I ran two builds live on this call, at the same time. One was a tool to find every franchisee owner in the country. The other was a set of outbound campaigns for a company that scans ad creative for legal exposure. Two different problems.
They turned out to be the same problem.
The data you want is almost never sitting in one clean place you can buy. But the ground truth is usually public. You find the root, chunk the work, and let the agents go build the thing that doesn’t exist yet.
There is no database — find the legal root
One member had been hunting the owners behind the franchises — the person who runs a hundred Taco Bells, not Taco Bell corporate. That ownership hides behind shell LLCs, one rolling into the next, and there is no master list you can buy. He’d already found roughly five hundred multi-unit operators and confirmed a fraction of them; the rest were four hundred anonymous LLCs he’d have to chase one at a time.
Here’s the move: there is no global database, but there is a legal root. The FTC makes every franchisor publish a Franchise Disclosure Document, and Item 20 is the one public, operator-grain list of who actually runs the units — current and former franchisees, with name, city, state, and business phone. It’s ugly. The roster is a detached exhibit, often OCR-scrambled into the wrong columns, and every outlet is its own SPV LLC. But it’s real, and it’s free. Fourteen state portals publish those registrations at the operator grain. And somebody has already posted an MIT-licensed corpus of the documents to HuggingFace — which means the first wave doesn’t even have to scrape, because the documents already exist. Run the whole thing end to end and roughly 1.3 million roster rows collapse into about 143,000 operators. That’s the database the vendors sell for real money, rebuilt from the free legal root. The moat was never buying it. It’s the extraction and the entity resolution.
You still gut-check it. Before you trust a corpus like that, open one state — pull Rhode Island, count the filings yourself with a PDF reader, and confirm the numbers are current. If it’s ninety-five percent right and refreshed, use it. If it’s stale, you rebuild it. Either way you never trust a listing you haven’t opened.
A better starting place beats “any industry, then qualify”
A consultant on the call had just landed his first client — a startup in the synthetic-data space, the kind that makes safe, statistically-faithful replicas of sensitive customer data so companies can train models without touching the real thing. The client sells to “anyone with sensitive customer data,” which is a total addressable market the size of the sky.
His instinct was the standard one: pull the usual firmographics into Clay, then qualify down with research agents. That’s where most people start, and it’s the weakest place you can start from. Where you begin decides where you end. A stronger root: aim at the Fortune 1000 and pull SEC filings — who is writing about this exact problem in their own disclosures? Now you’re starting from companies that already told the government the pain is real, instead of a giant list you whittle by guessing. The play is always to find the authoritative source that already names who you want, not to start broad and filter your way down.
One warning that came out of the same conversation: a big logo on a startup’s customer page is very often a free-trial signup or a founder’s old employer, not a real account. Don’t build your targeting on somebody else’s marketing.
Chunk the work into waves
Neither build works if you make one agent hold everything.
The way I picture it: if you send someone for a coffee and they narrate every step — “I’m taking a left, now a right, now a left again” — you’d tell them to stop reporting and go do the job. Most agent setups make the manager sit through exactly that narration, and the main agent gets dumber as its context fills with status updates.
Dynamic workflows fix it. You kick off a set of tasks — one finds where the data lives, one figures out how to structure it, one goes after the state-level sources — and each burns as much context as it needs, verifies its own work, fixes what’s broken, and only then reports the finished result up. The main agent gets to sleep. For the franchisee build that’s three clean waves: find where the franchise documents are, parse them, then resolve the LLCs and find contact info. Higher-fidelity subtasks get their own chunk — the way my PDF skill cuts a page so each context window only ever sees one row. In Claude Code you turn it on with ultra code or /effort.
Who Gets This
The worked examples below are the actual builds from the room — what we made, what I told each person to do with it, and the exact moves.
Free: the shift — find the root, chunk the waves, build the list that doesn't exist.
$50/mo (most readers start here): everything below the line in this post, and every paid article.
$2,499/yr: Every tool I ship. Edge Copilot is how you talk to all of it through Claude Code. Current tools: Edge Copilot, AutoClaygent, Agent 7, Who to Target and What to Say, Blueprint Cloud, Technology Finder, Video List Extractor, Competitor Monitor, LinkedIn Engagement, Domain & LinkedIn Finder, Dossier Builder, PDF Contact Finder, TAM Contact Harvester, Find a Rep, Blueprint Playbook, Crawford, JoJo. Whatever ships next is included. Plus all 3 courses + weekly Applied Office Hours. (Go annual — $2,499/yr.)
Below the line in this post:
Build the campaign backwards from the biggest fine — and why the CEO gets it and the compliance team doesn’t
Work a conference before you land: scrape the networking app, score, call ahead, host the dinner
Which tools to buy, which to scrape, and what Blitz is actually for
Every question from the room, with the answer
Every week I run Applied Office Hours on Zoom — bring what you're building and we'll work it live.
