Case studiesOutbound pipeline

Industrial software company

Every steel fabricator in the market, on one map.

Their reps prospected from whatever was already in the CRM, plus what they could find by hand. We merged more than twelve industry directories across four countries into one scored list, and found the good-fit fabricators the CRM had never seen.

7,817

good-fit fabricators new to the CRM, out of 10,623 above their fit bar, as of the August 2026 rebuild.

  • GTM engineering
  • Applied AI
  • Data and BI
Company
Industrial software company
Market
Steel fabricators
Work
GTM engineering, Applied AI, Data and BI
Engagement
Lead engine, summer 2026

What we found.

  • 12+

    directories across four countries held the market, each in its own format, with heavy overlap between them.

  • 1,332

    good-fit companies were hidden by one blended score, because they are hard to reach. Separating fit from reachability made them a work-list.

  • 525

    scraped “contacts” were address fragments posing as people.

What we built.

Quick wins first, then the bigger pieces, all in their own accounts.

  • One pipeline across every source

    Scrapers for trade associations, certification registers, government registries and map listings in the US, Canada, the UK and Australia, run on free sources under a strict spend rule.

    • Python
    • Playwright
    • Supabase
    GTM engineering
  • One golden record per company

    Normalisation and fuzzy-match deduplication, with a side-by-side review queue for merges and subsidiaries.

    • Deduplication
    • Review queue
    Data and BI
  • Two scores: fit and reach

    Fit weighted the way their sales team thinks about it, and a separate reachability score. A Claude-based classifier sorts fabricators from erectors, detailers and suppliers.

    • Claude
    • Scoring
    • CRM export
    Applied AI

What changed.

  • 7,817

    good-fit fabricators new to the CRM, out of 10,623 above their fit bar, as of the August 2026 rebuild.

  • 38,795

    unique companies from 52,842 raw records, with zero orphans on the rebuild.

  • 525 → 0

    junk contacts, after address fragments were filtered out of the people list.

771 automated tests guard the scoring and export logic.

The stack

The stack, all in their accounts

  • Python
  • FastAPI
  • Supabase
  • React
  • Playwright
  • Claude
  • CRM

Bring the list your reps work from.

On the first call we’ll size the market around it with you and show you what’s missing.