Full Setup • Daily Playbook • Infrastructure Map
300,000 cold emails per month. 153 calls booked. $1,200 total cost. The entire team is one AI agent running on a $7 VPS.
The complete system — daily operation, infrastructure, the math behind the 2-email philosophy, and every lesson from 4.7 million emails of data.
Data from 10+ B2B clients over 8 months. Based on Liam Sheridan's OpenClaw system, with post-install hardening from Moritz Kremb and multi-agent architecture from Sandra Leow.
The Core Insight
Only 3-5% of your Total Addressable Market is buying at any given moment. Most outbound teams email the same 20% of their list over and over — and miss the other 80% entirely.
The fix isn't better subject lines. It's reaching your entire TAM every 45 days so you mathematically can't miss a buying window.
“1,500-3,000 emails/day is the baseline for serious pipeline generation. Most people drastically underestimate.” Send to MORE people with fewer emails each — not fewer people with longer sequences.
The 2-Email Philosophy
Every cold email guru says “follow up 7 times.” 4.7 million emails of data says they're wrong.
| Email # | Share of Positive Replies | Verdict |
|---|---|---|
| Email 1 | 67% | Send |
| Email 2 | 22% | Send |
| Emails 3-7 | 11% | Kill |
Data: 4.7 million+ emails across 10+ B2B clients over 8 months.
Send 2 emails. Stop. Come back in 45 days with fresh copy and fresh timing. Result: 95%+ deliverability sustained over 8 months. Reply rates hold at 1.5-2.5% depending on market.
Daily Playbook
OpenClaw runs autonomously on a 24-hour loop. No human intervention required unless a hot lead comes in.
Apollo + Sales Navigator enriched via Clay, validated via ZeroBounce, segmented and loaded into Instantly. Takes 20 minutes. Used to take 2-3 hours by hand.
Reviews previous day's performance. Which subject lines got opens? Which variants got replies? Deliverability warnings? Auto-pauses underperformers, scales winners.
Emails go out via Instantly across 25 sending accounts at ~20 emails/account/day. 2 emails per prospect, then stop. No exceptions.
Hot lead = CRM task + WhatsApp alert. Soft positive = nurture. Objection = logged. Negative = instant unsubscribe.
Summary on WhatsApp: emails sent, opens, replies, hot leads flagged, issues detected. You wake up knowing exactly what happened.
Infrastructure Map
Everything you need to run 300K emails/month. No team. No office. One VPS.
Lessons From 4.7M Emails
Every failure came from adding complexity that sounded smart but didn't move the numbers.
Emails 3-7 destroyed deliverability while generating almost no replies. You're paying to make other emails less effective. Killed it. Went to 2.
Main domain got flagged. Regular business emails started going to spam. Took 3 months to recover. Now: dedicated outbound domains you're prepared to burn.
Scraped LinkedIn posts, news articles, job changes. Reply rate went up 0.5%. System became 10x more complex and broke constantly. Now: personalize first line on 3-4 data points max.
Uncanny AI smell. Too perfect, too generic, too obviously not human. Now: humans write the frameworks and angles. AI fills in personalization and handles mechanical work.
Coverage > cleverness. Volume > virtuosity. 2 emails > 7. The winning system is simpler than what most people expect.
Post-Install Checklist
Installation is the easy part. This is the 30-60 minute hardening pass that turns a fresh install into something production-grade.
Configure USER.md, IDENTITY.md, SOUL.md, MEMORY.md, and daily memory files. Without these, output quality degrades between sessions and context doesn't persist.
Set primary model + fallbacks. Reliability first, cost second. Claude for reasoning. Kimi K2.5 (5-8x cheaper) for structured retrieval and monitoring. Gemini Flash for simple ops.
Move secrets outside workspace. Strict filesystem permissions. On VPS: restrict inbound access, keep gateway token auth enabled. Never leave API keys in the workspace folder.
Set dmPolicy allowlist and allowFrom rules. Configure group/topic allowlists explicitly. Use managed browser profile by default — relay only for logged-in states.
Add heartbeat instructions for memory maintenance and cron-health checks. Cron jobs silently fail. Force-run missed jobs and report exceptions daily.
Dedicated accounts for agent environment. Install summarize skill early. Convert recurring successful workflows into custom skills for compounding efficiency.
Bonus: Multi-Agent Architecture
Once the outbound loop works, the next level is running multiple scoped agents with a shared coordination layer.
Every agent reads from a shared kanban-style taskboard. Agents check the backlog autonomously — if there's a task in their domain, they pick it up. No manual assignment needed.
Prompt Template
You are [role]. Your domain is [specific scope]. You have access to [tools]. Your quality bar is [standard]. You do not work outside your domain — if a task falls outside your scope, flag it. Check the dashboard every heartbeat for tasks in your domain. Be proactive, not reactive.
Running Claude on every agent is how you burn through credits in a week. Match the model to the task:
One chat for everything = one giant context window where conversations bleed together. Use Telegram Topics: 8 topics = 8 isolated contexts. Each topic has its own memory file. Route messages by topic ID, write to topic-specific memory, and let the dashboard extract tasks automatically.
Constrained-domain agents are easier to debug and more reliable than unconstrained general agents. Give each agent a narrow scope and a clear quality bar.
Bonus: Multi-Channel Expansion
Cold email is the engine. But the highest-converting outbound channel right now might be one most people ignore.
Decision-makers get 50+ LinkedIn messages per day but only 2-3 Twitter DMs. The arbitrage is real. Personalize from their last 5-7 tweets — it takes 30 seconds per DM and the response rates are dramatically higher than any other channel.
Proven Results
All metrics from Liam Sheridan's public data across 10+ B2B clients. Twitter DM data from Levi Munneke. Multi-agent architecture from Sandra Leow.
We help B2B companies deploy autonomous outbound systems — from infrastructure setup to daily operation. If you're running manual outbound and want the same output at a fraction of the cost, let's talk.
Book a CallOr connect on LinkedIn