Skip to content
Back to Projects
AutomationAI / MLFull-StackLiveClient project · anonymised

Cold-email outbound system

Production sequencer for a consultancy: six-touch campaigns, reply detection and a partner approval loop.

Diagram of the cold-email pipeline: prospect research, Claude Code drafts, guardrail checks, partner approval, sending and inbox polling

The Challenge

A consultancy's outbound programme needed personal, research-based cold email sent from the founder's real address – with nothing going out unapproved and deliverability put first.

The Approach

The system imports prospect research and renders a six-touch sequence per prospect. Custom drafts are written in Claude Code from a research brief, then pass hard guardrail checks – claims, length, opt-out line, CAN-SPAM signature – before anyone can approve them. The client's two partners review in an Excel workbook that round-trips into the system. Approved emails go out over SMTP2GO with correct threading of follow-ups, and an IMAP poller detects replies and bounces and pauses those prospects automatically. A Flask dashboard shows the campaign and edits settings.

Key Features

  • Six-touch sequences with reply threading
  • IMAP reply and bounce detection with automatic pause
  • Claude Code drafts from research briefs, with hard guardrails on claims, length, opt-out and CAN-SPAM signature
  • Partner approval through an Excel round-trip
  • Flask dashboard for campaign status and settings
  • Plain-text email, no tracking pixels

Results & Impact

  • Running in production for the client's outbound programme
  • Nothing is sent without partner approval
  • Follow-ups thread correctly and replies or bounces pause a prospect automatically

Working on something where people and data meet?

AI automation, data science, psychometric validation or sports psychology – tell me what you're building.