The Voice AI Proposal That Closes: One Page, Two Numbers, and a Paid Pilot (2026)
TL;DR: A voice AI proposal is not what sells the deal. The demo call sold it. The proposal exists to make saying yes easy and to make the scope impossible to argue about later. Mine fits on one page: the number the prospect gave me on the call, the exact call types the agent will handle, the things it will refuse to do, what the client sees in their own branded portal, a setup fee plus a flat monthly retainer, a start date, and one decision with a deadline. Anything longer reads as a hedge and buys you a two-week silence. I build production voice agents for US clients on Retell, n8n, GoHighLevel and Twilio, and I run VoiceDash, the white-label client portal agencies hand to their clients. Here is the structure I use, and the paid pilot that converts the prospects who will not commit to twelve months on day one.
Most voice AI proposals I see are ten slides explaining what a voice agent is. The buyer already knows, because they just spoke to one on the phone. Explaining the technology again after a good demo does not add confidence, it adds doubt.
The proposal has exactly two jobs
The first job is removing friction from a yes. One price, one start date, one thing to sign. Every extra option is a reason to postpone, and postponed voice AI deals do not come back warm.
The second is fixing scope in writing while the prospect is still enthusiastic. Every painful agency month I have watched came from a scope conversation that happened after go-live instead of before it. The proposal is where "can it also call our old leads" gets a price rather than a shrug.
If your proposal is doing a third job, like educating or persuading, something went wrong earlier. Fix the demo instead, using how to build a voice AI demo.
The one-page structure
1. Open with their number, not your capability
The first line of the proposal repeats a number the prospect said out loud on the call. "You told me roughly 40 calls a week go to voicemail after 5pm, and that a new patient is worth about $900 to you."
That one line proves you listened and anchors the price against their revenue rather than against other vendors, which is why a generic template sent to twenty businesses closes nothing. If you did not leave the discovery call with a call volume figure and a value-per-customer figure, you are not ready to send a proposal. Go back and ask, using the framing in how to sell AI receptionists.
2. What the agent will handle, as call types
Not features. Call types, in the client's own vocabulary.
Answers the main line 24/7 when your team does not pick up within four rings. Books new patient appointments directly into your calendar. Answers hours, location, insurance and pricing questions from your approved answer sheet. Takes a message and texts you within a minute for anything else. Transfers any caller who asks for a person, during business hours.
A client reads that and pictures Tuesday morning. Nobody pictures "intent classification and calendar integration".
3. What it will not do, in the same font size
The section everyone leaves out, and the one that protects the account. Four or five plain lines.
Does not handle billing disputes or discuss balances. Does not give clinical advice. Does not take payments over the phone. Does not make outbound calls in this scope. Does not answer questions that are not on the approved answer sheet, it takes a message instead.
Sophisticated buyers trust you more for it, because everyone selling them AI this year has promised everything, and the boundary is agreed before launch rather than negotiated during an incident. These lines become the scope clause when you paper the deal, which is the structure in the voice AI agency contract guide, and the refusals themselves get written into the agent the way I describe in the AI receptionist prompt guide.
4. What they see
Every prospect asks a version of "how do I know it is working". Answer it inside the proposal, before it comes back as a follow-up email.
You get a login at portal.youragency.com. Every call, recording, transcript and summary, live call volume and outcomes, and your usage for the month. Your team can have their own logins.
That line separates you from the freelancer quoting half your price with a monthly PDF. It is also why I built VoiceDash: it connects to your Retell account and pulls calls, recordings, transcripts and usage into a portal with your logo, on your domain, scoped so each client sees only their own data and never touches Retell. Live in under 10 minutes, no code, on Starter, Growth or Ultimate plans from $19/mo with a free trial. VAPI and Bland support are coming soon. Which numbers clients actually look at is in voice AI client reporting.
5. Price, in three lines
A setup fee, a flat monthly retainer, and what happens if usage runs past the included volume. That is it.
Setup and build: $1,500 one time. Monthly: $900, includes up to 1,000 minutes. Additional minutes at $0.35, billed at cost plus handling.
No tier tables, no good, better, best. A three-column grid invites the buyer to shop against themselves and lands you on the cheapest column. Pick the number that fits the value you just anchored and put it on the page. The reasoning behind flat retainers over per-minute billing, and what to charge by client type, is in voice AI agency pricing, and the underlying platform cost you are marking up is broken down in Retell AI pricing explained.
6. Timeline and what you need from them
Buyers stall when the next two weeks look vague. Make the whole thing look small.
Day 1: kickoff call and your answer sheet. Days 2 to 5: build and internal testing. Day 6: you call the agent yourself and we adjust. Day 8: live on your line. I need your calendar access, your FAQ answers and one hour of your time in week one.
Naming what you need from them is not a weakness in the proposal. It is the thing that makes the timeline believable, and the intake that collects it in one pass is in how to onboard voice AI clients.
7. One decision, with a date
End with a single action and a date attached. "Reply yes and I will send the agreement and the invoice today, and we go live on the 22nd." Not three options. Not "let me know your thoughts". The most expensive word in agency sales is "thoughts".
The paid pilot, for the buyer who will not sign for a year
Some prospects genuinely cannot commit to a year on the strength of one demo, and pushing harder just loses the deal. A pilot rescues it, but only if you structure it as a purchase rather than a trial.
Charge for it. A free pilot has no owner on the client side, and unowned pilots die quietly because nobody ever sends you the answer sheet. Something like $750 for 30 days, credited against the setup fee if they continue, works well.
Scope it to one call flow rather than a discount on everything. After hours only, or new patient booking only. A pilot that does less converts better than one that does everything badly.
Define the success metric in the proposal, in their words and with a number. "Success is 30 or more captured after-hours calls in 30 days that would otherwise have hit voicemail." Agree it before you build, because a pilot with no scoreboard becomes an argument on day 31, and you will lose that argument.
Give them the portal on day one. A pilot with a live branded dashboard converts better than the same pilot reported over email, because the client watches the number climb for four weeks instead of hearing your version of it at the end. That is the entire conversion mechanism, and it is why the portal goes live before the first invoice.
Put the renewal decision in the pilot itself. "On day 30 we either continue at $900/mo or we switch it off and you owe nothing further." Removing the trap makes the yes easy, and by then the client has usually already told their team the agent is theirs.
Three things to keep out of the proposal
Uptime guarantees you cannot control. You are reselling a platform, a carrier and a model provider. Promise a response time for issues, never a number of nines.
A build spec. The client does not need your architecture. It gives them the impression they are buying a codebase, and invites edits from whoever on their team once used Zapier.
Anything about consent or recording as a footnote. Recording disclosure and consent belong in the agreement as terms, decided with the client, using the checklist in voice AI call recording compliance. This is general information and not legal advice.
Following up without discounting
If there is no answer in three days, do not resend the price with a discount. Send new evidence instead: a fresh recording of your demo agent handling the exact objection they raised, or the count of calls their line has missed since you last spoke, if you measured it during the phone test in how to get voice AI clients. Evidence restarts a stalled deal. A discount teaches the client that your first number was fiction.
The proposal checklist
- One page, and their number is on the first line
- Call types written in the client's vocabulary, not features
- An explicit list of what the agent will not do
- The portal named, with what they will see in it
- Setup fee plus one flat monthly figure, no tier grid
- Overage handling stated in one line
- A dated timeline, including what you need from them
- One decision with a date attached
- A paid pilot option only if they hesitated on term length
- Nothing in it you would not want quoted back to you in month nine
The bottom line
The proposal does not have to be persuasive, it has to be small, specific and hard to misread. Anchor on the number they gave you, list the call types and the refusals in the same breath, show them where they will log in, put one price and one date on the page, and offer a paid, scoped pilot to anyone not ready for a year. Then let the demo you already ran do the closing.
Want the portal line in your proposal to be something you can hand over on day one? Start free on VoiceDash or book a demo and I will show you the portal I hand every client.