Using AI to write proposals works when the AI drafts and your team decides. Give it material you have already approved, have it produce a structured first draft that flags what it does not know, and have a named owner sign off scope, pricing, evidence, and commitments before anything goes to the prospect.
That is the whole workflow. The hard part is not the prompt. It is deciding what goes in, what the draft is allowed to leave blank, and who signs off. Most agencies never decide, because each proposal gets written against its own deadline by whoever holds the account.
An illustrative example: a 30-person creative agency has three good discovery calls, and three proposals are due the same week. The strategy lead assembles all three. One starts from a rebrand pitch for a retailer, one from a campaign proposal for a nonprofit, one from whatever turned up in the shared drive. Each pulls in a different version of the service descriptions, so the same deliverable goes out described three ways. None of that is judgment. It is assembly, and assembly is where AI helps, once the inputs are settled.
How to use AI to write a business proposal, in three steps
This is the workflow on our agency page, shown there as illustrative, not as a client result. It runs from discovery call to reviewable draft.
1. Approved inputs
The draft is built from four things you already trust: the discovery notes for this prospect, your proposal template, your service descriptions, and approved examples of past work. Nothing else.
That last line does most of the work. Ask a general AI tool to "write a proposal for a brand refresh" and it will. It will describe services you do not sell, at a scope you never priced, in a voice that is not yours. Give it your template and your service descriptions and it can only rearrange what you have already signed off. The draft starts from your offer, not an AI's idea of it.
If those four inputs live in six folders and two people's heads, gathering them is the first job. You do it once. Every proposal after that draws on the same set. Building a proposal content library from past work covers that first job, step by step.
2. A structured first draft
The draft comes back in four parts: the prospect's objectives, the proposed approach, the deliverables, and the questions still needing an answer.
The fourth part is the one most AI drafts leave out. If the discovery notes have no budget, the draft does not invent one. If nobody agreed a timeline, it does not propose a plausible six weeks. It lists the gap for the proposal owner to resolve. A blank you can see costs a phone call. A confident number nobody agreed tends to survive into the version the prospect signs.
A sample instruction for this step, to adapt to your own template:
Using only the attached discovery notes, proposal template, and service descriptions, draft the objectives, approach, and deliverables sections. Do not state a budget, fee, timeline, or past project outcome that is not in the attached material. End with a list headed "Questions still needing an answer" that names every gap.
3. Your team's approval
Your team confirms scope, pricing, evidence, and delivery commitments, and resolves every flagged gap before sending. In the sample review on our services page, that checkpoint is a single line: the proposal owner checks scope, pricing, factual claims, and client details before sending.
Who reviews what, in that illustrative agency of 30:
- The proposal owner, usually whoever ran the discovery call, owns the draft from start to finish, answers the open questions with the prospect, and checks every factual claim against the approved examples.
- Whoever sets fees confirms every price and makes any discount call. Fees are a decision about the account, not a drafting task.
- The delivery lead confirms the team can actually deliver the scope on the timeline as written.
Write those names down. A review step that belongs to everyone belongs to nobody, and in a three-proposal week the draft goes out unchecked.
Nothing in this workflow writes and sends a proposal on its own. It prepares a first draft from approved inputs only. Your judgment, your pricing decisions, and your relationship with the prospect stay where they were. What shrinks is the repeated assembly.
Why a general AI tool alone is not a process
Your team probably already uses ChatGPT or something like it to start proposals. That is a fair place to begin. It is not a process. What a general tool does not come with is everything around the draft: approved inputs, a named reviewer, a baseline, and a decision on whether the released time is real.
Without those, the trade stays invisible. Illustratively: your account director saves two hours on a first draft. Your head of delivery then spends ninety minutes rewriting a scope section that promised a deliverable you do not offer, and the owner spends another half hour removing a timeline nobody discussed. The draft was faster. The proposal was not. Nobody notices, because nobody timed the old way.
Proposal software has the same gap. It produces a draft too, and leaves open who feeds it, who checks it, and what counts as done.
How to tell whether it is working
Measure three things against how you prepare proposals today.
- Preparation time per completed proposal, from discovery notes to the version that goes out, counted across everyone who touches it.
- Corrections, meaning how much of the draft your reviewers had to change. Hours saved by the writer and spent by the reviewer were moved, not released.
- Adoption, meaning whether the team uses the workflow at all. If half the team still copies from whichever past proposal is nearest, you are measuring the old process.
For scale, published vendor benchmarks put agency proposal work at 8–12 proposals a month and 4–6 hours each (Pitchsite, 2026), or 32–72 staff hours a month. Those are assumptions, not GradeCircle data. Your own number is the one that matters, and you need it before you change anything. Otherwise you will never know what the change did.
Test the workflow on your own proposals
The AI Opportunity Review tests this workflow on your own work. It costs $1,500 and takes 10 business days.
Kickoff is your proposal owner walking us through a real proposal, from discovery notes to the version that was sent. Then come up to three interviews, up to three workflows, and representative work samples, with each workflow scored on benefit, effort, data suitability, and operational risk. The business case sets setup and ongoing costs against released capacity, counted in hours, not payroll savings, because released capacity does not automatically reduce payroll costs.
A decision meeting closes it. You leave with a recommended first pilot at a fixed price, or a recommendation to defer AI and what would need to change. What you get from a $1,500 AI Opportunity Review walks through it deliverable by deliverable. The $5,000 four-week pilot is a separate decision, made only after you have the findings, with software and usage fees disclosed before you commit.
The AI writes the first draft. Your team still writes the proposal.
GradeCircle builds and runs six products of its own: AfroSynergy, CallerBridge, HealthSquire, Heyzl, Lexfora, and LukeAPI, two of them in beta. The assumptions in our models come from operating those systems, not from a framework. See what we run.
If your agency writes proposals every month, scope your proposal workflow: five questions, and you see a preliminary fit summary before sharing contact details.



