Video 1 — Claude Code Lead Machine

Generated by Bloop 🫧 · S&V Preview Hub

Video 1 Analysis: "How I Get Unlimited Leads Using Claude Code (For Cold Email)"

Channel: Sales Automation (likely Eric Nowoslawski / Fixer team)
URL: https://youtu.be/Vo9VUnzYqpw
Topic: Building custom lead generation infrastructure with Claude Code to replace Clay at scale

📋 OVERVIEW

A cold email agency operator who was Clay's largest user (17.3M API hits/week) explains how his team—none of whom are developers—used Claude Code to vibe-code an entire custom lead generation system in roughly a week. The system processes 272,000 leads/second (vs Clay's 27 hours for 1M leads), costs ~$2,000/month to run, and includes Google Maps scraping, AI lead enrichment, ad library scraping, campaign analytics, and a 50M-lead private database. The speaker's core message: you don't need to code to build this—Claude Code and a clear pain point is all you need to start.

The ONE takeaway: You can replace expensive SaaS tooling (Clay, Apollo) by vibe-coding custom internal tools with Claude Code, even with zero coding experience, and the cost/speed advantage at scale is massive.

🎯 MAIN POINTS

1. Why They Left Clay [0:00-2:18]

Processing 9 million leads/month for Fixer AI alone
Clay limits: 50,000 rows per table, 12.5M rows per workspace
Deleted tables take days to actually clear
Clay was exploring charging per custom HTTP row/column
Clicking "run all" thousands of times, waiting days for processing
Decision: Build ahead of the pricing change, not after

2. The Switch from Cursor to Claude Code [2:20-2:54]

Started with Cursor ~2 months ago, spent $3K/month
Both speaker + James were spending $450/day on Cursor combined
Switched to Claude Code when it launched (~1-2 weeks prior to video)
Claude Code: $200/month per seat — "no-brainer"
James (never touched these tools before) built the entire core system in ONE WEEK

3. The Tech Stack [2:54-6:12]

GitHub — code storage & version control
Railway — deploys workers (processing engine), hosts Postgres databases. ~$2,000/month for the lead processor
50 workers running simultaneously — "little robots that process leads 24/7"
Postgres → migrating to Convex for real-time data handling
Vercel — hosts dashboards and visual interfaces, syncs with GitHub, auto-deploys when Claude pushes code
Claude Code — the builder/IDE ($200/month)
Railway for N8N — previously used Railway for N8N processing workers too

4. Speed Comparison [6:12-6:50]

Clay: 1 million leads = 27 hours (if perfect, often errors requiring reruns)
Custom system: 272,000 rows/second = 1 million leads in 5 seconds
This lets them pivot campaigns almost instantly when something isn't working

5. Custom Tools Built [7:06-14:23]

a) Google Maps Scraper [7:23-8:17]

Scrapes local business leads zip code by zip code (not city/state)
32,000+ zip codes to ensure results for every query
James built it in Cursor in 3-4 hours
After scraping companies, runs AI enrichment to find contacts
AI scrapes any public database to locate contacts at the company
Another AI layer for segmentation (multi-location? years in business? confidence scores)
Cost: $0.002 (1/5 of a penny) to find 3 qualified leads per company

b) AI Lead Finder [8:32-10:03]

When primary vendor (ARarc) doesn't have a match, AI searches the entire internet
Always tries to find 3 contacts per company
If it can't find a valid email for a person, looks for personal emails or emails at their other jobs
Returns LinkedIn profiles, confidence level, reasoning for fit
Runs LinkedIn URL through waterfall process to get email
Gets 95% valid email rate vs typical 30% from Apollo/LinkedIn

c) Ad Library Scrapers [10:16-10:57]

Scrapes Google and LinkedIn ad libraries at scale
Finds everyone running ads in last 30/60 days
Filter by how many ads a company is running
Logic: if running ads → actively trying to grow → has budget for client acquisition → warmer leads

d) Executive Summary / Analytics System [11:00-12:31]

AI analyzes all campaigns daily
Gives daily report on 15 randomly selected clients
Average rates, email-to-lead ratios, meetings generated
Building toward: analyzing copy by ICP — "what copy elements give highest multiplier for director of marketing at paper manufacturing company?"
Matches every email to a schema: subject lines, hooks, CTAs, body messaging categories, social proof categories, ICP categorization
Will generate weekly reports for clients
Looks at all sends from last week + historical data + what's working for similar ICPs

e) Instant Workspace Cleaner [12:32-14:05]

Automatically cleans Instantly workspaces (Instantly bills per stored lead)
Deletes old leads, loads proper lead lists, tracks everything
Sends "Friday analytics" report every Friday
Looking at moving to Email Bison to avoid this billing model

f) Private Lead Database [12:53-14:23]

Nearly 50 million leads
Catalogs every lead pulled for every client (unless NDA prevents it)
Most of the time don't need to buy new data — only fetch new market entrants
Can analyze which data vendor has best data for each ICP
e.g., "Lead Magic has best data for cybersecurity, Wiza performs better for e-commerce"
Auto-refill: when client gets low, system deletes old leads from Instantly and uploads fresh ones
Clients never run out of leads — runs on autopilot

6. Advice for Getting Started [14:23-16:00]

Start with ONE pain point — your biggest bottleneck
Don't try to build everything at once
Ask Claude Code "how can I do this?" — it'll walk you step by step
You don't need to understand the code, just know what you want to accomplish
Bought Claude Code licenses for entire team including executive assistant
EA uses it just to process CSVs faster
Used WorkOS for authentication on internal tools
Caution: wouldn't release these to the public — security isn't fully understood yet

7. Results / Social Proof [16:35-17:25]

RB2B: $0 → $4M ARR in 4 months, 42% of revenue from cold email
Fixer AI: $4.3M annual pipeline, could have been $32.2M ARR if hitting full TAM every 60 days
Directive Consulting: 15-20 meetings/day through cold email

💎 DEEP DIVE

Gold Nuggets — Things Most People Will Miss

"We were Clay's largest user, hitting their platform 17.3 million times per week" — This isn't just bragging. It means they have deep knowledge of what breaks at scale. If you're building cold email systems, listen to the person who actually stressed the infrastructure to its limits.
"James, who had never touched any of these tools before, built out the entire core system in a week" [2:48] — The real insight: the person who built their million-dollar lead system learned Claude Code 3 WEEKS ago. The barrier to entry for building custom SaaS-replacement tools is essentially zero now.
"When you build something for yourself, you can go crazy with it... if Clay errors 1% of the time that's hundreds of thousands of users impacted" [4:24-4:38] — This is the fundamental insight about internal tools vs. SaaS: you only need it to work for YOUR use case. Custom tools can be 1000x faster because they don't need to be safe for everyone.
"We're looking at moving everything over to Email Bison just because of that particular pain point" [12:38-12:42] — Casual mention that Instantly's billing model (per stored lead) is a big enough pain that they're migrating an entire operation. Email Bison and Smart Lead don't charge this way. Huge signal for anyone choosing a sequencer.
"She'll even use it just to process CSVs faster" [14:40-14:42] — The executive assistant using Claude Code. This reveals that the tool isn't just for building apps — it's becoming the default way to handle ANY data task for the entire team.
"For us, it was about processing speed and row limits... we'd have to do a bunch of pre-work for a month to actually get the leads ready. Now we don't need a month. We just need a few days" [15:01-15:15] — The real bottleneck wasn't just speed — it was LAUNCH TIME. Going from 1 month of pre-work to a few days means they can onboard high-value clients ($100K/month accelerator plan, 5M emails) in 1-3 weeks. The speed advantage compounds into a revenue advantage.
"All of these features to be put into Outfound if and when we decide to release Outfound.io" [14:14-14:21] — They're essentially R&D-ing their next product (Outfound) using their agency as the test bed. Every custom tool they build internally becomes a potential feature in a SaaS product.

Insinuated & Implied Information

Clay is about to become more expensive — The mention of Clay "playing around with charging per custom HTTP row and custom columns" suggests insider knowledge of upcoming pricing changes. They built ahead of this.
Outfound.io is the real play — The agency is the cash cow funding the SaaS product. Every tool they vibe-code becomes Outfound IP. The agency validates the tools at extreme scale before they become product features.
Their competitive moat is data, not code — 50M leads + performance data on which vendors work for which ICPs = compounding advantage. The code can be replicated; the data cannot.
They're making their entire cold email operation autonomous — The progression (auto-refill leads → auto-clean workspaces → auto-generate campaigns from data analysis → auto-executive summaries) points toward a system that needs minimal human intervention per client.
ARarc has replaced Apollo as their primary data vendor — Mentioned casually at [8:48] — "one of our favorite vendors now, we just recently found them and they've basically completely replaced Apollo for us."
The waterfall enrichment approach — They don't rely on one data source. They stack: primary vendor → AI internet search → personal email lookup → other-job email lookup. This layered approach is why they hit 95% valid email rate.
WorkOS for authentication — They're using WorkOS to secure internal tools, which means these dashboards are accessible to their team via the web, not just locally. This is more sophisticated than most vibe-coded tools.

Small Details That Matter

Zip code by zip code scraping — The key to Google Maps scraping is zip-code-level queries (32,000+ zips) rather than city/state, ensuring you get results for every query without Google's result limits cutting you off
50 workers running simultaneously — This is the Railway configuration that enables the 272K rows/second processing
Convex vs Postgres decision — Convex for real-time data, Postgres for complex products like Outfound. Different databases for different needs.
3 leads per company — Their target is always 3 contacts per company, not 1. This triples their chances of reaching someone.
Confidence scores on enrichment — The AI doesn't just find leads; it scores how confident it is in the match. This prevents bad data from entering campaigns.
Multi-company contact discovery — If Joe works at Company A (target) and Company B, and Company A email bounces, they'll email Joe at Company B about the Company A opportunity. Clever and unusual.
Friday analytics report — Automated weekly report to the team. Named specifically "Friday analytics."
Vercel auto-deploys from GitHub — When Claude Code pushes code changes, Vercel automatically deploys the updated dashboard. Zero manual deployment.

Mistakes, Warnings & Lessons Learned

Cursor was expensive — $3K/month for one person, $450/day for two. Claude Code at $200/month was a massive cost reduction.
Don't release vibe-coded apps publicly — Speaker explicitly warns about security concerns. These are internal tools only.
Clay's row limits will bite you — 50K rows/table, 12.5M/workspace, tables take days to delete. If you're scaling, you'll hit these walls.
Start with one pain point — Don't try to build everything. Build for your biggest bottleneck first, then expand.

Competitive & Market Intelligence

Clay — Still recommended for smaller users, but has fundamental scale limitations
Apollo — Being replaced by ARarc for lead data
ARarc — New favorite vendor, seems to have better coverage/pricing than Apollo
Lead Magic, Wiza — Different vendors have different strengths by industry
Instantly — Current sequencer but likely being replaced by Email Bison due to billing model
Smart Lead — Alternative sequencer they also use
Email Bison — Rising favorite, doesn't charge per stored lead
Outfound.io — Their upcoming SaaS product that will unify all these tools
Convex — Being adopted for real-time database needs

🔧 COMPLETE BREAKDOWN

Full Tool & Tech Stack

Tool	Role	Cost
Claude Code	Vibe coding / building all tools	$200/mo per seat
GitHub	Code storage & version control	-
Railway	Worker deployment + Postgres hosting	~$2,000/mo
Vercel	Dashboard hosting, auto-deploy	-
Convex	Real-time database (migrating to)	-
Postgres	Lead database (50M leads)	Hosted on Railway
ARarc	Primary lead data vendor	-
Lead Magic	Secondary data vendor (good for cybersecurity)	-
Wiza	Secondary data vendor (good for e-commerce)	-
Instantly	Email sequencer (being phased out)	-
Smart Lead	Email sequencer (backup/alternative)	-
Email Bison	Email sequencer (likely next primary)	-
WorkOS	Authentication for internal dashboards	-
N8N	Previously used for workflow automation	Hosted on Railway

Key Numbers

Metric	Value
Leads processed per second	272,000
1M leads processing time (custom)	5 seconds
1M leads processing time (Clay)	27 hours
Monthly lead volume (Fixer AI peak)	9 million
Weekly Clay API hits (peak)	17.3 million
Railway monthly cost	~$2,000
Claude Code monthly cost	$200/seat
Previous Cursor monthly cost	$3,000
Previous Cursor daily cost (2 users)	$450
Cost per 3 enriched leads	$0.002 (1/5 penny)
Private lead database size	~50 million
US zip codes scraped	32,000+
Simultaneous workers	50
Clay rows per table limit	50,000
Clay workspace row limit	12.5 million
Valid email rate (their system)	~95%
Valid email rate (Apollo/LinkedIn)	~30%
RB2B revenue (4 months)	$4M ARR
RB2B cold email contribution	42% of revenue
Fixer AI pipeline	$4.3M annual
Fixer AI potential (full TAM/60 days)	$32.2M ARR
Directive meetings/day	15-20
Accelerator plan client spend	$100K/month
Accelerator plan emails	5 million
Client launch time	1-3 weeks

⚡ ACTIONABLE TAKEAWAYS

If You Want to Replicate This:

Get Claude Code ($200/mo) — this is the builder
Identify your #1 bottleneck — for them it was processing speed + row limits
Start with one tool — e.g., a lead enrichment script or a custom scraper
Use GitHub for version control (even if you don't know what it is — Claude Code will handle it)
Deploy on Railway — cheap, handles scaling, hosts databases
Build dashboards on Vercel — auto-deploys from GitHub
Stack your data sources — don't rely on one vendor. Build a waterfall: primary vendor → AI search → personal email fallback
Process leads in bulk with parallel workers — Railway lets you run 50+ simultaneously
Build analytics from day one — track email-to-lead ratios, vendor performance by ICP, campaign performance

Minimum Budget to Start:

Claude Code: $200/mo
Railway: starts small, scales with usage
Data vendors: varies
Total MVP: probably $300-500/month

Key Transferable Principles:

Build for yourself, not for everyone — internal tools can be ugly and fragile, they just need to work for you
Speed of iteration > perfection — being able to reload lists instantly is worth more than a polished UI
Data compounds — every lead, every campaign result, every vendor comparison makes the next decision better
Non-coders can build production systems — the barrier is knowing what you want, not knowing how to code