Proposals & Fee Letters

Think about what actually goes into a good proposal. There is a cover note that sets the right tone, a paragraph or two showing you have understood the brief, a scope of work broken down by stage, a list of what is included and - just as important - what is excluded, a note on your team and approach, a deliverables list, a programme in words, and the terms boilerplate every practice reuses. Most of that is structure, courtesy and completeness: getting the shape right, striking a warm but professional tone, and not forgetting the exclusion that saves you an argument later. This is precisely the work Claude is excellent at. It never forgets a section, it writes a clean sentence on the first try, and it will cheerfully produce the same document in three registers so you can pick.

But notice what sits inside that shape: a fee, a set of dates, and a promise about what you will deliver. Those are not writing - they are commercial and professional decisions, and they carry real weight. If Claude invents a fee percentage because you did not give it one, the document will not look wrong; it will look completely convincing, which is the trap. A plausible number in a proposal is worse than a blank, because a blank you notice and a plausible figure you might sign.

So the mental model for this lesson is the one that runs through the whole course, sharpened for money: Claude assembles the document; you own every commitment inside it. Let it build the vessel at speed, and pour in the numbers, scope and terms yourself - deliberately, and checked. The figure on the right shows the split: a large safe zone Claude can draft with a light touch, and a small, non-negotiable zone that is yours alone. Get that division clear in your head and the speed becomes pure gain rather than a hidden risk - because you know, on every proposal, precisely where to slow down and read with a pen in hand.

The quality of the proposal is set almost entirely by what you feed Claude, not by clever wording afterwards. A good proposal prompt carries five things: who you are and who the client is, the scope in your own bullets, the fee and payment basis that you have already decided, the tone you want, and - if you have one - a past proposal to match. Give it all five and the draft comes back close to sendable. Give it a one-line request and it will fill the gaps with invented plausibilities.

Here is a directing prompt that keeps you in charge of everything that matters:

You are helping a small architecture practice draft a fee proposal.
Do NOT invent any figures, dates, percentages or commitments - use
only what I give you, and write [TO CONFIRM] wherever a number is
missing.

Client + project: <two lines>
Scope, in my words: <your bullets, by stage>
Fee (decided): <your number and basis>
Payment stages: <yours>
Exclusions to state: <your list>
Tone: warm, professional, concise; we are not the cheapest and
do not pretend to be.
Match the voice of the past proposal I am pasting below.

Draft the proposal with clear headings. Flag anything you think
is missing as a question at the end - do not fill it in yourself.

Two things make this prompt work. First, the explicit instruction to never invent, and to mark gaps with a token like [TO CONFIRM] rather than guessing - this turns Claude's biggest weakness into a checklist for you. Second, asking it to end with questions rather than assumptions: a good proposal often exposes something you have not decided (are prints reimbursable? who pays the authority fees?), and you want that surfaced, not silently resolved. Iterate from there - ask for a shorter version, a warmer opening, a firmer exclusions paragraph - the way you would guide a capable junior through a second and third pass.

Across every proposal and fee letter, three categories must originate with you and only you, and each fails silently if you let it slide.

Numbers. Fees, percentages, rates, lump sums, reimbursable caps, tax and payment amounts. Claude has no idea what your practice costs to run or what your market bears; anything it offers is a guess dressed as a figure. Decide the number yourself, type it in, and check that the one in the final document is the one you intended - it is astonishingly easy for a figure to drift across a redraft.

Terms and commitments. What you will deliver, by when, and under what conditions; liability, indemnity, copyright, termination, dispute wording. These are contractual. Claude can format them and can reuse your own approved boilerplate, but it must never author a new term or soften a liability clause, because a plausible-sounding clause can quietly give away something you would never concede. Anything with legal weight goes past you, and for real contracts, past a lawyer.

Facts about you. Credentials, registration, past projects, team CVs, awards, insurance. Claude, reaching for a strong proposal, may confidently state that your practice completed a project it never touched, or hold a certification it does not. This is not lying; it is the model producing the most impressive plausible sentence. You supply and verify every factual claim about your own practice - a fabricated credential in a proposal is a serious professional problem, not a typo.

The habit that protects you is simple: before anything goes out, read the document once as the client - is it clear and courteous? - and once as your own auditor, pen on every number, date, term and boast, asking of each: did I decide this, and is it right?

It helps to understand why these three fail silently while a grammar mistake does not. A typo announces itself; a plausible fabrication is designed, by the very nature of the model, to look exactly like the real thing. Claude is not trying to deceive you - it has no intent at all - but it is optimised to produce text that reads as correct, which means its errors are the hardest kind to spot: confident, fluent and in the right shape. That is precisely why the defence cannot be "I will notice if something looks wrong." Nothing will look wrong. The defence has to be structural: you know in advance that numbers, terms and facts about you are the three zones that must be traced back to a source you control, and you check them every single time regardless of how right the draft appears. Discipline, not vigilance, is what keeps a fabricated figure or an invented credential out of a document that carries your seal.

A proposal that reads beautifully and commits you to the wrong thing is worse than an ugly one that is correct. So the last step is never Claude's - it is a deliberate review pass you run every time, and it takes far less time than the drafting you just saved.

Work through four checks. Numbers against your own source: open the fee note or the spreadsheet where you actually decided the money, and confirm every figure in the proposal matches it - fee, stages, percentages, caps, taxes. Never check a number against the proposal itself; check it against where you set it. Scope against fee: read the scope of work and the fee side by side and ask whether they balance - has Claude, in making the scope sound generous, quietly promised more than the fee covers? Overstated scope is the commonest way a well-written proposal loses you money. Exclusions and assumptions present: confirm the document says clearly what is not included and what you have assumed; a missing exclusion is an argument waiting to happen. Commitments you can keep: every date, deliverable and promise - are they real, and yours?

Then the tone read: does it sound like your practice, or like a generic AI proposal? Strip the giveaways - the throat-clearing opening, the three-adjective flourishes, the phrase we are delighted to have the opportunity that every AI draft reaches for. A client who receives ten proposals can feel a machine-written one, and it undercuts exactly the trust a proposal exists to build.

One more habit is worth building in: check the whole document, not just the parts you edited. Numbers and clauses have a way of drifting across redrafts - you correct a fee in one place and Claude, generating the next version, quietly restates the old one in the payment table, or a stage you cut reappears in the deliverables list. Read the final version end to end as if you had never seen the drafts, because the client will. The goal is a document that is faster to produce, more complete than you would manage by hand, and unmistakably, accountably yours.