Last month I ran an experiment: delete the cold list, connect Claude to my buying-signal feed, and see what happens.
The numbers after 30 days:
- 187 messages sent (6 to 8 a day)
- 31% reply rate
- 35 booked meetings
- $99/month total tool cost
- 0 LinkedIn warnings
Here's the entire system, including the prompt and the failures.
The Problem With The Old Way
Cold outbound math is broken. Reply rates sit at 1 to 3%, which means 10 meetings requires 1,500+ messages. That volume needs multiple accounts and the exact robotic behavior LinkedIn's detection now catches instantly.
Meanwhile, real buyers were already raising their hands at me every single week:
- People viewing my profile
- People commenting on posts in my space
- Past champions changing jobs
- Target accounts raising funding, hiring, visiting my pricing page
I was ignoring all of it to go message strangers. That's the actual insanity of cold outbound: interrupting people who never looked at you while ignoring the people who did.
The Stack
Two tools. One connection.
Updately watches 30+ buying signals in real time: job changes, competitor engagement, funding announcements, hiring activity, profile visits, post engagement, website visitors. It also sends the outreach with account-protection limits built in.
Claude connects to Updately through MCP (Updately ships an MCP server, docs at updately.ai/docs/mcp). That means Claude can read my live signal feed directly. No CSV exports, no Zapier spaghetti.
Setup took 20 minutes. The only hard part was being honest about my ICP.
The Morning Prompt
Every day, one prompt:
"Check my Updately signal feed for the last 24 hours. Filter to my ICP. Rank by signal strength: job changes and pricing page visits first, then profile views, then comments, then repeated likes. For the top 15, research each person and tell me in one line why NOW is the moment. Draft an opener in my voice: casual, references their signal, no pitch, under 60 words, ends with a question. Flag anything stale or weak-fit. Don't force it."
That last line matters. An AI forced to produce 15 drafts pads the list with garbage. An AI allowed to return 9 stays honest.
Ten minutes of review. Approve. Updately sends, spread across the day, well under LinkedIn's limits.
Week By Week
Week 1: 2 meetings. Drafts were too formal. Fix: I fed Claude 20 of my real past messages and told it to learn my voice. Night and day.
Week 2: 7 meetings. Job-change signals started printing. A past champion moved companies, got a congrats message within 24 hours, booked the same day.
Week 3: 11 meetings. Added competitor-engagement signals. People commenting on competitor posts are literally in-market and nobody messages them, because no human can watch competitor feeds all day. Claude can.
Week 4: 15 meetings. The flywheel: my posts generate engagement, engagement becomes signals, signals become meetings, meetings become posts.
What Failed
Honesty section, because every "AI booked me a pile of meetings" post skips it:
-
Autopilot. I stopped reviewing drafts for 2 days. Quality dipped and one draft referenced the wrong company. The 10-minute review is the system.
-
Weak signals. Messaging every single post-liker converts at ~0 and reads desperate. Threshold: 2+ engagements in 30 days, or one strong signal.
-
Pitching in message one. The signal earns you relevance. Spend it on them, not on your product. Pitch after they reply.
Why This Doesn't Get You Banned
LinkedIn's 2026 detection flags volume, robotic timing, and fake browser sessions. This system sends 8 personalized messages a day to people who already engaged with me, through safeguarded sending. That's not automation behavior. That's just an active professional.
The safest outreach isn't the one with the cleverest evasion. It's the one that never needed volume in the first place. The full safety picture is in is LinkedIn automation safe in 2026.
The Takeaway
Everyone building "AI SDRs" is optimizing the wrong variable. More messages is how outbound died. The broader industry reached the same conclusion, covered in do AI SDRs actually work.
The unlock is AI that answers a different question: out of everyone in my market, who showed intent in the last 24 hours, and what exactly did they do?
Answer that every morning and 8 messages beat 800.
Frequently Asked Questions
Setup
Do I need to be technical to connect Claude to Updately? No. The MCP connection is a configuration step documented at updately.ai/docs/mcp, and took about 20 minutes end to end. Defining your ICP well takes longer than the technical setup.
Which Claude plan works for this? Any plan that supports MCP connectors. The workflow uses one conversation a day, so usage is light.
Can this run without Claude? Yes. Updately's own AI agents score, prioritize, and draft campaigns natively, and that autopilot mode is how most customers run it. The Claude layer adds a custom reasoning step and your own prompt logic on top for teams that want it.
Results
Are 35 meetings in 30 days typical? Treat it as an outcome of the inputs, not a promise. The drivers were an active founder profile generating signals, a clear ICP, job-change tracking on past champions, and daily consistency. Signal-based campaigns typically see 15 to 45 percent reply rates against 1 to 3 percent cold; your meeting count scales with your signal volume. Model your own numbers with the ROI calculator.
What does the whole stack cost? Updately starts at $79 per month and Claude at $20 per month, about $99 total. The stack it replaced, sequencer plus data provider plus enrichment, commonly runs $500-plus per month before payroll.
Daily operation
How much time does it actually take? About 30 minutes a day: read Claude's ranked queue and one-line timing rationales, edit two or three drafts, approve, done. Posting content twice a week adds another hour and feeds the signal flywheel.
What happens when I skip a day? Signals decay. Profile views and comments cool within about 48 hours, so a skipped day means a colder queue tomorrow. The system tolerates a missed day; it punishes a missed week.
Setup docs: updately.ai/docs/mcp. Full pricing from $79/month. See pricing or book a demo.