All posts
Plays·Jun 2, 2026·7 min read

We pulled 1,600 SaaS companies from a funding DB and sorted them by what they actually sell

The Verbiflow teamBy The Verbiflow team

Lists from funding databases are noisy. “B2B SaaS” covers everything from devtools to compliance to vertical CRMs. So when a customer asked us to find SaaS companies that fit their very specific ICP, we didn’t buy a list. We built a play that read the homepages.

1,600
SaaS companies pulled from a funding database
4 stages
seed → crawl → extract → enrich
1 SQLite DB
resumable, restartable, no orchestration layer

The problem with buying the list

The customer was selling into B2B SaaS, but only a specific slice: companies whose product itself touches developer or infrastructure workflows. A standard funding database returned 1,600 “B2B SaaS” companies. Maybe 200 of them were a real fit. The other 1,400 were a tax on every sequence.

Manual qualification at that scale isn’t worth it. So we built a four-step pipeline that did the qualification automatically and let the customer sequence only the survivors.

The pipeline

  • Seed. Pull 1,600 companies from the funding DB into a SQLite table. Every row has a name, a domain, and a funding stage.
  • Crawl. Visit each company’s homepage, customer page, and case studies with Playwright. Extract clean markdown.
  • Extract. Hand the markdown to Claude with a structured-output prompt: who are this company’s customers, what does the product do, what category does it fit. Write the answer back to the DB.
  • Enrich. For each extracted customer, look up funding stage via SERP. Roll that back up to rank the parent company by ICP fit.
Why SQLite
Every stage reads and writes the same DB. Kill any step mid-run, re-run it, and it picks up where it left off. No orchestration framework, no queue, no temporal worker. The pipeline is four Python scripts and a database.

What the customer actually got

Three ranked CSVs, exported from the DB:

  • Top 100 by ICP fit. Companies whose own customers were the customer’s exact buyer profile.
  • Series A targets. Subset sized by funding stage, where the buying committee is small and the deal cycle is short.
  • Series B and below. The broader long-tail to sequence at lower priority.

The customer loaded the top 100 into Verbiflow as the audience for a new sequence, and let the lower tiers feed into a slower-cadence track. Sequencing started the next day.

Why this isn’t “just a list”

Two things are different from buying a list. First, the ranking is based on what the company actually does today, not on what someone tagged it as five years ago. Reading the homepage gets you a fresher signal than any vendor’s industry taxonomy. Second, the pipeline is re-runnable. Six months later, when the customer wants the same view but with new entrants, re-run the four scripts. The DB updates itself.

The best lists aren’t bought. They’re built once, well, and re-run when the market moves.

The bigger pattern

This is one of the “plays” we build with customers. Other plays in the same family: every regional security firm in the US sorted by Google review sentiment; every YC company’s trust page read for compliance gaps; every B2B SaaS site with a careers page that mentions “Series A.”

Different shapes, same idea: data that’s public if you go look for it, structured into a sequenceable audience, then run through Verbiflow’s outbound stack. The play does the qualification. The platform does the sending. A CSV or an API call connects the two.