What belongs in an AI consultant's first proposal
6 minute read. Updated 2026-08-08.
A first proposal should be two pages that name the decision the client has to make, the option you recommend, what it costs, and what happens if they do nothing. Twenty pages of methodology reads as a hedge against being wrong later. Clients hire people who commit to a position on page one, not people who defer the decision to an appendix.
The long proposal is a hedge
A 20-page proposal usually exists because the person writing it has not yet decided what they would do in the client's position, so they lay out every option and let the client choose between them. It looks thorough. It is actually a way of avoiding the one sentence that matters: here is what I would build, and why.
The client already knows the problem, roughly. They hired you to bring a decision, not a survey of the field. A director of data at a 200-person retailer does not need three pages comparing fine-tuning against retrieval-augmented generation. She needs to know which one you would build for her, and what that choice costs.
Length also signals uncertainty about price. Consultants pad proposals with methodology sections when they are not confident the number at the end will survive scrutiny on its own. A shorter document has nowhere to hide the number, so it forces the fee to be right.
Four things, two pages
The decision, stated as a sentence a non-technical board member could repeat back. Should we build a customer service assistant on our own ticket history, or buy a vendor tool and accept its limits. Not "AI strategy for customer service", which describes a topic rather than a choice.
The recommendation, argued in one paragraph, not a table of three options with ticks and crosses. If a genuine trade-off exists, show both sides of it in three or four lines and then still pick one.
The cost, tied to a specific piece of work with a specific end point, not to a day rate multiplied by a guess at effort.
The cost of doing nothing. What does the client lose by taking six more months to decide. Most freelancers skip this paragraph because it feels like pressure. It is not pressure if it is true, and it is the paragraph that gets proposals approved.
Price it as arithmetic, not a rate card
Quote the fee as a calculation the client can check, not a number that appeared from nowhere. A 10-day discovery and design phase priced against the median Machine Learning Engineer contract rate of £575 a day, as advertised in UK contracts over the six months to August 2026, comes to £5,750. Say it that plainly.
This does two things. It shows you are pricing against a market the client can look up themselves, which builds trust faster than a bespoke number would. It also protects you: if they push back on the total, the conversation becomes about scope, 10 days versus seven, rather than about whether your day rate is fair.
Do not present the advertised rate as what you will personally be paid. It is a reference point for how comparable work prices in the market. What you charge above or below it is your decision, made openly rather than buried in a total.
What to cut
Methodology sections describing how you generally work, agile ceremonies, a RACI chart. None of that tells the client anything about this engagement specifically. Keep it for the kickoff document, not the proposal.
A list of technologies you might use. If the recommendation is right, the technology choice follows from it and can be one line, not a slide with logos on it.
Your biography beyond a sentence. If this came through a shortlist process, the client has already read the reasoning behind your inclusion. Repeating your CV in the proposal wastes the two pages you have to make the argument.
When a longer document earns its place
Some procurement processes genuinely require a formal document with sections a compliance team expects: data protection approach, sub-processor list, insurance cover. That requirement is real, and fighting it wastes goodwill you need later.
Even then, put the one-page decision memo at the front, as a covering document. The person deciding whether to proceed reads that page. The compliance team reads the rest, separately, on their own schedule.
The two documents serve different readers with different jobs. Merging them into one file means neither reader gets what they need in the order they need it, and the decision gets delayed by paperwork that was never meant to block it.
The test for your own proposal
The risk of a short proposal is that it reads as thin if the recommendation underneath it is weak. That is a real cost, but it is the correct cost to bear: a weak recommendation dressed in 20 pages is still a weak recommendation, just harder for anyone to spot, including you.
Write the two pages, then read them back and ask whether you would sign the fee on the strength of the argument alone. If you would not, the fix is not to add pages. It is to change the recommendation.
What to do about it
- Open the proposal with the decision the client has to make, in one sentence.
- Recommend one option, not a menu, even where a genuine trade-off exists.
- Show the fee as arithmetic against a stated benchmark, not a number with no working shown.
- Cut methodology, technology lists and biography from the proposal document itself.
- State the cost of inaction as plainly as the cost of taking action.
- Where procurement needs a long document, put a one-page decision memo at the front of it.
Questions people also ask
How long should a first AI proposal actually be?
Two pages is the working target: one for the decision and recommendation, one for cost and next steps. Add supporting material as an appendix only if the client's procurement process explicitly requires it, and keep the decision memo as a separate page at the front so the person approving budget does not have to read past it to find the answer.
Should I include a full technical architecture in a first proposal?
No. Save it for after the client has agreed the decision and the fee, then send it as a technical annex or produce it in a working session. Detailed architecture attached to a first proposal usually distracts from the recommendation, and it tends to become redundant the moment scope shifts in the first meeting, which it almost always does.
What if the client's procurement team insists on a long document with fixed sections?
Give them the sections they require, but attach a one-page decision memo at the front stating the recommendation and cost plainly. The person who approves the budget is usually not the person checking the procurement boxes, and each of them should get the document shaped for how they actually read it.
Isn't a short proposal risky if the client doesn't know me yet?
It is more risky to bury the recommendation in pages the client will not read closely. A short proposal that states a clear position, with a fee tied to visible arithmetic, tells a new client more about how you think than 20 pages of general methodology ever will.
Where the figures come from
Every rate and salary quoted in this article is a median or percentile of figures advertised in UK job postings over the six months to 8 August 2026. They are not rates paid, and the gap widens at the top of a range.
- IT Jobs Watch, UK contract rates, 6 months to 8 August 2026, read 2026-08-08.
- IT Jobs Watch, UK permanent salaries, 6 months to 8 August 2026, read 2026-08-08.