What to put in an AI services proposal (and what to cut)
A proposal for a custom AI buildout needs five pages, not twenty. Here is what belongs on each one, what the data says about length and pricing presentation, and the page most agencies skip that costs them the deal three weeks after signature.
Lumina13 min read
An AI services proposal for a local business needs about five pages and four answers: what you are building, what it costs every month, when it goes live, and who fixes it when the agent says something wrong. Everything past that is padding you wrote to feel prepared. The proposal is not where you sell. You sold on the demo call. It exists so the owner can forward it to a spouse or a business partner, without you in the room, and have it still make sense.
Most of these proposals fail one of two ways. Either they are a 20-page deck about artificial intelligence with the price buried on slide 18, or they are a one-line invoice with no scope, so every change request after signature turns into an argument. Here is the structure we use, what the data says about length and pricing, and the page most agencies skip that costs them the deal three weeks later.
What is a proposal for, and when is it the wrong move?
A proposal does three jobs. It proves you listened. It fixes the scope so both of you know what "done" means. It gives the buyer something to say yes to in writing.
Notice what is missing. It does not educate anyone about AI, and it does not persuade. If the owner still needs persuading after talking to a demo receptionist in their own trade, the proposal will not fix that, and a longer one will fix it less. Send a shorter proposal and book a second call.
Sometimes a proposal is the wrong tool. Selling a single service, a website chatbot at $300 a month, send a payment link and a one-paragraph email instead. A formal proposal for a $300 recurring sale tells the buyer this is a bigger decision than it is, and invites them to think harder than the purchase warrants. Below roughly $800 a month in retail, a short email with a start date and a link closes faster.
The other wrong moment is before you know your own cost. If you are ordering the build through our fulfillment platform, your wholesale number comes out of a 30-minute pricing call. Writing a proposal before that call means guessing your own margin, and the guess is usually low. Have the call, then write.
The five pages that earn their place
Page one: what they told you
Their problem, in their numbers. Three or four sentences.
"Northside Heating and Air takes roughly 40 calls a day in August and misses 12 of them. Your average job is $340. You told me you stopped answering after 6pm two years ago because the office staff would not do it. The last time a customer called at 8pm with no heat, they called the next company on the list."
No adjectives. If you cannot write that paragraph without inventing a number, your discovery call was too short. Our demo call playbook covers the questions that produce these numbers.
This page matters more than it looks, because it is written for the second reader. Proposify's 2026 report shows buyers spend 25.3 minutes inside proposals they sign versus 15.8 minutes inside the ones they reject, and that close rates rise 20 percent when more than one person at the company opens the document. The person who was not on your call decides from page one whether to keep reading.
Page two: what you are building
A plain list, no technology names. The owner does not care which model answers the phone. They care what happens at 8pm.
A line item reads like this: "Phone line answered 24 hours a day by a receptionist trained on your services, pricing bands, service area, and hours. Books into your calendar. Transfers to your cell for anything it cannot handle."
Four to seven of those. Each names what happens, what triggers it, and where the result lands. A line item that does not say where the output goes is a wish, not a line item.
Name the trade pack you are using. We ship more than 100 of them, and the pack is why the demo sounded like their business instead of a template with the trade left blank. "Built on our HVAC configuration, including emergency versus maintenance triage" belongs on this page.
Page three: what it will not do
This is the page nobody writes, and it is the one that saves the relationship.
Write the limits down before signature, in the owner's language:
The agent does not quote jobs that need a site visit. It books the visit.
It does not give medical, legal, or diagnostic advice.
It transfers to a human for anything it has not been trained on, instead of improvising.
It will occasionally mishear a street name. Here is what happens when it does.
Voice calls are billed per minute, so the monthly figure moves with call volume.
Two reasons to be this blunt. The buyer already suspects there are limits, so naming them yourself is the cheapest trust you will ever buy. And this page is your scope line. Six weeks in, when the client asks why the receptionist will not process a credit card refund, you point at page three instead of eating the work. We wrote more about change requests after signing, and nearly all of them trace back to a missing page three.
ServiceTitan's 2026 State of AI in the Trades survey, run by Thrive Analytics across 1,017 verified US contractors between May 26 and June 15, 2026, found that 11 percent of contractors report high or complete trust in the AI tools they use, and 58 percent report little or no trust. Among contractors already using AI, 35 percent named accuracy and reliability as their leading concern. That is who you are selling to. A proposal that pretends the system never fails reads like a sales document to a skeptic. One that says "here are the four things it will get wrong and here is what happens when it does" reads like an operator talking.
Page four: the money
One monthly number, large. One start fee. Nothing else on the page except what the monthly includes.
Put the recurring figure first, because that is the decision. Proposify's data on proposals with recurring fees shows an 18 times lift in deal value against one-time pricing, and companies using interactive pricing, where the buyer selects options, win twice as many proposals. A fixed monthly with two optional add-on lines is the cheap version of that.
Show the arithmetic against their own numbers, using the figures from page one and nothing invented:
"$1,450 setup, then $2,100 a month. At your $340 average job, the system needs to recover six jobs a month to pay for itself. You are currently missing around twelve calls a day in peak season."
That is arithmetic the owner can check. It is not a promise, and you should never write it as one. Do not put a projected revenue figure, a percentage lift, or a case study result in a proposal unless you can source it to a named customer who agreed to be named. We have no published customer results and we do not invent them, and neither should you. For how to set the retail number in the first place, see how much to charge for an AI buildout.
One more thing about this page. Put the e-signature on it. Proposify found proposals with e-signatures close 15 percent more often and 60 percent faster, and the ones signed by the sender before sending close 65 percent more often. Sign yours first. It costs nothing and it reframes the document from a quote into an agreement waiting on one party.
Page five: the start date and what you need from them
End with a date and a list.
"We start Monday the 10th. Live in your office by Friday the 21st, assuming we have the five items below by Wednesday the 12th: current call forwarding setup, service list with price bands, calendar access, logo files, and who gets the leads."
That list predicts whether the build ships on time better than anything else on the document. Deals do not die at signature, they die in intake, and the fix is naming the five items on the page the client signs instead of emailing for them later. Our client intake checklist is the long version, and onboarding assistance exists because this is where closed deals stall.
Proposify's numbers back the urgency: their average time from a buyer opening a proposal to closed-won is 2.5 days. If your proposal does not contain a date, you are adding a week to a two-day decision.
How long should it be?
Shorter than you think, and the data is messier than the usual advice suggests.
Proposify's report, drawn from 742,137 proposals worth about $3.06 billion sent through their platform in 2025, found winning proposals averaged 11 pages and losing ones averaged 13. Their separate look at the proposal database found an average five-page proposal closes at roughly 50/50, while a 30-page proposal drops to around 35 percent. Those two cuts do not agree on an ideal number, which is worth saying out loud rather than picking whichever one supports a tidy rule.
What they agree on is the direction. Past roughly a dozen pages, more document buys you nothing and probably costs you. For a local service business buying a $2,000 a month system, five pages is plenty. If you cannot fit it in five, the scope is unclear, not large.
One thing worth stealing from the same data: 83 percent of winning proposals include images, averaging 12 each. For an AI buildout that means screenshots of the demo you already sent, the booking confirmation, the audit output. Not stock photography of robots.
How should you show the price?
Three rules.
Lead with the monthly. Owners budget in months, and a $1,400 setup fee next to a $2,100 monthly reads as reasonable. The same fee alone on a page reads as a wall.
Settle your own cost before you write the page. Our wholesale tiers run from $100 setup and $15 a month for a single service up to $1,150 to $1,400 setup and $315 to $460 a month for the top buildout, plus $0.35 a minute where a line takes calls. You set retail, the gap is yours, and the proposal shows one number. Yours.
Price the voice minutes separately and say so. This is the only truly variable cost in most buildouts, and hiding it inside a flat monthly is how agencies end up losing money on a client in August. If you assemble voice yourself instead, the per-minute math is public: ElevenLabs lists $0.08 a minute for additional agent calls on every self-serve tier, with language models and telephony billed separately on top. That separately is where the surprises live, and it is the reason we quote one per-minute rate that includes the stack.
The trust problem no proposal can close
Adoption is not the barrier anymore. Goldman Sachs surveyed 1,256 small business owners through its 10,000 Small Businesses Voices program between January 27 and February 4, 2026, with Babson College and David Binder Research, and found 76 percent already use AI and 93 percent of those report a positive effect. The gap is somewhere else: only 14 percent said AI is fully embedded in their core operations, and 73 percent said more training and implementation support would help them.
Read that as a sales instruction. Your buyer is not asking whether AI works. They want to know whether this particular thing will still be working in March, and whether they will be the one keeping it alive. A proposal that answers that wins against one that argues the first question.
So put the maintenance terms in writing. Who they call, how fast you respond, what the monthly covers, what gets billed. One paragraph. It is the dullest part of the document and the part a cautious owner reads twice.
One caution in the other direction. You have seen the statistic that 95 percent of enterprise AI pilots fail to deliver measurable return, from the MIT report that went around in August 2025. Keep it out of your proposal. The figure came from 52 interviews, the report called its own findings directionally accurate rather than statistically valid, and Paul Roetzer has picked apart the methodology. Scaring a buyer with a shaky number also hands them a reason to do nothing.
What this approach costs you
Two downsides, because a guide with none is marketing.
You cannot send a precise proposal in the same hour as the demo. Every build is priced on a 30-minute call with us, so there is a step between "they said yes" and "here is the number," usually one business day. If your sales motion depends on quoting a final figure live on the first call, that lag will annoy you, and some reps prefer a published rate card they can read off a screen.
The sandbox disclosure stays visible. On a published demo, nothing carries our branding, not the logo, not the subdomain, not the booking link. The one thing that is always on the page is a line stating that the AI is a demo and that bookings made in it are simulated. We do not remove it. If your pitch relies on the prospect believing the demo is their live production system, this is not the stack for you, and we think that pitch is a bad idea anyway.
A five-page outline you can copy tonight
What you told us. Their numbers, four sentences, no adjectives.
What we are building. Four to seven line items, each with a trigger and a destination.
What it will not do. Five bullets, including the per-minute billing and the escalation path.
Investment. Monthly first, setup second, payback arithmetic in their numbers, e-signature on the page, signed by you already.
Start date and what we need from you. A date, five items, a deadline for the items.
Build it once in whatever document tool you already pay for, save it as a template, and change pages one, two, and four per client. Proposify clocks the average proposal at 17 minutes to create. If yours takes an hour, you are rewriting pages three and five every time for nothing.
FAQ
How long should an AI services proposal be?
Five pages for a local service business buying a system in the $1,000 to $3,000 a month range. Proposify's 2026 data has winning proposals averaging 11 pages against 13 for losers, and their database analysis shows close rates falling as page counts climb past that. Nobody in the data wins by writing more.
Should I include a contract term?
State the term on page four, whatever it is. Month to month closes more deals with small businesses and leaves you exposed on the setup cost, which is why most agencies set a setup fee high enough to cover the build and then go month to month. Pick one and write it down instead of leaving it implied.
Do I need to tell the client which AI models I use?
No, and on page two you should not. It invites a comparison you do not want, where the owner starts pricing API calls instead of outcomes. If they ask directly, answer directly, because dodging it costs you more trust than the answer costs you margin.
What if the client wants to negotiate the monthly down?
Change the scope, not the price. Take a line item off page two and requote. Proposify's data shows discounted proposals carry higher average values, which mostly means discounts show up on bigger deals, not that discounting wins them. Cutting scope keeps your margin honest and teaches the client what each piece is worth.
Can I send the proposal before the demo?
You can, and it closes worse. The point of a live demo is that the buyer tries the thing before reading about it. A proposal lands differently when the reader has already talked to the receptionist you are quoting.
The next step
Pick your most recent closed deal, the one you already know the scope of, and write pages three and five for it tonight. Not the whole proposal, just the limits page and the intake page, because those two are the ones you are missing and they are reusable across every client after this one.
If you want the demo that goes out ahead of the proposal, talk to the live receptionist on our homepage, switch it to your trade, and look at what the page becomes under your brand. The platform is $49 a month, builds are priced at wholesale on a short call, and whatever you charge above that is yours.
