Turn call notes into proposals without starting from scratch

Who this is for

This is for owners and sellers who still rebuild the same proposal from a transcript, a notebook, and last quarter’s PDF. Agencies and service firms feel it first: the call went well, then two days disappear into formatting and forgotten exclusions. If the wider leak is packets and handoffs, start at /document-workflows. This guide is the narrower job: conversation to priced follow-up without starting from scratch.

It is not a license to generate contracts, NDAs, or statements of work that change legal terms without counsel. Keep legal language in a reviewed block. Generate around it. Do not let a model rewrite indemnity because the call “felt collaborative.”

Symptoms

If several of these are true in the same month, this is a live operational leak — not a tooling preference.

• A strong call still means a blank document and a hunt through old PDFs for a paragraph that sounded right.
• Prices and timelines in the proposal do not match what was said on the call, or they match a different client’s deck.
• Notes live in a recorder, a CRM comment, and a personal doc — none of them complete.
• The only person who can send a proposal is the owner, because the template is in their head.
• A tool already drafts from the transcript, and at least one draft has gone to a client with a wrong number.
• Onboarding later argues about scope because the proposal was a remix, not a reviewed source.
• Estimates, recaps, and full proposals are used interchangeably, so clients cannot tell what they were asked to sign.

What done looks like

Done is a working operating change, not a purchased seat or a dashboard nobody opens.

• One document type is in use for this motion. The others are named and parked.
• A source packet exists for each draft: notes, CRM fields, and prior scope.
• The draft is produced into a locked template, not a free-form essay.
• Money terms, dates, and exclusions are reviewed by a named person before send.
• The sent file is stored as the file of record, with a link from the opportunity.
• Nobody auto-sends a proposal, estimate, or change note.

Merge first, then draft

Models are good at combining text they can see and bad at remembering a call they were not given. Merge before you generate. The packet is boring on purpose: transcript or notes, the opportunity fields, the last signed scope or rate card, and a short list of “said on the call, not yet in writing.” If a fact is not in the packet, it does not belong in the draft. That rule is how you stop inventing a discovery workshop the client never asked for.

The merge is also a CRM hygiene test. If the opportunity has no next step and no owner, you do not have a proposal problem yet. You have a follow-up problem. Write the next step, then open the template. Proposals that try to resurrect a dead thread are expensive fiction.

One document type

Pick the artifact this motion actually needs. A recap email is not a proposal. An estimate is not a master services agreement. A slide deck is not a signature path. Teams that “just generate a proposal” after every call produce a stack of documents that cannot be compared and cannot be handed to delivery.

Freeze one type for thirty days. If you sell two motions — a small estimate and a larger scoped proposal — that is two templates, two review rules, and two send paths. Do not ask one prompt to decide which document the client is looking at. People decide. The draft fills the chosen shell.

Constrained draft means the template has sections the model may fill and sections it may not. Fill: problem restatement, proposed work in the seller’s language, timeline options already on the rate card. Do not fill: legal terms, custom pricing below a floor, promises about staffing that were not said, or a start date the calendar cannot support. Leave those as blanks or locked blocks for the reviewer.

Review is the product

The review pass is not a rubber stamp. Read money, dates, and exclusions against the packet and against what you can actually deliver. Read names. Read the client’s words back to them; do not upgrade their request into a larger project because the template had a bigger section. If the call was ambiguous, the proposal should say so and offer a decision, not a confident invention.

Store the sent file where onboarding can find it. The onboarding path should open the same document the client signed, not a reconstructed memory. That is also why trusted sources matter: the current rate card, the current exclusions, and the current case studies (if you use them) need owners. A model that pulls last year’s prices from a random deck is doing what you asked — badly.

Never auto-send money terms

Auto-send is the failure mode that turns a useful draft into a commitment. Even a “quick estimate” is money language. Put a human on the last click. If you want speed, shorten the template and sit the reviewer on the same afternoon as the call. Do not remove the reviewer.

Logging helps. Keep the packet, the draft, the reviewed file, and the send timestamp. When a client says “but on the call you said,” you want the packet, not a debate. This is control, not theater.

The source packet, item by item

Transcript or notes: the client’s words for the problem, the constraints they named, and anything they said was out of scope. If you only have a seller’s memory, write that down as memory, not as fact.

Opportunity fields: name, company, owner, stage, next step, and any commercial notes that already survived a human. If those fields are empty, stop and fill them. A proposal should not be the first place the company writes who the buyer is.

Prior scope or rate card: the current, owned version — not last year’s PDF from a search. If two rate cards exist, you do not have a drafting problem. You have a knowledge problem. Resolve the source, then open the template.

“Said on the call, not yet in writing”: a short list. This is where forgotten discounts and casual start dates hide. Each item is either added to the draft as a decision or dropped on purpose. Do not let it leak in as flavor text.

What the reviewer actually checks

Names and entities first. Wrong company, wrong contact, leftover client from the last draft — those errors destroy trust faster than a clumsy sentence. Then money: fees, payment timing, what happens if work expands. Then dates: start, first deliverable, anything that implies a calendar the team does not have. Then exclusions: the work you are not doing, written in the same tone as the work you are.

Read the problem restatement against the packet. If the draft upgraded a simple request into a transformation program, cut it. Sellers do this when a template has a bigger section than the call deserved. The client should recognize their own words.

Read delivery capacity. A beautiful proposal that the studio cannot staff is a future onboarding fight. If staffing was not in the packet, do not invent a named team. Offer a start window you can keep, or leave the date blank for the reviewer.

Failure modes on the way to send

The team generates three document types from one call and the client does not know which one is binding. Freeze the type before you draft. If you need a recap the same day and a proposal later, say so in the recap. Do not send two priced artifacts that disagree.

Someone pastes model output into an old deck that still carries another client’s appendix. Lock the template. Archive the rest. Search should not be able to surface a deprecated shell as “the usual file.”

Auto-send gets turned on “just for small estimates.” Small estimates are still money. If you want speed, sit the reviewer on the afternoon of the call. Removing the last click is how a wrong number leaves the building.

The sent file never gets stored, so onboarding rebuilds scope from the thread. The file of record is part of done. Link it from the opportunity before you mark the next step “awaiting signature.”

DIY vs hire

DIY one template, one merge checklist, and a same-day review habit. A seller who already writes proposals can do this with the recorder, CRM, and docs they have. The first win is reuse of reviewed language, not a new platform.

Hire when you need the draft to pull live from several systems, when several sellers must share a locked template with permissions, or when you want an agent that prepares the packet unattended. That is a later build. The $997 assessment can rank this against follow-up and onboarding. Implementation stays off that fee.

Controls before automation

Recommendations that touch customers, money, contracts, private data, or systems of record need human review, limited permissions, a log, and a fallback. Do not automate a broken path because the tool is ready.

• Money terms, discounts, and start dates require a human send.
• The model may not invent scope, staffing, or legal language.
• Drafts cite the packet. If a sentence has no source, cut it.
• The sent file is stored and linked from the opportunity.
• Old templates are archived so they cannot be pulled as current.
• If the rate card and the call notes disagree, stop and ask. Do not average them.

Same afternoon, not same minute

Speed here means the reviewed file leaves the same day as the call, not that a model sends while you are still in the parking lot. Block a review window on the calendar the way you would block the call. If the owner is the only reviewer, the constraint is their afternoon, not the generator.

If same-day is impossible because scope is still moving, send a recap that states the open questions and the date of the priced document. Do not send a half-priced proposal to “keep momentum.” Momentum that commits the wrong number is just a faster mess for onboarding.

Train anyone else who will draft. The packet checklist is the training. If they cannot assemble the packet, they are not ready to generate. Generating into an empty packet is how last quarter’s appendix shows up with a new logo.

If two sellers produce different documents from the same packet, the template is still too free. Tighten the sections they may fill. Shared language is the point. Personal style can live in the cover note, not in the priced body.

Among the fourteen asks

Call-to-proposal is a stop-rebuilding ask. Rank it with follow-up and onboarding so you do not polish a document for a pipeline that has no next step. If you want that ranking written down, book the assessment.