Proposal Generation: 70% Template, 30% AI Customization

If you're a Head of Sales who's been told by a vendor that AI will write proposals from scratch in 90 seconds, my honest take is: please don't. The teams that actually ship proposals faster and win more are not the ones where AI invents the document. They're the ones where AI customizes the 30% that should be customized and leaves the 70% that shouldn't.

Why does "let AI write the proposal" go wrong?

Three failure modes. First, AI invents legal language because it pattern-matches against documents it has seen — and your MSA is not those documents. Second, AI sees pricing as text to optimize rather than a discipline to defend; it tweaks structure in ways that erode commercial logic. Third, the rep loses the muscle of explaining the deal in their own words because the AI has done it for them.

Definition: Proposal slop — fluently written, structurally complete proposals that nobody on the deal team can defend line by line, often produced by handing the entire document to a generative model.

The teams I see winning with AI on proposals have done the opposite — they've spent two weeks tightening the template and only then layered AI on the small surface where customization matters.

What's the 70% that should be a locked template?

This is the part that should never vary call to call. If it varies, you have a pricing or legal discipline problem, not a writing problem.

• Legal language: MSA, SOW boilerplate, confidentiality, IP assignment, termination, jurisdiction
• Pricing structure: pricing tables, package logic, discount-approval thresholds, payment terms
• Methodology: how you deliver, what stages exist, what the customer is responsible for
• Deliverables list: the format and definition of each named deliverable
• Case-study insert section: one or two pre-approved stories per vertical, picked by tag
• Team/company background: who you are, why this work is in your zone

These 70% should sit in version-controlled blocks. The rep does not edit them. Legal owns changes. Pricing owns changes to the pricing logic. If a rep wants something different, it goes through an exception process, not into the proposal.

What's the 30% AI is good at customizing?

The parts that should vary by account. AI is genuinely strong here because the inputs are already in the CRM and the discovery notes.

Account-specific value framing

The opening section that says "here's why this matters for [you specifically]." AI takes the discovery notes — pain points, quantification, named stakeholders — and produces 2-3 paragraphs in the company's voice that name the buyer's situation. Reviewed by the rep, edited if needed, but the draft saves an hour.

Objection pre-emption

If discovery surfaced specific objections — "we've been burned by [vendor type] before," "our procurement takes 12 weeks," "our CFO will ask about ROI in months 1-3" — AI drafts a paragraph in the proposal that addresses each one head-on, in the relevant section. The rep doesn't need to remember; the discovery notes do.

Definition: Objection pre-emption — addressing a known buyer concern in the proposal text before they raise it in the next call, reducing the back-and-forth cycles between proposal and close.

Commercial framing of the same price

The price tables are locked. But the way the price is positioned — "the equivalent of replacing 1.5 FTEs at a fully-loaded cost of X," "less than the cost of the consulting engagement you described avoiding" — that's customized per account from the discovery data.

Executive summary

The one-page exec summary at the top, calibrated for the named economic buyer's role and concerns. AI takes the discovery + the locked sections and produces a 200-word summary the rep edits down to 150.

Copy/paste proposal-generation pattern

Inputs (collected from CRM + discovery notes):
- Account name, industry, size (employee count)
- Named stakeholders: economic buyer, champion, end users
- Top 3 pain points (with quantification if available)
- Top 3 objections surfaced in discovery
- Compelling event + date
- Pricing tier selected (from rep, must match locked options)
- Selected case studies (1-2, from approved library by vertical tag)

Locked sections (no AI touch):
- Cover + table of contents
- Methodology
- Deliverables table
- Pricing table (sections, not numbers — numbers come from the tier input)
- Legal: MSA, SOW boilerplate, terms
- Team bios

AI-customized sections:
- Executive summary (200 words, calibrated for economic buyer role)
- "Why this matters for [account]" — 2-3 paragraphs from discovery
- Objection pre-emption — one paragraph per objection, placed in relevant section
- Commercial framing — 100-word reframe of price for this account
- Case-study lead-in — 50 words connecting picked case to this account's situation

Output: rep reviews the AI-customized 30% in 15 minutes, edits if needed, hits send.
Turnaround target: under 24 hours from discovery to sent proposal.

Tool tip (Course for Business): The pattern here is Augment, don't replace at its cleanest — AI does the 30% the rep would have spent the most time on, and humans guard the 70% where AI would have caused trouble. In our 6-week program, week 3 is the proposal-template lockdown session, and week 4 is the AI-customization layer. Skipping the lockdown and going straight to "let AI write it" is the single most common failure mode we see in sales orgs trying this. Walk through the program at https://course.aiadvisoryboard.me/business.

Good vs bad examples

Account-specific value framing

• Bad: "We help companies achieve operational excellence through transformative solutions."
• Good: "You mentioned that your three-week onboarding cycle is costing you 6-8 weeks of productive time per quarter across your 12-person new-hire intake. The work we're proposing targets that specific cycle and reduces the median onboarding time to under five days within 90 days."

Objection pre-emption

• Bad: "We are confident in our delivery."
• Good: "You shared that your last vendor engagement stalled in week 8 because internal stakeholders disengaged after the initial workshop. We address that with weekly stakeholder check-ins and a written 30/60/90 review with your COO — not optional, not skippable."

Commercial framing

• Bad: "Total: $147,000 for the engagement."
• Good: "Total engagement: $147,000. For context, this is the equivalent of about one FTE for the year, fully loaded — and the projected hour-savings you described in discovery exceed two FTEs per quarter. Payback inside the first quarter, by your own numbers."

Team scan (what AI champions report after week 1)

• 18 reps in the org, 1 AI Champion (sales operations lead) maintaining the template + customization layer
• Adoption: 100% of new proposals routed through the 70/30 pattern within 10 days
• Use case: AI-customization layer reduces rep prep time per proposal from ~3 hours to ~45 minutes
• Saved time per rep: 6-9 hours/week previously spent on proposal Frankensteining
• Use case: legal review queue shrunk by ~60% because locked sections don't trigger review
• Friction: 2 reps initially edited locked sections; AI Champion ran a 15-min reset on why
• Win-rate pattern: first-proposal close rate moved up 4-7 points in 4 weeks, driven mostly by objection pre-emption
• Template upkeep: AI Champion runs a 30-min monthly review of locked sections with legal + pricing
• Disclosure: reps know which sections AI touched, which sections it didn't — full transparency
• After 6 weeks: turnaround time discovery → proposal sent dropped from 4 business days to under 1

Micro-case (what changes after 7-14 days)

A 130-person managed services company averaged 4.5 business days from discovery to proposal, with a 28% first-proposal close rate. Each proposal was 60-80% rewritten from scratch by the rep, with legal review on every one. They locked down 70% of the document over a weekend with legal and pricing, set up an AI-customization layer reading from CRM discovery notes, and trained the reps in a single 90-minute session. Week 1: average turnaround dropped to 22 hours. Week 3: legal review queue was down 65%, first-proposal close rate had moved to 34%, and three reps reported that the objection pre-emption section had been quoted back to them in close calls by buyers. By week 8, turnaround averaged under 12 hours, close rate sat at 36%, and the sales team had freed up an aggregate ~25 hours/week previously spent on proposal mechanics.

Note on this case: This example is illustrative — based on typical patterns we observe with companies of 30-500 employees, not a single named client. Specific numbers are rounded approximations of common ranges, not guarantees.

Tool tip (Course for Business): The fastest way to land this in a real sales org is Shoulder-to-Shoulder — the AI Champion sits next to one rep, runs through the AI-customized 30% live, and the rep narrates "what I'd change in the value framing." Two of these sessions per rep, two weeks apart, is enough to lock in the muscle. We see the 70/30 pattern stick when reps personally edit the AI-customized section in their second hot seat; we see it fail when the company tries to roll it out by email blast. Book a 30-min mapping call at https://course.aiadvisoryboard.me/business.

FAQ

Won't proposals feel templated and impersonal? The opposite, in practice — the 30% customization is what buyers notice, and it's higher-quality customization than reps used to write under time pressure. Buyers don't read the legal language; they read the executive summary and the "why this matters for you" section. Those are the AI-customized parts.

Can AI also negotiate the price? No. Price discipline is exactly the area where AI does the most damage. Pricing tiers are inputs, not outputs of the AI layer. If the rep wants a non-standard price, it goes through human exception approval — not through a "let me see if the AI suggests this."

What if our proposals are wildly different per deal? They probably shouldn't be. The "different per deal" instinct is usually proposal craft hiding pricing or scoping discipline that should be fixed at the catalog level. If after fixing the catalog they're still different per deal, you're in custom-services territory and the 70/30 ratio shifts — but the principle holds.

Doesn't this require a fancy proposal-generation platform? A docx template, a few merge fields, and a small script calling an LLM with the discovery notes will work for a team of 5-20 reps. Tools like PandaDoc or Proposify make it tidier; they're not required.

How does this interact with discovery-call coaching? Tightly. The discovery notes feed the AI-customization layer; bad discovery means bad customization. The 70/30 pattern surfaces discovery weakness — if a rep's "why this matters" section keeps coming back generic, the discovery was generic. Worth a separate coaching loop.

Conclusion

The promise of "AI writes the proposal" is mostly bad. The reality of "AI customizes the right 30% of the proposal" is excellent. Lock the template. Identify the 30% that varies by account. Let AI handle that 30% from the CRM. Watch turnaround drop and close rate move.

Pick the 70% to lock down this week. Identify the 30% to AI-customize. Ship the first proposal under the new pattern by Friday.

If you want every employee — including every sales rep — to ship their first AI automation in five days, with proposal patterns that hold up under real deals, book a 30-min call and we'll map your team's first week at https://course.aiadvisoryboard.me/business.