The proposal writes itself, and every claim shows its source
Yes, if every claim shows its source. The draft opens on the buyer's words, prices only the scope they stated, keeps cut lines out, and waits for the rep.
Jared Hawkins-Young · Founder, Lost Star TechnologyOctober 8, 20265 min readPart 8 of 13Key takeaways
- AI can draft a sales proposal from recorded calls and emails, provided every claim in it resolves to a timestamp or a numbered email the rep can play back.
- A sourced proposal opens on the buyer's own sentence, answers the objection in the section where it would resurface, and runs the timeline backward from the buyer's date.
- Stated requirements become priced lines from the seller's own price book, and a line the buyer declined on the call is never quietly added back.
- Three measures show whether generated proposals beat hand-built ones: call-to-proposal time, sections the rep edits, and reply rate per section.
Yes, AI can write a sales proposal from your call notes, and it can write a good one, as long as every claim in the draft points back to something the buyer actually said. The proposal opens on the buyer's own sentence, prices only the scope they asked for, answers the objection where it would come back up, and waits for the rep to approve it. This part shows one proposal built that way: the Northwind deal from part 1, drafted from its calls and emails.
What does a proposal look like when every claim has a source?
- Seats ×4, annual$12,000
- Onboarding and integration, one-time$12,000
- 3Seats 5 and 6removed on the call
- The 14th · Security review starts, SOC 2 report already in hand
- Week 6 · Review complete, legal starts
- Jan 30 · Signed, one annual invoice
- Early Feb · HubSpot connected, four reps live before the kickoff
Four passages carry the weight, and each one does a specific job.
- It opens on her sentence. Dana Reyes, Northwind's VP Sales, said at 12:40 that two deals slipped last quarter before anyone saw them. That line is the first thing in the proposal, verbatim. A proposal that opens on your pitch reads like every other vendor's. One that opens on the buyer's words reads like someone was listening.
- It answers the objection in place. In her third email she wrote that the recorder already pushes a summary into HubSpot. That objection will come back the moment the CFO reads the price, so the answer sits in the comparison section, in her own words from the call: the sync writes a summary, not the fields.
- It keeps the scope cut. At 18:02 she said they would start with four reps, not six. Seats five and six are gone from the price, which is why the amount is $24,000 and not $30,000.
- It runs the timeline backward from her date. She wants four reps live before the February sales kickoff, and at 31:14 she said the security review takes six weeks and has not started. So the plan starts with the review on the 14th and works forward to a signature by January 30. If the review slips, the proposal says February slips with it.
Why does every claim need a source?
Because the proposal is the document that gets forwarded. Dana already sent the one-pager to her CFO "for visibility." The proposal will reach the CFO and the security lead, neither of whom was on the call. Every line the rep can defend with a timestamp is a line nobody has to take on faith. A line the rep cannot source is a line they will hedge on in the next meeting.
Buyers are checking more than they used to. Gartner surveyed 645 B2B buyers in August and September 2025 and reported in May 2026 that 69% turn to sales reps to validate insights they got from AI tools. Your buyer is reading AI summaries of your category and then asking your rep what is true. A proposal where every claim traces to the buyer's own words is the easiest document in the stack to verify.
The same mechanism runs through the whole series. In part 7, the rep presses a CRM field and hears the sentence behind it. Here the rep presses a paragraph of the proposal and gets the same thing back. Nothing in the draft is unattributable, and nothing leaves until a person presses send.
How do stated requirements become priced lines?
- Four reps to startREC 18:02
- Works inside HubSpot, no new loginREC 15:30
- The rep approves every record firstREC 14:05
- Deals and contacts only, never marketing dataREC 22:15
- One annual invoiceREC 28:30
- Seats five and sixDeclined on the callREC 18:02
- Seats ×4, annual$12,000
- Onboarding and integration (HubSpot, recorder)$12,000
- Approval step, deals and contacts scopeincluded
- Billed once a yearin the terms
- Seats 5 and 6not priced
The rule is simple and strict. Every priced line has a stated requirement behind it, and every stated requirement either maps to a line or is marked as not priced. The prices come from your price book, not from the model, and the model never invents a discount. A line with no requirement behind it is flagged for the rep before the draft is ready.
Declined lines are not quietly added back. This is the most common way a proposal loses trust. The buyer said four seats on the call and the proposal arrives priced for six, because the template had six in it. Now the champion has to explain to her CFO why the vendor did not listen. The same line drives the amount recommendation in part 3: the CRM says $30,000, the call says four seats, and the system argues for $24,000 with the quote attached.
The open issues travel with the proposal too. The SOC 2 report Dana asked for in her second email is on the open-issue ledger from part 5, so the proposal's timeline assumes it was sent before the review starts, and says so.
How would you know generated proposals are working?
None of the measures below is a result. They are the instruments to put in place before the first generated proposal goes out, with a month of hand-built proposals as the baseline.
| Measure | What it tells you | The bad sign |
|---|---|---|
| Call to proposal sent | Whether the proposal lands while the conversation is still fresh for the buyer | Still measured in days |
| Sections the rep edits | Whether the extraction is right | The rep rewrites the opening every time |
| Reply rate per section | Which sections the buyer quotes back | A section nobody ever mentions: cut it from the template |
| Declined lines re-added | Whether scope discipline holds | Any number above zero |
Scoring a proposal section by section is the same idea the outbound series applies to cold email, where openings, value lines and asks are scored separately inside comparable cells. A falling edit rate on the opening is the honest signal that the draft is landing. A rising one means fix the extraction, not the rep.
Who this is not for
If your deals close on a single call from a public price page, a template is fine and this is overkill. Generated, sourced proposals earn their keep when the scope moves during the deal, more than one person on the buyer's side reads the document, and a security review or a CFO sits between the champion and the signature. That describes most B2B software deals above a few thousand dollars a year.
Next in the series
The proposal is the artifact a rep sends after the call. The deck is the artifact they present during it. Part 9, Use last quarter's demos to fix next quarter's, reads a quarter of recorded demos to find where the room goes quiet and turns that into an edit list with owners.
Common questions
- Can AI write a sales proposal from call notes or recordings?
- Yes. The recorded calls, emails and rep notes already hold the buyer's pain, requirements, scope, objections and dates. A proposal drafted from them can open on the buyer's own words and price only the stated scope. The condition is that every claim links back to its source, and that the rep reviews the draft and presses send.
- Does the AI set the prices in the proposal?
- No. Lines are priced from your own price book, and the model never invents a discount. Every priced line has a stated requirement behind it, and a line the buyer declined on the call stays out. Any change to price or scope is the rep's decision.
- How do you measure whether generated proposals work?
- Baseline a month of hand-built proposals first, then track call-to-proposal time, how many sections the rep edits before sending, and reply rate per section. A falling edit rate on the opening is the clearest sign the extraction is right.
Sources
Written by Jared Hawkins-Young from the system he builds and runs. Diagrams are drawn from the same specs as the software. Published October 8, 2026. Illustrative numbers are labelled as such; every other figure carries its source.