AI proposal generator for agencies: turn a client brief into a priced scope of work

Every agency owner knows the Thursday-night proposal. The brief came in on Monday, the call went well, the client wants something "by end of week", and now you're in a spreadsheet guessing how long a website redesign takes while the scope section gets copied from the last proposal you sent.

Then the project starts. The client asks for "just one more page template". Content arrives four weeks late. Somebody mentions that the old site had a members area that never made it into the brief. And the job you priced at a healthy margin finishes at break-even.

This post walks through an AI proposal generator for agencies that fixes both halves of that problem: the hours it takes to write the proposal, and the money you lose after it's signed because the scope of work was vague. It's a full build blueprint you can make in Lovable in a day or two, for your own agency or to sell to others.

Get the full blueprint: the AI Brief-to-Proposal Generator with Scope of Work & Pricing pro prompt includes the complete master prompt, 11 build steps, an 8-step sell-and-run playbook, required tools with real pricing, and a cold outreach script you can send today.

The real cost of a proposal is the stuff you forgot to write down

Start with the obvious cost: time. Loopio's research puts the average formal RFP response at 23 hours across 9 people. Most agency proposals are lighter than a formal RFP. But even a third of that is a full working day from your most expensive person, on work that might not win.

The bigger cost shows up later. Ignition's 2025 agency pricing report (surveying 270+ US agency leaders) found that 57% of agencies lose $1,000 to $5,000 a month to unbilled work, 30% lose more than $5,000, and 78% rarely or only sometimes charge for out-of-scope work. Average agency net margin in the same research sits around 13%.

Put those together and the maths is ugly. A 13% net margin on a $20,000 project is $2,600. One unbilled week of a designer's time wipes most of it out.

Most of that leak starts in the proposal. If the scope doesn't say "up to 8 page templates, 2 rounds of revisions each", the client reasonably assumes 12 templates and unlimited tweaks. If nothing says "stock photo licences are extra", you're paying for them. And if the hours came from gut feel instead of what similar projects took last time, you were under before you started.

What existing proposal tools do well, and the gap they leave

There are good proposal tools. Proposify's Build plan is $49 per user per month (or $35 billed annually) with 10 sends a month, and Grow is $325 a month for 5 seats. PandaDoc's Starter plan is $35 a seat monthly or $19 billed annually, with Business at $65 or $49. Better Proposals starts around £10 to £15 per user per month for a single user sending 10 documents.

They make proposals look good, track opens and collect signatures. What they don't do is decide the price. The numbers still come from whoever's writing the proposal, usually from memory or a spreadsheet, and nothing stops a junior account manager discounting a job below what it costs to deliver.

That's the gap this blueprint fills. It reads the brief, prices from your own past project actuals, writes the scope of work, and blocks anything under your margin target before it reaches the client.

How the brief-to-proposal generator works

There are two parts: a one-time setup, and a flow you run for every new brief.

The one-time setup

The agency loads its service catalogue: services, the deliverables inside each one, and default hours per role. It adds its roles with two rates each, the internal cost per hour (salary plus overheads divided by billable hours) and the bill rate the client sees. It sets up three package templates (Good, Better, Best, labelled however the agency likes) with a contingency percentage for each. And it sets a margin target and a margin floor.

Then the important bit: 10 to 20 finished projects with the hours they were quoted at and the hours they took, by role, plus a one-line reason for any that overran. Most agencies can export this from Harvest, Toggl, Clockify or Productive in an afternoon.

This past-project data is what makes the whole thing work. The AI can only price as well as your history, and your history knows things your gut doesn't.

Brief in

Someone pastes the approved brief, uploads the client's PDF or Word document, or drops in a discovery-call transcript. The AI reads it and pulls out every requirement, each with the exact sentence from the brief that supports it. That quote matters. It means the AI can't invent a requirement the client never asked for, and you can click any line to see where it came from.

Each requirement gets sorted into must, should or could, and matched to one of your services and deliverables with a confidence score. Anything vague ("a modern feel", "integrate with our systems") gets flagged as unclear with one specific clarifying question. One click turns all the open questions into a friendly email to the client.

If you'd rather fix vague briefs before they reach this stage, the sibling Client Intake to Brief with Gap Chaser blueprint handles the intake side.

Hours priced from what past projects took

This is the part I like most. The hours estimate is done by plain code, with the AI kept away from the arithmetic. For each deliverable, the app finds similar past projects for the same service and takes the median of their actual hours by role, ignoring what was quoted. It then gives a low (25th percentile), likely (median) and high (80th percentile) figure.

Every line item shows its working in plain English: "Median actual of 6 similar past projects: 34h. Quoted range 28-41h." With 5 or more matches the estimate is high confidence. With fewer than 3 the app falls back to your catalogue defaults and says so.

The AI's only job here is a complexity factor between 0.7 and 1.5 for this particular brief, with a one-sentence reason. A 40-page site with three integrations isn't the same job as a 12-page brochure site, even in the same service.

Three packages, priced and checked

Good covers the must-haves. Better adds the should-haves and is marked as recommended. Best includes everything, the could-haves and (if your template says so) a 3-month retainer. Each is priced from your bill rates plus that tier's contingency, never below the service minimum, rounded to clean numbers.

Then the checks run. A margin meter per option turns green at target, amber between floor and target, and red below the floor. Red blocks sending until the owner approves with a written reason. Separate warnings fire if a line item is cheaper than similar past jobs sold for, if the likely hours are under what similar projects took, if a must-have has no line item covering it, if the recommended option is more than 15% over the client's stated budget, or if a service has a habit of overrunning by more than 25%.

The scope of work generator

Here's where scope creep gets handled before it starts. The AI writes five sections, all editable: in scope (with quantities and revision rounds), out of scope (at least 6 items), assumptions, client responsibilities and change-request terms.

Two details make this scope better than a copied template. First, every unanswered clarifying question becomes a written assumption, so "we assume the client supplies all page copy" is on paper before anyone signs. Second, the out-of-scope list draws on the overrun reasons from your own past projects. If three past jobs overran because "client added extra templates mid-project", the next proposal says plainly what happens when they do.

The change-request terms are short and clear: anything outside scope is quoted in writing before work starts, at your change-request hourly rate or a fixed price, and late client inputs move the timeline by the same amount.

The client side

The client gets a private link to a branded web proposal that works on a phone. They see a "what we heard" summary in their own words, the three packages side by side, optional add-ons they can toggle (the total updates live), the full scope, a week-by-week timeline, the payment schedule, case studies, team and terms. Hours, roles and margins never appear.

To accept, they type their name and title, tick the terms box and sign. The app records the timestamp, IP address, browser details and a SHA-256 hash of the exact proposal they saw, locks it, and sends them straight to Stripe Checkout to pay the deposit. After payment they get a signed PDF with the signature record on the last page.

Back at the agency, every open, section view and option click is logged. Reminders go out automatically if the proposal hasn't been opened in 3 days, hasn't been decided in 7, or is 3 days from expiry. And once the project is running, any extra request becomes a one-page change request the client approves from a link, with a running total of approved extras.

Want to build this? The full proposal generator master prompt specifies all of it in one paste: the data model, every screen, the AI rules, Stripe and email setup, edge cases and seed data, so it works on first load.

A worked example: Northlight Studio and the garden centre brief

Northlight Studio is a 6-person web and SEO agency. A regional garden centre chain sends a two-page brief: redesign the website, keep their Google rankings through the move, add a store finder for their 4 locations, and "maybe a plant care blog later". Budget: "around £20,000". They want to launch before spring.

The account manager pastes the brief in. The app comes back with 11 requirements. The redesign, the rankings and the store finder are must-haves, each with its supporting quote. Copywriting for key pages is a should (the brief says "we'd like help with the words"). The blog is a could. Two items get flagged as unclear: "integrate with our stock system" (which system, and does it need live stock levels?) and "launch before spring" (a hard date or a hope?). The red-flags panel notes a fixed seasonal deadline.

The account manager sends the two questions to the client with one click and starts on the proposal while waiting.

The app finds 6 similar past redesigns. Their median actual effort is 142 hours, against a typical quote of 118. That 24-hour gap is the story of this agency's last year, and it's now visible on the line item instead of buried in a time-tracking export.

Here's where it earns its keep. The account manager, wanting to land near the £20,000 budget, drags the design and build hours down to 120. A warning appears: likely hours are below the median actual of similar past projects, with a one-click fix to use 142. The margin meter on the Better option turns amber. They leave the hours at 142 and instead follow a different suggestion: move the copywriting into an optional add-on, which brings the recommended package close to budget without pretending the work takes less time.

The scope section now says the redesign covers up to 10 page templates with 2 rounds of revisions each, the store finder covers 4 locations, and the stock integration is listed as an assumption ("product pages show stock status from a nightly file export; live stock levels are out of scope unless confirmed") until the client answers. The out-of-scope list includes extra templates, photography, ongoing hosting and the blog content itself, partly because two past projects overran on exactly those.

The client opens the proposal on a Tuesday evening. The owner gets a phone notification. The client views the Better option three times and toggles the copywriting add-on on. On Thursday they sign and pay the 50% deposit through Stripe. Four weeks in, they ask for a gift-card page. That becomes a £900 change request, approved from a link the same day, and the agency gets paid for work it would previously have done for free.

Total time from brief to sent proposal: about 20 minutes plus the wait for two answers.

Who should build this, and the real economics

Three kinds of people should look at this blueprint.

Agency owners building it for themselves. If you send 4 or more proposals a month and price at least partly on hours, the tool pays for itself the first time a change request gets approved instead of absorbed. Running costs are small: Lovable Pro is $25 a month, Resend's free tier covers 3,000 emails a month, Stripe charges per payment (2.9% + 30c in the US, from 1.5% + 20p for standard UK cards), and Lovable Cloud and AI usage for a small agency runs from a free allowance into usage billing.

AI app builders and freelancers selling to agencies. The suggested pricing in the blueprint is a $2,500 to $4,500 setup fee and a $297 to $497 monthly retainer. The setup covers the build, loading the catalogue and past projects, branding, Stripe and email, a 2-week supervised pilot and team training. The retainer covers hosting and usage, monthly win-rate and margin reporting, and catalogue and template updates.

Sign 4 agencies at the midpoint and that's around $14,000 in setup fees plus about $1,600 a month recurring, for a build you reuse with different data each time.

There's also a performance option that's easy to sell: drop the monthly to $197 and take 5% of approved change-request value, capped at $1,000 a month. You only earn more when the agency recovers money it used to give away, which is a very easy conversation with an owner.

Freelancers and fractional consultants. A cut-down version (one user, no approval flow, smaller catalogue) sells at $1,200 to $1,800 setup and $97 to $147 a month.

The best prospects already pay for Proposify, PandaDoc or Better Proposals. They've accepted that proposals deserve software. They just still price in a spreadsheet. You can find them on Clutch, DesignRush and Sortlist, in agency owner communities, and on LinkedIn by searching for owners and managing directors at 2 to 50 person agencies.

The blueprint includes a ready-to-send cold email that opens with a question every owner has an uncomfortable answer to: did your last project come in on the hours you quoted?

What to expect, honestly

Building takes 1 to 2 days for someone comfortable with Lovable, plus a day loading real data. Then run a 2-week pilot: build the next 5 proposals both ways and compare. Where the app disagrees with the team, check the estimate basis. It's almost always a past project in the wrong service, or a cost rate that's off. Fix the data and the estimates follow.

Estimates stay weak until each service has at least 3 similar past projects behind it. The app says so on every low-confidence line, and the monthly routine of adding finished projects keeps improving it.

The AI will occasionally mismatch a requirement, so a person always reviews before sending. And win rates won't jump overnight. The faster, clearer proposals help, but the money you'll see first is in margin: fewer under-priced jobs, and change requests that get paid.

Get a lawyer to read the terms and signature wording before going live. Typed-name e-signatures with a timestamp, IP and document hash are the standard approach for commercial agreements under the US ESIGN Act and UETA and the UK Electronic Communications Act 2000, but the agency's own terms are the agency's call.

Ship it as a real app your team carries around

The last step of the playbook wraps the web app as a native iOS and Android app (plus a PWA) through AppBuild.DIY. The reason is timing. The moment a client opens a proposal, signs it, or asks a question is the moment a quick phone call wins the deal, and a push notification on the owner's phone catches that moment far better than an email sitting in an inbox.

Turning on the camera too means someone can snap a printed brief or a whiteboard at a client meeting and start a new proposal from it on the spot. And the owner can approve a below-target proposal from their phone instead of finding a laptop.

Why build it now

Ignition's same research found 97% of agencies plan to raise prices this year. Raising your rate card is the easy part. Holding the margin depends on whether your proposals price from real hours and your scope says clearly what's included.

This build does both, it's a weekend-sized project in Lovable, and the market is every small agency that's ever finished a "profitable" project at break-even.

If you also run briefs for other teams, the rest of the Agency Systems set covers the upstream work, including the SEO Content Brief Generator and the Website Redesign Brief & Sitemap Planner.

Build your proposal generator today: grab the AI Brief-to-Proposal Generator with Scope of Work & Pricing pro prompt. Fill in your agency name, currency, margin target, deposit and change-request rate, paste it into Lovable, and send your next proposal from it.