It doesn't have to be email

For me, the problem turned out to be email - that's what came up on the first call, so that's what the audit went after and that's what most of my examples are. Yours could be scheduling, quotes, social content, or the fact that the whole business lives in someone's head. The method is the same. Wherever I use email as the example, read it as "the thing they named".

The shape of it

Everything below is that table, expanded, with the templates in the appendix.

Getting to a free audit

1.Start with people who already know you

My first paying client is someone I used to pay. She supported me through my pregnancy and the birth of my son, and we'd kept in touch every six months or so since. I wasn't cold pitching a stranger, I was emailing someone who already liked me and had already seen me be competent at something.

That second part matters more than the first. Warm isn't "knows your name". Warm is "has watched you think". People you've paid, people you've worked beside, the friend who runs a practice and has heard you talk about your job at dinner.

The exercise, and I mean actually do it: scroll your phone contacts from A to Z and write down every person who runs something on their own. Not "could use AI". Runs something. You'll get to 10 faster than you expect. For what it's worth, the 40 cold emails I sent in the same fortnight are still sitting at zero conversions.

2.Ask for advice, not for work

The email I sent is in Appendix B, near enough word for word. The important part is what it isn't: there's no pitch, no deck, no "I'd love to help you with". It says I've started a business, I'd love to pick your brain, could you squeeze me in for 20 minutes, no stress if you're flat out.

It works because it's true. I did want her brain. Someone who has run a business on her own for years knows things about the admin side that I didn't, and going in as the one asking questions changed the whole temperature of the call.

3.The first call: be honest, then shut up

I told her what the business was shaping up to be - two sentences, not a pitch - and that she was one of the first people I was talking to. Then I asked what actually eats her week and let her talk.

Questions that got the most out of the call:

• "Walk me through last week. What did you spend time on that wasn't the actual work?"
• "If you could hand one job over tomorrow and never do it again, what would it be?"
• "What have you tried and given up on?"

Listen for three things. Anything they complain about twice. Anything with a date attached. And the phrase "it's all in my head", or any version of it - that's where the real problem usually lives.

This is where your audit gets its subject. Mine was email. Yours is whatever they name.

4.The offer, in the last five minutes

Near the end I said something very close to this:

"There's a lot to unpack here. Would you let me do a proper audit for free - 90 minutes on a screen share, I map out how it actually works and tell you what I'd fix - in exchange for a testimonial if it's useful?"

She said yes on the spot.

Why it lands: they risk nothing, and you're not asking for free work forever. Say the second part out loud on the call - the audit is free, and if there's something worth building afterwards, that part has a price. It's more comfortable for them than pretending.

Before the audit

The pre-audit prep form

13 questions, sent as a form a week before the call. The full thing is in Appendix A, ready to paste into Google Forms or Tally. What each group of questions is doing:

Keep it to one screen. The point isn't the data, it's that they've already started thinking about their week before you turn up.

Ask for real examples of the thing

10 to 20 real examples of whatever they named. If it's email, 10 to 20 emails they've actually sent - enquiry replies, follow-ups, a decline or two. If it's quotes, 10 quotes. If it's social, the last 10 posts. Not tidied up. This is what the demo runs on. You can't show someone their own voice, or their own pattern, without it.

What you build before the call

Three things. None of them take long, and the call is very different with them than without.

1. A pre-filled process map. A table - trigger, decision, action, tool, friction - filled in from the form and the examples, for the two or three flows you think matter. It will be wrong in places. That's the point; see step 3 in Part 3.
2. A run sheet for yourself. The order of the call, the questions per section, the timings. 90 minutes goes quickly and the commercial part is at the end, which is exactly the bit you'll run out of time for if you don't plan it.
3. The demo, tested. Whatever you're going to show them, build it on their real examples and test it before the call. For email, that was a Claude or ChatGPT project holding two files kept separate: a voice file (how they write - openers, sign-offs, rhythm, derived by counting, not by impression) and a context file (what they know - prices, packages, boundaries). I tested it on one email I held back: gave it that situation, compared the draft to what she'd actually written. If it can't reproduce a real one, it isn't ready. Have a backup version ready in case the live one falls over. Mine did, briefly.

Running the 90 minutes

Screen share throughout. Record it - the page you send is built from what they said, not what you remember.

The five steps

1. Pick one pain, not the business. On the first call they named a few things. Take one - the one they complained about twice, or the one with a date attached - and say out loud that you're parking the rest. An audit of "the business" produces a list of observations nobody acts on. An audit of one thing produces a number and a fix. Mine: emails only, everything else parked.

2. Watch, don't interview. Don't ask "how do you handle X?". Say: "Show me the last one that came in, and walk me through what you'd normally do. Including the boring bits." Then shut up. Most of what you learn comes from the screen - the tab they open first, the thing they copy from, the thing they retype every time. Make them go through every step, painfully, even when they say "and then it's just the usual". Mine: the last enquiry, from opening it to pressing send. Then how she schedules, then how she sets up her course emails.

3. Map it while they watch. Your pre-filled table, on the shared screen, corrected live. Being visibly wrong in front of them produces better information than any question, and the corrections become a section of the page you send - the one that earns the most trust. Press save. I didn't, lost the map, and rebuilt it from the recording. It cost an evening.

4. Count the hours, with their numbers. For every row that costs time: how often (a week or a month) and how many minutes each. count × (4.33 if weekly, 1 if monthly) × minutes ÷ 60. Ask for the busy month, not the average. Let them delete a row that varies too much to estimate - that refusal is honest and belongs on the page. Watch for per-unit versus per-batch: if "20 minutes" covers three things for one client, the row says one, not three. Getting this wrong inflates the total in your favour, which is the worst direction it can go. Read the total back to them out loud. Mine came to about 14 hours a month, and most of it was four or five emails per client just to land a meeting time.

5. Show them one thing on their real stuff. Not a deck. One live demo on one real example. The demo isn't about the tool. It's the moment they see their own work come back at them, done, and realise the thing is possible. Mine: she dictated why she was turning down an enquiry, in her own words, and the project drafted the reply in her voice. About five minutes for something she said normally takes her 30.

Adapting it to their business

The five steps don't change. What changes is what you watch, what a row in the map is, and what you demo. Find the row that matches what they named on the first call:

Three questions decide what goes in scope, and they're the same for every row:

1. Did they name it, or did you raise it? Only the first is demand. "That would be amazing" when you describe something is enthusiasm about a hypothetical, not a request.
2. Does it bite this quarter? If they're coping now and it won't hurt for months, it goes in "later, not now" with a reason. Something built before the pain exists becomes furniture.
3. Are they already doing it themselves? Then you don't charge for it. If it stays in for a technical reason, say why on the page and offer to take it out.

And one thing to say out loud before you leave the call: what you won't touch. Every business has a part that is the product - the consultation, the session, the personal message at the right moment. For mine it was the meet-and-greet and the hour she spends finding the right research paper for a particular client. Naming the parts you'd never automate is what makes the rest of your proposal believable.

The page you send the next day

Send it while it's warm. One web page, not a PDF and not a deck.

The top half is what they decide from. Their number at the top - the hours you counted together. Two problems, in their language. One card per thing you'd build, numbered in the order you'd build it. What it costs. And what happens if they say yes: real dates, and the one thing you can't start without.

The bottom half is the evidence. How each piece actually works. The full hours table, as recorded. What you had wrong before you watched them work. The things you're deliberately not proposing, and why. What happens next.

Three rules that hold the page up:

• Name the outcome, not the mechanism. "Course emails that send themselves", not "a Wix automation". They're buying the result, and the tool might change.
• Say what each piece can't know. Every automation is blind to something. A booking link can only see what's actually in their calendar - if their shifts aren't in it, it will happily book a client over one. A drafting system can't know a price they changed last week unless someone tells it. Put those on the page, in their words, before they discover them. It makes the rest of the page more credible, not less.
• Every number and every quote traces back to the recording. That's where the page gets its authority - it quotes them, it doesn't describe them.

The prompt that writes it

You don't need to be able to code. Appendix E is the prompt I'd give Claude or ChatGPT to turn your call recording, your map and your hours table into that page as a single HTML file. It has two deliberate stops in it: it reads everything and reports back before writing anything, and it shows you the top half before it touches the evidence. Both stops exist because the value of the page is your judgement on the findings, not the prose. A model can write a polished page from a bad scope in one shot. Don't let it.

Getting a link. In Claude, ask for the page as an artifact and use its share link. Or save the HTML file and drop it on Vercel or Netlify's free tier - both give you a link in under a minute. Two honest notes: a free link is public to anyone who has it, so keep the client's name out of the URL and add a noindex tag so it stays out of search. And open it on your phone once before sending. That's how they'll read it.

The covering email

Short. They read it on a phone between other things. It carries the link and where to stop reading, one line on the headline finding, an invitation to mark it up ("I'd rather be corrected than have you be polite"), and the next call with an escape hatch on the time. Leave out anything that turns it into a second pitch. Template in Appendix F.

The follow-up and the yes

Give them a few days. I sent the page on a Wednesday and we spoke the following Monday. Don't chase in between.

The decision call. Ask what they'd change before you ask what they think. If they want to drop a piece, ask why before you answer, because "drop that one" has two meanings. No time for it - park it, or reduce what it asks of them. They wanted it; deleting it solves the wrong problem. Not worth it to them - cut it. Ask for the one thing you can't start without on this call, not afterwards.

What mine cost, with an honest caveat. $1,300 for the full build, $750 for just the drafting system. Half to start, half when it's working and she's happy. Plus a guarantee on the riskiest piece: if the drafting system doesn't sound like her, she doesn't pay for that part.

I'll be straight with you: I didn't have a framework to price from. I said a range out loud on the audit call before I'd properly scoped anything, and then came in at the bottom of it. I'm still learning how to price this properly, so please don't take those numbers as what you should charge, and please don't treat them as a ceiling. They're what a first client cost when I had no track record. The guarantee is the one part I'd copy - it took the risk off her, and it's the reason the yes was easy.

The terms that made it easy to say yes. Deposit invoiced the afternoon of the yes, work starts that week whether or not it's cleared - the deposit confirms the slot, it doesn't gate the work. Balance on completion. The guarantee above.

She said yes to the full build.

The tools I used

Most of this is free. Where a paid plan actually matters, I've said so.

What I'd stand behind.

1. Talk to a potential client like you're building a relationship, not making a sale. The first call wasn't a pitch, the audit wasn't a pitch, and the page wasn't a pitch either - it told her what not to buy as well as what to buy. For a client-based business, being the person they trust to tell them the truth is the highest-return thing you can be. Everything else follows from it, including the yes.

2. As a woman, reaching out to women-run businesses has been the easier way in. More open to a conversation, more willing to say "I've got no idea how to use this thing", more supportive of someone starting out. That's been my experience, and it shaped who I emailed first. If it matches yours, use it.

3. Be confident in what you can deliver, and show up as yourself. I told her she was one of the first people I was talking to. I said "I don't know" on the call more than once. I was wrong about her business in front of her and let her correct me. None of that cost me the client - it's most of why she trusted the parts I was sure about.

4. Don't be discouraged if the free audit doesn't turn into a paid gig. A testimonial is excellent to have on your website, and you will learn a ton from the first audit alone - about how a real business runs, about what people actually struggle with, and about how you work under a bit of pressure. Mine converted. Yours might not, and it's still worth doing.

If you run one, I'd genuinely like to know how it goes.

The templates

You are helping me write the deliverable from an AI audit I ran for a small business owner. The output will be ONE web page, as a single self-contained HTML file, that they decide from. I will host it and send them the link.

FIRST, before writing anything, read everything I've attached and report back to me in plain text:
1. The hours table exactly as recorded, with the total. Flag any arithmetic that doesn't survive the transcript, and say which direction the error runs.
2. The lines where they described the problem in their own words. Quote them. These carry the page.
3. For each thing I could propose building: did THEY name it, or did I raise it? Quote the moment either way.
4. Anything they said they're already doing themselves.
5. Anything with no urgency - they're coping now and it won't bite for months.
6. Any deadline of theirs, and any comparison they made (a VA, a tool they pay for, a person they'd hire).
Then STOP and ask me what goes in scope and what the price is. Do not draft.

SECOND, once I've answered, write the top half of the page only, and show it to me before going further:
- Their hours, at the very top, as a big number with a simple bar per row.
- "What is actually going on": exactly two problems, in their language. If one has no fix in what I'm proposing, say so.
- "What I would build": one card per piece, numbered in build order. Title names the OUTCOME, not the mechanism, and must not overclaim. Body says what they do, what it does, what they do next. Add a line for what it can't know, and one thing to check first, only where that's real.
- "What it costs": the full build, and one lighter option. If a piece is something they could do themselves, say so and show what comes off the price.
- "What happens if you say yes": a week-by-week plan with real dates (check the weekdays), when the deposit is invoiced and that work starts that week regardless, and the ONE thing I can't start without.
Then STOP and let me review.

THIRD, the evidence, below a clear break:
- How each piece actually works, one section per card, same numbering.
- "Where the time goes": the full hours table as recorded, plus a line for anything we deliberately left out because it varies too much.
- "What I had wrong before I watched you work": every correction from the map. Never cut this section.
- If they abandoned a tool: why it wasn't their fault, and what's different now.
- "Later, not now": what I'm deliberately not proposing, and why, worded so it credits them for coping.
- "What happens next".

RULES that apply to every line:
- Every figure, quote and characterisation must trace to a line in the transcript or the form. If you can't point at it, don't write it.
- Never invent a figure, especially prices they mentioned, client counts, or anything attributed to them. Ask.
- Write it as me, to them, in plain warm English. Second person. Their vocabulary, not consultant vocabulary. Australian spelling. Spaced hyphens, never em-dashes.
- No sections they'd have to ask about. If a term needs explaining, explain it in one line.
- Style: warm off-white background, near-black text, one accent colour, a serif for headings and a clean sans for body. Numbered section labels. Mobile first - they'll read it on a phone. Add <meta name="robots" content="noindex">. No external scripts, no frameworks, no images.