Drafting sales proposals from discovery notes with AI

A proposal is mostly reassembly: your standard scope language, the buyer's words about their problem, and prices from a book that already exists. This playbook has AI do the reassembly so the account executive spends their time on the parts that win the deal.

01the problem and who owns it

Senior sellers rebuild proposals from the last similar deal: copy the file, find and replace the client name, rewrite the problem statement, and hope nothing from the previous buyer survives. It takes hours, slows the deal, and every so often a proposal goes out with another company's name in paragraph four.

The AE owns the document, deal desk owns pricing and discount policy, and legal owns the terms. Without a shared source, each proposal drifts a little further from the approved language.

02what the AI does, step by step

1. Gather the deal contextThe system collects discovery call summaries, the opportunity fields, emails the AE marks as relevant, and the buyer's stated goals. The AE can add a few lines on what they want emphasized.
2. Choose the closest approved templateBased on product line, deal size, and industry, the workflow selects a template and two or three anonymized past proposals as style references. Client names in references are stripped before the model sees them.
3. Draft the narrative sectionsThe model writes the situation summary, objectives, and approach in the buyer's own terms, citing which note each claim came from. It does not add capabilities that are not in the approved service descriptions.
4. Build scope and assumptionsScope items come from a structured catalog of what you deliver, selected to match the stated needs. Assumptions and exclusions are pulled from the catalog entry, not improvised.
5. Price from the book, not the modelLine items and amounts are computed by deterministic code from your price book and the selected scope. The model never writes a number. Discounts outside policy are flagged rather than applied.
6. Flag gaps for the AEA short checklist lists what the draft could not support: missing timeline, unclear decision maker, a requested item not in the catalog.
7. Route for approvalThe draft opens in your proposal tool or a document editor, with pricing sent to deal desk and any non-standard terms sent to legal.

03systems it connects to

• CRM. Opportunity data, contacts, and the approval status of the proposal.
• Notes and transcripts. Call summaries and discovery notes linked to the opportunity.
• Proposal or document tool. PandaDoc, Proposify, Google Docs, or Word for the final document and e-signature.
• Price book and service catalog. The structured source of scope items, prices, and approved assumptions.

04human checkpoints

• Deal desk pricing approval. Every proposal with a discount or a custom line item goes to deal desk before it is shared.
• Legal on non-standard terms. Any edit to standard terms routes to legal; the model cannot draft new contract language.
• AE final read. The account executive reads the whole document and owns what goes to the buyer.

05what to measure

• Days from discovery to proposal sent. Compared with the period before the workflow.
• Revision rounds. Internal edits before approval, and buyer-requested changes after.
• Discount deviations. Proposals that went out outside policy, which should trend to zero.
• Win rate by template. Which templates and scope bundles close, to guide the next revision of the library.

06risks and guardrails

• Cross-client leakage. Past proposals contain other clients' details. Anonymize the reference library and run a final check for known client names before export.
• Invented commitments. The model must only describe services that exist in the catalog. Treat any unsupported claim found in review as a bug to fix in the prompt or catalog.
• Version sprawl. Store the approved version in one place and lock it once sent so the signed proposal matches the approved one.

07build vs buy

Proposal platforms such as PandaDoc and Proposify offer templates, content libraries, and AI writing help. For productized offers with a short price list, that is usually enough.

A custom build earns its keep when scope is configured from many interacting services, when pricing logic is complex, or when the draft needs to pull from your CRM, call notes, and catalog together in a way a template cannot.

08related playbooks

Browse every sales playbook or the full library.