A cold email agency founder explains how his team replaced Clay with a custom lead generation system built entirely with Claude Code — processing 272,000 leads per second at a fraction of the cost.
Why They Outgrew Clay
- Scale limits: Processing 9 million leads/month for one client alone; Clay caps tables at 50,000 rows and workspaces at 12.5 million rows
- Operational friction: Deleting tables took days, requiring thousands of "run all" clicks and weeks of pre-work before campaign launches
- Economics: Clay was exploring per-row charges for custom HTTP columns, making the cost trajectory unsustainable
The Tech Stack
- Claude Code ($200/mo): Replaced Cursor (was $450/day for two people); James, with zero coding experience, built the core system in one week
- GitHub: All code stored and version-controlled
- Railway (~$2K/mo): Hosts 50+ parallel workers that process leads 24/7 — 1 million leads in 5 seconds vs. Clay's 27 hours
- Vercel: Hosts dashboards and visual interfaces, auto-syncs with GitHub pushes
- Postgres → Convex: Migrating to Convex for better real-time data handling
Custom Tools They Built
Google Maps Scraper
Scrapes business leads zip code by zip code (32,000+ zip codes) instead of by city/state to get complete results. Built in 3-4 hours with Cursor.
AI Lead Finder
- When primary data sources (like Arc) don't have a match, AI searches the entire internet for contacts
- Always tries to find 3 leads per company with LinkedIn profiles, confidence scores, and fit reasoning
- Falls back to personal emails or contacts at other companies the person works at
- Cost: $0.002 per company (a fifth of a penny) to find 3 qualified leads
- Gets 95%+ valid email rate vs. typical 30% from Apollo/LinkedIn
Ad Library Scrapers
Scrapes Google and LinkedIn ad libraries to find companies actively running ads — warmer leads since they already have acquisition budgets allocated.
Executive Summary System
- AI analyzes all campaigns daily, providing snapshots of client performance, email-to-lead ratios, and meeting generation
- Working toward copy analysis that recommends subject lines, hooks, and messaging by ICP
- Will programmatically create campaigns based on historical performance data
Private Lead Database
- Nearly 50 million leads cataloged across all clients
- Tracks which data vendor performs best for each ICP and industry
- Auto-refills campaigns when leads run low and cleans old leads to reduce Instantly billing
Key Takeaways
- Start with one pain point: Don't try to build everything at once — solve your biggest bottleneck first, then expand
- You don't need to code: Neither the founder nor James who built the system know how to write code; Claude Code handles it
- Internal tools only: These are private tools, not production apps — security isn't battle-tested for public use
- Ask Claude good questions: Point it at APIs, ask for step-by-step setup guidance, and be willing to learn
"You don't need to understand the code. You just need to know what you want to accomplish."