Reusing past proposals with AI, without mixing clients

Build a proposal library from your past work, with client details stripped out and each outcome recorded, then let an AI agent draft the next proposal from that library plus the notes of the one client you’re writing for. The agent never needs another client’s files open, and the price and the send stay with you.

Why “just adapt the old one” goes wrong

The quick way is to open last year’s winning proposal and ask an AI to rewrite it for the new client. It works until it doesn’t: the old client’s name survives in a paragraph, their numbers shape the new price, or a detail from their engagement ends up in a document someone else reads. Engagement letters often limit where client information can go, and some now name AI tools directly.

Step 1: build the library

One file per past proposal, rewritten so it keeps what’s reusable and none of the client:

---
type: ProposalPattern
description: Margin audit for a direct-to-consumer brand. Won, fixed fee.
tags: [Proposals, Business]
outcome: won
fee_type: fixed
duration_weeks: 6
client: redacted
---
# The problem, in the buyer's words (anonymized)
# The options offered
# Why it won
# What we'd change next time

Keep the frameworks, the structure and the reasoning. Remove names, figures you can’t share, internal details, and anything the engagement letter covers. If a detail is the reason the proposal won, describe it in general terms instead.

Step 2: record what won and why

A library without outcomes teaches an agent your style, not what works. For each proposal, add the outcome, the reason in one line, the fee type, and who decided. After a handful, patterns start to show: which option buyers pick, which part of the scope gets cut, which fee structure closes.

Add every proposal when it closes, win or lose. The losses are half the lesson.

Step 3: draft for one client at a time

Give the agent two things: the library, and this client’s folder. Nothing else.

Draft a proposal for the client in clients/lone-sock-supply/ using
proposals/library/. Take the problem and constraints from the client's
discovery notes. Borrow structure and options from library files
marked won, and say which ones you used. Don't open any other client
folder. Leave fees as [fee]. Save the draft to
clients/lone-sock-supply/proposal-draft.md.

Telling the agent to stay out of other client folders is a request, not a boundary. For confidential work, make it a boundary: run the agent where it can reach only the library and this client (an AI agent on one client’s files and nothing else).

Step 4: what stays yours

• The fee. The library shows what similar work carried. You set the number.
• The fit. Whether the options match what you heard on the discovery call.
• The last read. Look for any sentence that sounds like another client, then send the proposal from the tool you already use.

The pipeline stages around the proposal, from discovery to signed, are in the CRM alternative for independent consultants.

Where Oknola fits

An agent allowed one client reads that client’s files and nothing else, and nothing is sent without your yes. In Oknola, frameworks, past proposals, call notes and results stay linked to the client they came from, and an agent you allow your proposals and frameworks drafts the next one from past work. Oknola types, tags and files what lands in the folder, so the library grows as engagements close. Keep your proposal tool for sending and signing. See it for consultants, and what is available today is on the roadmap.

Questions people ask

Is it OK to reuse past proposals with AI?

Your frameworks and your structure are yours to reuse. Client details belong to the engagement. Check what each engagement letter permits, and strip what it covers before a proposal goes into the library.

What if a client’s contract rules out AI tools?

Then that client’s proposal stays out of the library, and their files stay out of the agent’s reach. You can still use AI on the library and on your own templates.

How many past proposals do I need to start?

A few won proposals are enough. The library gets better with every proposal you add after it closes.

Should the agent set the fee?

No. Let it show what comparable work in the library carried, then decide the number yourself.