AI Wins Me the Deal—One Client Interview, a Full Proposal in a Dozen-Odd Minutes

There’s a line I strongly agree with: never get lost in AI as a technology, chasing every update. Put AI into real business instead—create value, and turn that value into money.

However brilliant your idea, however impressive the thing you built—building it isn’t the impressive part. Selling it is.

Whether you’re a million-a-year executive, a one-person company, or a founder, nothing in daily business is more of a headache than client meetings, sorting out requirements, reporting, pitching. So today, let’s see what the very best agents and models can do for us.

1. When you save the meeting, two things are worth the most

Every mainstream online meeting tool lets you save the meeting locally. Two of the files it leaves behind matter:

• The transcript text—accuracy may only be 60-70%, but it preserves attendee names and timestamps, and that’s a useful anchor
• The raw audio file—so we can re-run recognition with a better model

2. Transcribing with Gemini 3 Flash—but cut it into chunks

I had it turn the meeting audio into plain text with Google Gemini 3 Flash. The first pass took all 24 minutes in one go, and at just past the 9-minute mark it hit attention collapse. Recognition fell apart.

The standard fix: cut the audio into chunks with FFmpeg, usually 7-8 minutes each, then run recognition. This time it came out clean—who said what, all correct.

Why it beats ordinary meeting software: we fed in the context up front—meeting background, attendees, where the discussion was headed—so the transcription model had something to lean on.

3. Drop 2.5 hours of meeting notes into Claude Code

That meeting wasn’t a full client conversation. So I took the notes from an earlier session where I’d talked with a client for two and a half hours, dropped them into Claude Code, and gave it the brief:

This text is my on-site conversation with a client. I prepared a lot of questions and interviewed him—his upbringing, his career, his company’s clients and business.

The goal is to help him with a complete analysis of his domestic China business model + the full business canvas. Please spin up a multi-agent team to research the business model from every angle, and call up my business canvas analysis experience for the full write-up.

Also map out his upbringing and personal traits. Next I need to do complete creator-business positioning for him, analyze topic dimensions, and produce sample topics—the skills have the matching experience for that too.

When it’s done, present it as a web page. Real case—redact the client name, only say what business they’re in.

4. The client: a Singapore firm helping Chinese companies go global

This client is a Singapore company that specializes in helping Chinese firms expand abroad, and the founder wants to build a personal brand to grow business back in China. As his consultant, my job is to:

• Run industry research for him first
• Do a business canvas analysis against where he stands today (that method is already a skill, so I just call it)
• Once I know the client’s details, work out personal brand positioning + what makes him different
• Help him pick his topics

All of this knowledge has been turned into skills. If you’re interested, leave a comment asking for the skill.

5. A complete proposal page in a dozen-odd minutes

Task done. Open the web page—“Building certainty and a sense of safety for Chinese entrepreneurs in Singapore.”

Business operations analysis

• The client’s clients, their pain points, the value returned
• Demand segmentation—three vertical customer groups
• Market research—macro trends, competitive landscape, competitive advantage
• Sizing—TAM / SAM / SOM
• Core selling points, solution, product substance, delivery flow
• Business model, unit economics, core metrics
• Acquisition and growth—the WeChat Channels conversion path / the Xiaohongshu conversion path
• Building moats, finding partners and channels
• The key assumptions the business rests on, and how to test and validate them

Creator business plan

It settled on five dimensions, each with sub-directions, and each direction with 10 topics. Complete proposal, delivered.

6. Why a complete plan comes out in a dozen-odd minutes

Because what Claude Code had wasn’t just this one meeting:

• Input: the content of every consulting conversation I’ve ever had (even the pre-meeting questions were AI-prepared)
• Output: the report I sent each client afterward—used as the best-practice reference
• And I added: how to search and research, how to think, how to build up to the final report step by step

Once those examples were in, I distilled my own expertise in business canvas analysis, creator-business positioning, and topic selection into skills. That’s how this got done.

7. A client is only an opportunity, not a sure thing

This one matters. Running a one-person company or a startup, openings like this come at you constantly and you can’t tell which ones are real. You have no choice but to try, and trying means sinking real time into prep.

This client only paid for a single consultation. But working this way, I handed him years of accumulated experience fast—not a framework anymore, closer to a full plan.

So a genuinely high-value client conversation now costs me almost no time, I get to screen clients quickly, and off one conversation the client trusts my expertise completely → the long-term relationship locks in.

AI is here to help us. It makes what I can do bigger, more repeatable, faster. It does not replace the person. Everything AI pulled off today rests on my own expertise.

Never forget it—use the best models and the best AI tools to extract your own professional expertise, then point it straight at your real business.

Source: EP0013_audio.mp3 · ASR model gemini-3-flash (chunked parallel) · full text of the 9:34 original recording

[00:00] There’s a line I really agree with: never get so absorbed in AI tech that you’re endlessly chasing the latest AI update—instead you should apply AI tech to real business, create value, and turn that value into money. No matter how brilliant your idea is, or how impressive the product you built is, building it isn’t the impressive part. Selling it is.

[00:16] Whether you’re an exec on a million-a-year package, or an OPC, or a founder, nothing in the day-to-day is more of a headache than meeting with clients, then sorting out requirements, reporting back to the client, pitching. So today let’s look at what the very best agents and models can do for us in that process.

[00:32] Now, the usual online meeting tools all have this feature where you save the meeting content to your own machine. And the most valuable parts of that content are really two things. The first is the full-transcript text. And although it’ll have errors in it—the accuracy is usually only around 60-70%—it’s still an important reference. Usually that kind of transcript leaves you the attendees’ names and the timestamps, and that’s genuinely meaningful. And the second is the pure audio file.

[00:58] And that file lets us easily run the content through a better model to transcribe it. So let’s actually demo it. I’ve already downloaded the file—I’ve downloaded the audio file from my last client conversation to my D drive.

[01:16] Please use the Google Gemini 3 Flash model to turn the meeting content into plain text for me. This meeting was my conversation with a client, Teacher Chen, about how, at the current trade shows, clients in the health and wellness industry can do better at acquiring customers and recommending services, using AI agents to handle pre-sales consultation work. Good, so let’s do this meeting-notes transcription first.

[01:50] Alright, let’s look, it’s finished. This process actually didn’t go all that smoothly—to record this tutorial for you, I cleared out some of my skills and rolled some memory back. The first time around it used the Gemini 3 Flash model to transcribe the whole 24 minutes in one pass. And what we found was that around the 9-minute mark, because of the attention-collapse problem, its recognition went bad.

[02:08] So when we do these recording-to-transcript jobs, the usual practice is to use FFmpeg to cut the audio into multiple segments, usually seven or eight minutes each, and then have it recognize them that way. And we can see the recognition result is very, very good, really good. Who said what across the whole conversation, all of it, done.

[02:29] So why would a recognition result like this be better than what the ordinary online meeting software produces? The reason is that at the very start, we fed in the meeting’s background information, its participants, and roughly what direction they discussed. So during this recognition run, that gives the transcription model some extra context as support, which lets it recognize the meeting content more clearly.

[02:50] So next let me do another demo, because that meeting just now wasn’t actually a complete client conversation. So I’m going to bring in the notes I took from an earlier conversation with one of my clients—about two hours plus, two and a half hours or maybe three—and show you how I used those meeting notes to produce a consulting report.

[03:18] So let’s expand these meeting notes. And looking at these notes—right, notes from a two-and-a-half-hour meeting—we drop them into Claude Code, and then add: the text file you’re looking at now is an in-person conversation between me and a client, where I prepared a lot of questions and interviewed him.

[03:43] Including his upbringing and his personal career, plus his company’s current clients and business. And my purpose in doing this is to help him fully analyze his business model in China, and the whole business canvas. So please organize a multi-agent team and research this business model from every angle.

[04:14] And at the same time we want to lay out his personal upbringing, including some of his personal characteristics and traits. And next I’m going to need a complete positioning for him on this creator-business thing, plus help working out the angles he could take and putting together some sample topics. And for this, I’ve got matching experience in my skills you can pull from.

[04:39] And once you’ve finished pulling this information together, I need you to give me a complete presentation of it, in the end, as a web page. And since this is a real case, I need you to hide the client’s name in the page—just mention what kind of business it is, that’s all. And if you have anything you need clarified you can ask me, and if not, take the goal I just gave you, the finished web page, as the benchmark, and report back to me once we’ve finished this project.

[05:01] Right, so you can see I ran an interview with a client. And this client is a Singapore company that specializes—very professionally—in helping Chinese businesses go overseas. And their founder wants to build a personal brand to expand his business in China. So first, as a consultant, I obviously have to research the industry for him. And then, given his current situation, do a business-canvas analysis.

[05:25] And this analysis method I’ve already turned into a skill, inside Claude Code, so it can just call it directly itself. And at the same time I have a very complete process for helping a client with personal-brand positioning once I know their details, plus some analysis of their differentiated advantages. And it can also help them with topic selection. And all of that knowledge has been turned into skills. So if you’re interested, leave a comment and I’ll share what’s in these skills with you.

[05:52] So now we just have to wait for it to finish the web page. Let’s fast-forward again. Good, the task is done. So let’s look at this, at this web page file.

[05:59] Chinese entrepreneurs—building a sense of certainty and security for Chinese entrepreneurs in Singapore. OK. This is a report built from the online client meeting. Part one is an analysis of the business operation. The client’s clients, what their pain points are. What their value gains are. Demand, audience segmentation, which three vertical customer groups there are.

[06:20] Market research, what the macro trends are, what the competitive landscape looks like, what the competitive advantage is, his size estimates, total addressable market, serviceable market, and obtainable market. What the core selling point is, what solution he provides, the product core, what his delivery process looks like, his business model, what his unit economics are, what the core metrics are.

[06:42] How to do customer acquisition and growth. What the WeChat Channels conversion path and the Xiaohongshu conversion path look like. How he builds a moat, how he finds partners and channels. What the key assumption is that makes this business work, and how to test and validate it.

[07:01] And then the whole content plan for his creator business, with five angles mapped out for him. And then for the topics he can do under each angle, it listed sub-directions, with 10 topics for each direction. A complete proposal, right there.

[07:19] So we can see this plan took about a dozen-odd minutes to produce, and it got to something this complete. The reason is that it takes the content of every consulting meeting I’ve had in the past, including the questions I prepared before those meetings (which AI even prepared too)—every one of those conversations goes in as input, as an input.

[07:37] And the report I gave the client after each meeting goes in as an output, called output, a best-practice reference. So it knows how I talk with clients, and how those conversations have gone in the past. And then I added how, after I get hold of that material, I search and research, how I think, how I step by step produce a final report like this.

[07:58] And once I’d given it these examples, this skill’s model—it’s the equivalent of taking my skill, my professional skill, at analyzing a business canvas and doing creator-business positioning and nailing down topics, and distilling it out. That’s what that process accomplished.

[08:14] And once that’s done—everyone please note, this client is only a lead. It’s only a lead, not a confirmed client. So when we’re doing this one-person-company thing, or in the middle of a startup—especially these last two years—there are a huge number of leads like this, but we can’t be sure they’re solid, so sometimes we have no choice but to run trial and error, and pour a lot of time into all this upfront prep.

[08:34] So this client only paid for one consulting session, but by doing it this AI way, I quickly turned my past experience into a solution framework for him—barely even a framework, honestly closer to a complete proposal. And that lets me have a really efficient conversation with the client for very little of my time.

[08:51] So I can screen clients quickly, to the point where after this one conversation the client trusts my expertise completely, and from there we lock in a long-term working relationship.

[09:01] So it’s that line again: AI is here to help us, to give me a boost, to take what we can already do and blow it up bigger and faster. So it isn’t there to completely replace people. The fact that AI can produce results like today’s is built on the foundation of my expertise.

[09:18] So you all need to keep that line firmly in mind: use the best models and AI tools to distill out your own professional ability, and use it to give your own real business a lift and make it faster.

[09:30] So that’s it for today. See you next episode, bye-bye.