Hey,
Two weeks ago we finished building something we have been planning for a long time.
A full GTM outbound system that lives entirely inside Claude. From building the client playbook to sourcing leads, writing copy, launching campaigns, and having the whole thing optimise itself automatically based on what is actually working.
Here is how it works (full video breakdown here)
THE FOUNDATION: Playbook and Leads
Every client engagement starts with a living playbook. Claude builds it using call recordings, website research, and all the materials clients share with us. It lives in GitHub and syncs automatically every week, updated with new campaign results, sales call insights, and client feedback.
This playbook drives everything that comes after it.
When it is time to source leads, Claude follows a specific priority order we define: CSV first, then our Supabase database, then BlitzAPI, then Exa. The targeting is always pulled directly from the playbook. The ICP is never an assumption.
THE EXECUTION: Enrichment, Copy and Launch
Once we have the leads, Claude runs waterfall email enrichment through MagicFind, DeBounce, and BounceBuster. Valid emails only before anything moves forward.
For the copy, we use a split setup. DeepSeek does the research. Gemini 2.5 Flash writes the first draft. Claude then applies the playbook rules, reviews the output, and generates examples for 10 leads, email 1 and email 2 for each.
This is where the human in the loop matters. You review the examples, chat with Claude like you would with a GTM engineer, and push back on anything that needs adjusting.
Once the copy is approved, Claude builds the campaigns directly in Smartlead or Instantly.ai. ESP-specific, correct senders, right tags, custom variables. Before full launch, a human checks 10 leads to confirm everything looks right.
Then you start the campaign.
The Loop (Auto Optimisation)
This is the part that changes everything.
Once the campaign is live, Claude reads the replies, positive and negative, the sales calls from the client team, and the feedback we receive from clients directly. Based on all three, it keeps the playbook continuously updated and the campaigns continuously optimised.
No manual review cycles. No quarterly playbook rewrites. The system learns as it runs.
What Clay couldn’t do
Clay is great for enrichment and pushing leads to the sequencer. We still use it for 90% of our client work and it does that part well.
But the playbook creation, the lead sourcing, the campaign building, the self-learning loop, that never lived in Clay. Now it all lives in one place.
We are still early…
We built this in under two weeks. There is a lot more to build. But the core system is already running for clients and the results are coming in.
If you want to understand how we are implementing this and what it could look like for your team, book a call with us.
