Six moves, in order: pick a niche that bleeds money, build one proof system, film it working, find the businesses that need it, send DMs they can't ignore, and close. No portfolio. No cold pitch deck. Proof.
Skip SaaS and ecom. Every agency on earth is already pitching them. Local service businesses are the opposite: real revenue, obvious problems, almost zero AI competition. The rule is simple. Rich and boring beats cool and broke.
Paste this into Claude and let it do the scoring:
# Prompt: niche radar I'm starting an AI automation business in [YOUR CITY]. Score these local niches: roofers, med spas, plumbers, HVAC, dentists, landscapers, auto detailers. For each: (1) the single most automatable money leak (missed calls, no-shows, slow follow-up), (2) rough monthly cost of that leak for a typical shop, (3) average job value, (4) do they already pay vendors for software or marketing? Rank them and pick one winner for me. ✓ one niche, one city, one problem. target locked.
Sanity-check the winner yourself: read 10 Google reviews in that niche. If customers complain about "nobody answered" or "took days to call back," you found the leak.
Proof first. Portfolio never. Build a missed-call textback demo: when a call rings out, the caller gets a text back in seconds, an AI qualifies them, and hot leads get pushed to the owner. It attacks the exact leak you just scored.
The stack is n8n plus Twilio, and the demo runs on a Twilio number that costs about a dollar a month. Have Claude scaffold it:
# Prompt: proof machine Build me an n8n workflow for missed-call textback: 1. Twilio webhook fires when a call to my number goes unanswered 2. Send an SMS back within 30 seconds that sounds human, not automated 3. An AI step qualifies the reply for roofing: roof age, active leak, insurance claim 4. If qualified, SMS the "owner" number (my cell) with the lead summary Give me the exact nodes, the Twilio setup steps, and the webhook config. ✓ then call your own demo line and watch it fire
Test it until a call to your demo number gets a text back in under 30 seconds, every time. That's the whole build.
One video beats ten case studies. Open Loom, screen-record your phone and n8n side by side, and follow this script. No slides, no intro, no "hey guys."
# proof/loom-script.md, 90 seconds, no slides 0:00 call the demo line on camera, let it ring out 0:20 the SMS lands on screen in real time 0:45 reply as a customer, AI qualifies and books the estimate, no human touches it 1:15 face camera: "want this running on your line by friday?" ✓ this video IS the portfolio
Re-record until it feels effortless. Five takes is normal. The moment that sells is 0:20, when the text arrives while the viewer is still processing that the call just rang out.
Don't pitch 100 random roofers. Prove which ones have the problem first. Pull a list of roofers in your city from Google Maps (Claude can help you structure the scrape, or use Apollo to enrich owner names and emails). Then run the filter that matters:
# The verify pass 1. Take your list of local roofers 2. Call each business after hours or at lunch 3. Log two things per call: - did it ring out to voicemail? - did anyone text or call back within a day? 4. Keep ONLY the ones that rang out with no follow-up. Delete the rest. ✓ leads.csv trimmed to the bleeders. focus wins.
Call them yourself. It takes an afternoon, and every rang-out call is documented proof of the leak you're about to pitch. The businesses that answered? Delete them. They don't need you yet.
Now the outreach writes itself, because you're not opening with a compliment or a pitch. You're opening with their own missed call.
# DM / email / voicemail script "Hey [NAME], I called [BUSINESS] Tuesday at 6:12pm. It rang out and I never got a text or callback. That's probably happening with real customers too. I build a system that texts missed callers back in 30 seconds and books the estimate automatically. Here's a 90-second video of it catching a live call: [LOOM LINK] Want it running on your line by Friday?" ✓ specific time, verifiable claim, proof attached
Send it everywhere they live: email, Instagram DM, Facebook page, even a voicemail on the number that rang out. Follow up twice, two days apart, then move on. You have a whole list of bleeders. No single reply matters.
When an owner replies, don't send a proposal. Get them on a 15-minute call, share your screen, and run the demo live: call the line, let it ring, let them watch the text land. Then make one offer with one deadline.
# The offer "I'll install this on your business line and have it live by Friday. Flat monthly, cancel anytime, first invoice goes out after it catches its first missed call." ✓ one system, one price, one deadline. no menu.
Price it as a monthly subscription in the low hundreds. One saved job in most trades covers the month, and you should say exactly that on the call. First client? Trade a small discount for a testimonial and permission to screen-record the results. That recording becomes the proof video for client two.
The full library of 50+ playbooks covers niches, automations, pricing, tools, and the systems behind all of it. Grab the rest free.
Get all 50+ free playbooks