Automated LinkedIn Outreach with AI Reply Agent

The challenge

Manual LinkedIn outreach forces a person to repeat the same sequence every day — finding prospects, sending connection requests, watching for acceptances, and writing follow-up replies — which does not scale and is easy to get throttled for when done too aggressively.

Replies also need to remember the full history of each lead, so a person juggling many conversations loses context and momentum across a long outreach campaign.

The solution

An n8n-orchestrated outreach system handles the full cycle: prospect scraping through Unipile, a deliberately gradual connection ramp, post engagement, acceptance tracking, and campaign messaging.

A Claude-powered reply agent reads the stored conversation history for each lead and drafts context-aware responses, while a PostgreSQL database keeps every prospect, message, and state durable across runs.

Implementation

• Designed a 10-table PostgreSQL schema so the system can track prospects, connections, messages, and campaign state at multi-user scale.
• Built a gradual weekly ramp that grows from 3-5 to 15-20 connection requests per day to stay within safe usage patterns.
• Added randomized delays, sponsored-post filtering, and audit logs so the automation behaves naturally and every action is traceable.

Results

• Prospecting, connecting, engagement, and replies now run as one continuous automated system instead of daily manual work.
• Every lead's history is retained, so replies stay context-aware no matter how many conversations are open.
• The system runs live on a Hostinger VPS with built-in throttling and audit trails.

Impact

• 10-table PostgreSQL schema designed for multi-user scale
• Gradual weekly ramp from 3-5 to 15-20 connection requests per day
• Randomized delays, sponsored-post filtering, and audit logs