LinkedIn lead generation automation with conversational DMs
An outreach pipeline that finds prospects, connects, opens conversations and replies to what people actually write.
The problem
Manual LinkedIn outreach does not scale. The automation tools we tend to see fail in one of two ways: accounts get flagged by LinkedIn's detection, or the tool cannot hold a real conversation once the first message is sent.
Challenges
Staying clear of account flags
Many automation tools get accounts flagged by LinkedIn's detection. The build had to pace activity with rate limits, realistic delays and conservative thresholds, and it puts account safety ahead of raw sending volume, accepting that no approach removes the risk entirely.
Holding a real conversation
The second common failure is a tool that sends a first message and then cannot cope when someone replies. The system had to read each reply and write a response based on what the prospect actually said, instead of running a fixed script.
Not over-messaging anyone
Follow-ups help when a lead goes quiet and annoy when they repeat too often. The build needed scheduled follow-ups with stop conditions, so each person is contacted a sensible number of times and then left alone.
How the system works
Lead scraping
An automated process finds LinkedIn profiles that fit set targeting parameters such as industry, job title and keywords, and stores the qualified ones in a database.
Targeted connection requests
Screened leads receive personalised connection invitations, sent deliberately rather than as indiscriminate mass outreach.
Acceptance-triggered outreach
As soon as someone accepts, an automated opening message goes out.
Conversation continuation
The system watches for replies and writes an AI response based on what the prospect actually said.
Automated follow-up
Leads that go quiet get scheduled follow-ups, with safeguards that limit how often anyone is contacted.
Design decisions
Targeted rather than mass connection requests
Leads are screened against set targeting criteria such as industry, job title and keywords before anyone is invited to connect. We chose deliberate, personalised requests over indiscriminate volume because it protects the account and gives better conversations.
Outreach triggered by acceptance
The opening message goes out only once a connection is accepted. That ties the first contact to a real signal of interest and keeps the sequence feeling like a natural exchange rather than a cold blast.
A database as the record of every lead
Qualified leads and their conversation state are stored in Supabase, and n8n orchestrates the steps. Keeping state in one database makes the process observable and lets the system know who has been contacted and when.
Result in production
A LinkedIn outreach pipeline that runs continuously without manual sending, while still reading and responding to real replies instead of using generic scripts.
Outreach runs continuously without anyone sending messages by hand.
Prospects who reply get a response that reflects what they wrote, instead of a fixed script.
Follow-up is scheduled and limited by stop conditions, so the team does not have to track who has been contacted.
What we would do differently
We would separate the enrichment pipeline from the sending pipeline more strictly. Coupling them made it too easy for a bad enrichment run to poison outreach quality for a day.
Who this pattern fits
We built it for sales and marketing teams, but the same architecture applies to any department with similar volume, compliance and coordination needs. It can be self-hosted and run by a small team without dedicated platform engineers.
Tools used
n8n
Workflow orchestration and scheduling, with version control and observability.
Unipile
The integration layer used for LinkedIn in production.
Supabase
PostgreSQL database, authentication and persistent record storage.
Related service, use case and industries
Frequently asked questions
Does it only send a fixed message sequence?
No. After the opening message, the system watches for replies and writes an AI response based on what the prospect actually said. Leads who go quiet get scheduled follow-ups, and stop conditions limit how often anyone is contacted.
How are leads chosen?
An automated process finds LinkedIn profiles that fit set targeting parameters, such as industry, job title and keywords, and stores the qualified ones in a database. Only screened leads receive a personalised connection request.
Will this get my LinkedIn account flagged?
The system uses rate limiting, realistic timing gaps and conservative safety thresholds. No tool removes the risk entirely, but this approach puts account safety ahead of the usual alternatives.