Alien: Engineering a 90-Inbox Cold Email Infrastructure for Sub-$400/Month

Cold email at scale is a deliverability engineering problem before it is a sales problem. The infrastructure question that determines whether cold email works is not which tool do I use but how do I get 2,700 to 3,600 emails per day into primary inboxes at a cost that a single closed deal justifies.

The Alien outreach system was built to drive customer acquisition for Avo's custom development practice. The target was funded startups and agencies willing to pay $15,000 to $25,000 for a 2 to 3 week custom build. At that deal size, the entire infrastructure cost was recovered in under 30 minutes of a single closed project.

The decision to build rather than buy was explicit. Apollo, Clearbit, Hunter, NeverBounce, ZeroBounce, and Sales Navigator were all excluded from consideration. Every paid SaaS tool in the outreach stack charges per-seat or per-send. At 3,000 sends per day, per-send tools become expensive faster than sales covers them. The only exception was Smartlead at $39/month flat, used exclusively for inbox warmup, because building a warmup network requires thousands of cooperative real inboxes, a network effect that cannot be replicated in-house.

Inbox infrastructure used Maildoso 90 Google Workspace inboxes at approximately $330 per month. Google Workspace was chosen over SMTP inboxes after testing showed GWS inboxes had 92 to 95% inbox placement versus SMTP at 85 to 90%. At 3,000 sends per day, a 10% inbox placement gap was 300 emails per day landing in spam.

Each domain carried a maximum of 2 to 3 inboxes. Spreading 90 inboxes across 30 to 45 domains kept per-domain send volume low enough to avoid automated reputation flags. SPF, DKIM, and DMARC records were configured on each domain. Sending from avogrowth.com was prohibited from day one. The risk was asymmetric: a deliverability incident on a cold outreach domain costs a $10 domain replacement. A deliverability incident on avogrowth.com costs brand trust across every email sent from that domain.

The 30 to 50 cold sends per inbox per day limit was inviolable. Research showed spam rate jumping from 0.7% to 3.9% above 100 sends per inbox per day. At 90 inboxes times 40 sends per inbox, system capacity was 3,600 sends per day.

The alien/leads-pg PostgreSQL schema implemented an explicit state machine with nine states: raw, enriched, verified, scored, ready, in-sequence, replied, booked, closed. State transitions were atomic PostgreSQL updates with a check constraint preventing backward transitions. The state machine made pipeline health observable: at any moment, a single SQL query showed the count of leads in each state, the transition velocity, and the bottlenecks.

Each outbound email was personalized using the company website content fetched at enrichment time and the ICP score factors. The alien/personalization Rust binary called the Claude API to generate a 1 to 2 sentence opening hook referencing something specific to the prospect's website, then appended a templated pitch body under 80 words. Plain text format was mandatory. No HTML. No images. One link maximum. HTML emails from cold outreach domains trigger spam filters at 3x the rate of plain text emails.

Reply detection used IMAP polling on each of the 90 GWS inboxes, running every 5 minutes. Each reply triggered a Claude API classification call to label the reply type (positive, neutral, objection, bounce, or stop), a state machine transition, an AI-drafted response generation if positive or an objection, and routing to the /admin/inbox tab for operator review.

90 inboxes at 92 to 95% inbox placement. Theoretical capacity: 3,600 cold sends per day. Operating target: 2,700 sends per day with 25% headroom. Total infrastructure cost: approximately $369 per month. Break-even: 1 closed $15,000 project per 40 months of infrastructure cost. Warmup timeline to full volume: 22 days for initial ramp, 30 days to reach sustained 30 to 40 cold sends per inbox per day. Zero domain blacklistings during the warmup phase.