposal going out

Eight zones, and a final node that always belongs to the same person: whoever met the client. Click a zone to open its stations; otherwise, the sequence plays on its own.

04The twenty-four stations, and who does what

The circuit unfolded, station by station, with whoever owns each one. The drafting takes up a single zone, the fifth, between what prepares it and what checks it.

05What the AI does, and what it never does

Five stations out of twenty-four. The model judges whether your notes are enough, separates what the client said from what you inferred, drafts the summary around the points they stressed, links each of them to one of your services, and flags what it is unsure about. A proposal is writing, and writing is the kind of task people readily picture handing over to a model entirely. The tally does not take the drafting away from the model. It says the drafting is the core, not the whole.

Three hard limits, the same for every proposal. It puts no words or figures in your client’s mouth: something inferred can be used, but it is never written as something the client said, and a sentence that points to no line in your notes stops the draft. It does not set the price: the draft leaves it blank, and the figure comes from your quote or from a person. Pricing itself belongs to another circuit, the one for your quotes. It sends nothing: the draft goes to whoever met the client, and they send it themselves, with their own covering message. A proposal is a letter from someone, not a notification.

Besides Claude and Make, the circuit uses what you already have: your client records or CRM, the voice recorder on your phone, your email, your calendar, your word processor and your document template. I connect to them, I do not replace them. Two limits are worth knowing before you start. How well a photo of a notebook can be read depends on the handwriting, and whatever is illegible is not guessed: it goes back to a person. And many small businesses have no CRM: the prospect’s record is then a folder and a spreadsheet. The circuit adapts, but it has no history to read before it writes.

06What cannot be bought, and what holds everything together

Three pieces that neither the scenario nor the model produces. Your library of services: what you actually sell, described once, with what is included, what is not and the terms, plus the references you are allowed to cite, with the agreement of the client concerned. It is not your website or your brochure; it is a table every proposal draws on. Your proposal outline, and its voice: the structure you follow, from the client’s situation to the next steps, and three or four winning proposals chosen as examples of tone. And the rules of commitment: what is never promised, such as a quantified result, an unchecked deadline or exclusivity; which references may be cited, and to whom; the amount above which you read the proposal yourself; how long notes are kept.

They are built with you, because they are made of what you know and have never written down. No template “filled in by AI” supplies them: they describe what you sell and what you refuse to promise, and no tool knows that for you. Personalising a proposal starts with knowing what you sell.

Add to those one field on the note-taking form: “said in confidence”. Whatever the client said in confidence stays there: it never goes into a proposal and is never sent to the model. What never goes into the model cannot come out of it.

07What can go wrong

Four failures. Two come from the drafting, one from the person doing the final read, and one from time passing.

Another client’s name. The proposal was started from the previous one, and one paragraph still names the client it was written for. Once it has been sent, there is no taking it back. The draft is checked against your client records before anyone sees it. A client whose name is also an everyday word may stop the draft needlessly, and that is better than the reverse.

Something the client never said. The notes read “budget?”, and the proposal says “you told me your budget was tight”. The client never said so, and knows it. Hence the split, as soon as the notes are read, between what they said and what was inferred, and the check that stops any sentence without a source. That check confirms that the sentence points to a line of your notes and that any words quoted appear there. It does not judge whether the sentence is right. The final read decides that, which is why it is never skipped.

The final read becomes a glance. After a few good drafts, people skim, then send without reading. That is the human failure, and no check removes it. What limits it: the draft arrives with the doubtful points at the top, the inferences, the points with no matching service, the open questions, so that even a glance lands on them. And above the amount set in the rules, the managing director reads it too.

The library of services goes out of date. A service dropped, a scope revised, a client withdrawing permission to be cited, and nobody updates the table. The proposal is still well written, and it offers something you no longer sell. The library is kept up to date like your catalogue, by someone whose job it is, and the gap captured after each proposal goes out shows what had to be corrected by hand.

08How long it takes, and what it costs

This page gives no timescale, and that is deliberate. The timescale depends less on the scenario than on what you already have: notes taken on a shared form or by each person in their own way, a list of services written down or still in your head, winning proposals to reread or to track down. Two or three of your proposals would tell me, and I have not seen them yet. The audit sets the timescale, and the price with it: it is free because, before it, I cannot put a figure on anything.

We start with a single case, the proposal that follows a first meeting, and nothing else until that one holds. The frequently asked questions cover your data, the cost and the timing, and my page says where I come from.