You commented JOB, so here's the build from the reel: a form in, Claude in the middle, the result out. Then you charge per result instead of £29 a month.
I'll walk through it with one real example, start to finish: an AI follow-up service for a local trades business. Swap the job and the same shape works for bookings, invoices or quotes.
Why this works when the app doesn't
A business owner won't log into another tool. They already pay someone (or themselves at 10pm) to chase leads, book jobs and send invoices. They don't want software. They want that job to be done.
So you don't sell access. You sell the finished job, one at a time. The software is your cost, not their product.
Step 1: Pick the job (the 3-question test)
Only build it if all three answers are yes:
Is a human doing it today? If nobody does it now, nobody will pay for it.
Can you count one finished result? "One follow-up sent" counts. "Better marketing" doesn't.
Is a lost result expensive? A lead that never gets a follow-up is a lost £500–£5,000 job. That's why they'll pay £5 for a follow-up.
Follow-ups pass all three for most trades, clinics and agencies. That's why I start there.
Step 2: Build the back end only
1. Form in. Use the client's existing website form or a free Tally form. Every new enquiry fires a webhook. That's the only thing the client has to change, and you can do it for them in 10 minutes.
2. Claude in the middle. An n8n workflow (or one small script) receives the webhook, adds the client's business rules and sends both to Claude. Claude returns the finished follow-up.
3. Result out. The follow-up goes out as a text or email from the client's own number or inbox. The client gets a copy. Nothing to log into.
4. Log every result. Write one row to a Google Sheet for every result: date, client, lead, what was sent and whether they replied. That sheet is your proof and your invoice.
The prompt in the middle
This is the core of my follow-up prompt. Fill in the brackets per client:
You write follow-ups for [BUSINESS], a [TRADE] in [AREA]. A new enquiry just came in: [FORM DATA]. Rules: reply in under 60 words, sound like the owner, mention the specific job they asked about, offer two time slots from [AVAILABILITY], never quote a price, never promise a date you can't see. Output only the message, ready to send. If the enquiry is spam or out of area, output SKIP and the reason.
The SKIP line matters. It stops you billing for junk, and that's what makes per-result pricing trustworthy.
Step 3: Price per result
Set the price at 1–5% of what a won result is worth. If a lead is worth £1,000 to the client, £5–£20 per follow-up is easy to say yes to.
Bill monthly from the sheet. Count the rows, send one invoice or Stripe payment link, and attach the sheet.
Do the maths before you build. 10 clients × 20 results × £10 = £2,000 a month. Your Claude cost on that is a few pounds.
The first two weeks: run it with a human check
For the first 14 days, every message lands in your inbox for a one-click approve before it sends. You'll catch the edge cases (wrong tone, missing details), tighten the rules and then switch to auto-send. Most builders skip this step, and that's how they lose the first client.
What I'm not covering here
That's the shape, and it's enough to build your first one. What takes it from a working demo to something clients keep paying for:
The n8n workflow itself, node by node, including retries and the approve step
My prompt pack for the other three jobs: bookings, invoices and quotes
How I find the first three clients, and the exact message I send them
What to do when a client says "why don't I just use ChatGPT?"
I'll cover these in the next issues of Automate & Hustle. You're already subscribed, so they'll land in your inbox.
Reply to this email with the job you're thinking of building and I'll tell you what I'd charge for it.
Don't ship the app. Ship the result.
Mike