Sales proposals in minutes: one agent drafts, another checks, and it waits for your sign-off

The problemEvery proposal takes days, every rep builds it their own way and nobody knows whether it follows the pricing policy.

Tuesday, 10:00 a.m. Your prospect wraps up the meeting with a line that sounds like a win: "Send me the proposal." The rep promises to send it "as soon as possible." He digs up an old proposal from another client, copies sections, asks finance for the updated price, waits, adjusts the scope and sends it Friday at 7:00 p.m. On Monday you discover he applied a discount nobody authorized and cited a case study from three years ago. That is how a lot of sales proposals work, and it is why so many opportunities go cold between "send it over" and "received."

Sales proposals with AI are not about taking work away from the salesperson. They are about having the right draft ready in minutes, so the rep can spend their time on what only they can do: read the client and make the call.

Why a proposal takes three days

The time does not go into writing. It goes into searching and waiting:

• Scattered information. Prices in one spreadsheet, templates in a folder, case studies in someone's head and payment terms in an email from months ago.
• Every rep has their own version. One uses the 2024 template, another builds it from scratch, another copies the proposal that worked best for him. The brand and the promise change depending on who handles the deal.
• Late validation. Finance or leadership review at the end, if at all. Pricing errors are caught after the client has already seen them.
• Competing priorities. Reps build proposals between calls, meetings and follow-ups. The proposal competes with everything else.

The root problem is that the proposal depends on each person's memory and style, not on a system.

How it works: one agent drafts, another checks, you sign

An agent is a program that uses an AI model to carry out a specific task within rules you define. For proposals, the design that works separates three roles:

1. The drafting agent. It takes the meeting notes, the client file and the type of service. It consults the current price list, approved templates, comparable case studies and standard terms. With that, it builds a complete draft: client context, scope, price, timeline, reference case and terms.
2. The evaluator agent. It checks the draft against your policy: that prices are current, that the discount is within the allowed range for that type of client, that payment terms are the approved ones, that the cited case is still valid and that no section is missing. If something is off, it sends it back to the drafter or flags it for a person. We explain why this role must be a separate agent in generator and evaluator.
3. The salesperson. They receive a draft that has already been validated, adjust the tone, add what only they know about the client and approve it. If the discount exceeds their authority, the proposal goes up to the manager. That is how human approval gates by role work.

Nothing goes to the client without a human sign-off. And every step is logged: who requested what, what the rep changed and who approved it.

Checklist to get your sales proposals ready for AI

Before you build the agent, organize what it will use. This work alone already improves your proposals:

• A single, current price list, with an owner and an update date.
• A written discount policy: how much a rep, a manager and leadership can give, and under what conditions.
• Two or three approved templates by type of service or client size.
• A case study library with industry, problem, result and whether the client authorized you to cite it.
• Standard terms: payments, timelines, warranties, offer validity.
• A definition of "proposal ready": which sections it must include, no exceptions.

Illustrative example: the corporate training firm

Illustrative example. Picture a firm that sells training programs to companies. Its eight reps each build proposals differently: some include materials and some do not, some price per person and others per group. The sales director spends every Friday reviewing proposals that have already been sent.

Before, a proposal went out two to four days after the meeting. Every so often, one went out with an outdated price and had to be corrected in front of the client, with the credibility cost that comes with it.

Now, the rep uploads their notes when the meeting ends. The drafting agent builds the draft with the right program, the current group price, a case study from the same industry and the standard terms. The evaluator spots that the rep requested a larger discount than allowed and routes it to the director for approval. The director approves from his phone, and the rep sends the proposal that same afternoon, with their personal touch. The Friday review disappears because the review happens before, not after.

Common mistakes

• Asking the model to "write a proposal." Without prices, templates or rules, the result is polished and dangerous text.
• Letting the same agent draft and review itself. An agent grading its own work tends to approve it. Validation must be independent.
• Skipping the human sign-off to move faster. Speed is gained in the draft, not by removing approval.
• Having three "current" price lists. The agent is only as good as the information it consults.
• Not measuring. Track how many drafts pass the evaluator on the first try, how much the rep changes and how many proposals are won. That is how you tune the system; we cover it in evals for AI agents.

What to do on Monday

Take the last ten proposals your team sent and compare them: time from meeting to send, price applied vs. current list price, and structure. If you find ten different ways to sell the same thing, you already have your business case.

At Addabra, we build this workflow inside a Sales OS: a drafting agent connected to your sales brain, an evaluator that applies your pricing policy and approval gates by role, with every change logged and costs under control. Your team still signs; it just stops losing days along the way.

salesproposalsAI agentshuman approval