Yusuf Ahmed
← Builds
Client build · In production

Hidden Business Finder

Every seller in a vertical buys the same lists from the same databases, so by the time a new business is listed on Google Maps it has already been pitched. Public records show the business forming months earlier. The build is everything between a pile of raw records and a business a seller can actually call.

Since2026StackClaude Codepublic-record readersmodel-judgment qualificationone-button operator screenCSV exportLinksSee a real run ↗
The problem

What was actually hard

Anyone can point a scraper at a public dataset. Match volume is not lead volume: most of what clears a keyword filter is an existing business, not a new one, and a system that reports raw matches as leads is lying to its own sales team.

How it works

In order

  1. Pick a metro and a window. the operator chooses where and how far back.
  2. Read the records. permits, licences and build-outs from the public sources that cover that metro.
  3. Score for relevance. keyword and category matching narrows the pile, but never gets the final say.
  4. Confirm the business is new. the qualification layer separates a forming business from an existing one changing something.
  5. Keep only what can become a contact. licensed businesses, not private individuals named on a permit.
  6. Export. one file, from one button. The team never runs a script.
Judgment calls

Decisions that made it work

  • The highest keyword score is often wrong.One top-scoring record turned out to be a temporary Pokémon Go event. Keyword relevance and buyer relevance are different things, which is why a model judges the shortlist and the scoring model never decides alone.
  • Plenty of records can never become a contact.On permit records the named party is often the owner or applicant as a private individual. Useful as signal, unusable as a lead, so the deliverable is licensed businesses only.
  • Report leads, not matches.The output is judged on its own: three exports from one run are on the public page, and the run reproduces.
What it did

Outcomes

In production for two GTM teams. The recording on the public page is the run, then the file it produced.

Why it matters

Scanning is cheap. Qualifying is the build. The system is honest with its own sales team about the difference between a match and a lead.

What stays off this page. Which records get read, how a business is confirmed as genuinely new, and how the qualification layer decides what reaches a seller stay off this page. That is the client's edge.