The one change that increased my client close rate by 63%

I've failed to close yet another client.

Had a great discovery with the potential client one Tuesday. Great brand. Real budget, and they actually know what they want. The dream. I came off the call feeling like the deal had basically been done.

Friday afternoon rolls around and the proposal still hasn't gone out.

Not because it's difficult. Writing a SOW isn't hard. It's just tedious and takes time to get it right.

But somewhere between that call ending three days ago and today - I never got round to sending the proposal to the client.

It's not because I didn't want to. It was the 3 discovery calls I'd had since, the emails, WhatsApp messages, delivery for other clients, the dog walks, the food shop. The rest of my week got in the way.

The proposal sat on my to-do list. The lead cooled. Then emailed me at lunchtime saying they've decided to go with someone else.

This isn't the first time this has happened.

The problem is 'speed-to-lead'. I'm sure you've probably heard this term before. Productivity and biz-dev gurus love to throw it around. It's basically the gap between the initial client interaction ending and the proposal landing in their inbox. Top performers measure this gap in minutes. Most people (myself included) measure it in days. The shorter the gap, the higher the close rate.

I see this pattern in every AI consulting client I work with. The owner is sharp on the call, capable of turning a client's fragmented ideas into real outcomes - but bottlenecked entirely on the time it takes to write the proposal doc. They lose deals to operators who aren't better, just faster.

The fix? Use automation to cut your proposal writing time down by up to 90%. Call ends, document is in your prospect's inbox within the hour.

WHAT I BUILT

An agentic skill called the AI Proposal Drafter (link here if you want it).

It does one thing. It takes whatever you captured during the discovery call - a Granola or Fireflies transcript, an email thread, a page of typed notes - and turns it into a first-draft SOW based around the ones you're already sending. If it's not sure about something - it'll ask you some followup questions in order to produce the most accurate first-draft possible.

You set it up once. Train it on all your existing offer structure, rates & proposal templates. After that - it takes approximately 2 minutes to execute end-to-end.

Setup phase

(one-time, takes about ten minutes)

1. Pricing

Drop in your rate card if you have one. If you don't, the skill researches public market rates for the services you offer (using sources like the YunoJuno freelancer rates report, the Motion day rate survey, AOI guides, the Wethos pricing benchmarks), proposes a structured rate card with you, and asks the questions it needs to fill the gaps: typical engagement length, retainer or project preference, day rate or fixed-fee, common deliverable bundles. Even if you do have a rate card, it'll suggest pricing-structure tweaks and possible upsells based on the same benchmarks.

2. Tone of voice

Tell it how you write proposals that sound like you. Formal? Conversational? Dry? Warm? Or upload a recent SOW you wrote and the skill mirrors it. Tone is captured once, not asked every run. You can update it whenever you want to.

3. Past briefs and SOWs (optional)

Upload two or three you've sent before. The skill mirrors the section headings, structure & framing you're already using in order to personalise the output to your specific needs.

Operating phase

(run after every discovery call)

1. Drop in the input

A call transcript is the strongest signal because it carries phrasing, emphasis, and context you'd lose otherwise. The skill is optimised for transcripts. There are a tonne of free AI meeting notetaker apps out there right now - Granola, Fireflies, tl;dv and Otter all have free tiers worth trying. If you didn't record the call, an email thread or a page of typed notes works too. More detail is better here. The output quality scales with the input quality.

2. Answer the Gap Analysis questions

The skill identifies the points where the client was vague or where you'd need a judgement call before drafting, and asks you. Three example questions from a real run:

"The client said the budget is roughly £40k. Draft against that as a ceiling, or set it as a range?""They mentioned an existing image library, but it's unclear whether they own it. Do we budget for new photography?""Launch is Q3 but they are asking for a lot of deliverables here. Do we want to have them prioritise the most important deliverables first?"

You answer the questions. The skill writes the proposal with your answers folded in.

3. Get the draft

Two minutes later, a structured proposal document, ready to edit:

• Executive Summary, two or three sentences in the client's own language, pulled directly from phrases they used on the call.
• Scope of Work, bulleted, filtered into must-haves and optional add-ons so blue-sky ideas don't sneak into the must-haves.
• Assumptions and Exclusions, the boundaries that prevent scope creep three weeks in.
• Phased Timeline, with any dates the client mentioned interpolated correctly.
• Investment Summary, calculated from your rate card.

You read, tweak & send. Or, chat with the AI agent to have it run another draft. The model does the structural work. You do the judgement work.

Worked example

One of the AI consulting clients I work with runs three to four discovery calls a week. Manual proposal drafting at 2-4 hours apiece is 6-12+ hours of unbilled writing per week. Over the span of a year - that racks up to thousands of pounds of unbillable labour.

With the AI Proposal Drafter, transcript ingest is ten seconds, Gap Analysis takes three minutes, the first draft is out in under two minutes, the human edit is ten to fifteen minutes per SOW. Roughly an hour and a half total for three SOWs. The proposals get sent the same afternoon, not three days later.

I've recently implemented this in my own consultancy business and I've seen my close rate increase by a staggering 63% over the last two weeks (from an admittedly small sample size - but it's quite a sexy statistic so I'm running with it).

PRO TIP: PASS IT TO GAMMA

Pretty proposals close more clients. Especially for any creative work.

The AI Proposal Drafter outputs as a structured markdown document. To take it further - drop that markdown doc into Gamma along with your brand template, and you've got a beautifully formatted, on-brand proposal deck in under another minute - fully automated.

I may cover the full Drafter-into-Gamma flow in detail in a future issue. But for now, just know it works cleanly.

STACK IT WITH AI EDITOR IN CHIEF

If you're using the AI Editor In Chief skill from last week, point AI Proposal Drafter at your voice fingerprint when you generate the SOW. The Drafter handles structure. AI Editor In Chief reshapes the prose to read like you. Together, the two skills do a lot of the heavy lifting.

WHO ELSE IS DOING THIS

The transcript-to-document pattern isn't new. Anyreach, an AI go-to-market vendor, has published a case study describing how they generate Statements of Work for their own pilots: discovery call in, structured SOW out (cover, business context, objectives, scope, integrations, risks, roles, commercial terms). Their published numbers are self-reported, not audited - 87% reduction in proposal-creation time, 4.2 hours to first proposal versus two to three days manually - so I'd treat the figures as directional rather than gospel. The pattern is the point. They identified speed-to-lead as an issue, and built an automation to improve it.

At the enterprise end, Ironclad and Docusign CLM have both shipped AI clause-generation tools for in-house legal teams. Same instinct, but for much larger contracts and much higher price tags.

The gap I saw was the middle. A self-service, voice-matched proposal skill that any solo operator or small creative team can install once and run after every discovery call. No monthly fees to a big SaaS company.

WHERE TO START

You can get the AI Proposal Drafter via the link below. It runs natively in your LLM of choice (Claude, ChatGPT, Gemini). Install takes about 30 seconds. Input in, clarifying questions answered, first-draft SOW out in two minutes, human edit, send. Close more clients.

If you'd prefer to build this yourself, you should be able to do so using the playbook above. Take the input flow, the setup-and-operate split, the Gap-Analysis-before-drafting structure, the rate-card-fallback framework, and build it in your tool of choice. It will probably take you an afternoon. The skill above just saves you that afternoon and gives you the version I've already tested extensively and am running in my business.

Catch you next week!

Sam

P.S. If you are a creative professional running an AI workflow I could feature - designer, editor, studio owner, freelancer - drop me a line at hello@madewithmachines.com. Every issue needs a case study. Yours could be Issue #004.

​ ​