Teardown No. 17 · Outbound

How Ilya Azovtsev had Claude run a week of outbound to HubSpot integrators

Ilya Azovtsev let Claude run his outbound for a week. Claude pulled pain points from 3 discovery call recordings and wrote the sequence in the prospects' own words. It then ran the campaign in Jason AI (Reply) through MCP and flagged the positive replies. He reports 60 contacts, a 40% positive reply rate, 24 interested and 17 booked calls so far. He puts his own time at about 2 hours for the week.

Ilya AzovtsevVerified Oct 16, 2026Markdown copy
Part of playbook: Run cold outbound with agents without burning your domain

The problem

His target customers were HubSpot integrators, which he describes as “agencies implementing HubSpot for clients”. He wanted to test one question: “Can Claude run an entire outbound campaign - from research to replies - with minimal human input?” He says the goal was to have Claude “Actually operate the system” rather than just help write an email.

The blueprint (as he describes it)

  1. Mine real calls for language. Claude analyzed 3 discovery calls with HubSpot partners. It extracted pain points such as “net new is hard,” “inbound hit a ceiling,” and “SDR agency didn’t work”.
  2. Write the sequence from those calls. Claude wrote the full sequence in the prospects’ exact language. In his words: “No templates. No “personalization tokens.” Real pain points from real conversations.”
  3. Build the list. The HubSpot Partner Directory was scraped and the contacts were enriched in Clay. The post doesn’t say whether he or Claude did this step.
  4. Run the campaign through MCP. Claude controlled the Jason AI (Reply) campaign. It added contacts, launched the sequence and monitored deliverability.
  5. Triage replies. Claude monitored replies and flagged the hot leads. He says: “I just showed up to book the calls.”

Stated results

  • 60 contacts.
  • 40% positive reply rate.
  • 24 interested.
  • 17 booked calls, which he marks as “(so far)”.
  • Total human time: about 2 hours for the week.
  • No open rates, deliverability figures or closed deals are stated.

Caveats

  • This is one self-reported LinkedIn post. The source includes no screenshots, raw data or post date.
  • The sample is small: 60 contacts in one niche, over one week.
  • The booked-call count is marked “(so far)”, so it may have changed after posting.
  • The post ends with a comment-for-DM offer: “Comment “CLAUDE” and I’ll send you the exact MCP setup + prompts I used.” The setup and prompts are not in the post itself. The source mentions no paid product, course or tier.
  • The post calls the sending tool both “Jason AI (Reply)” and “Replyio”.
  • The post doesn’t say who did the directory scrape or the Clay enrichment.
  • X: Unverified not checked.

Sources

Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).

Get one teardown a week.

Same format, a new operator every issue. Double opt-in.