-
1,000 Cold calls a day7 Tools she can use mid conversation0 Card digits ever spoken out loud
-
The problem
Outbound calling is brutal work. A human rep burns out, has bad days, gets rattled by the fourth rejection in a row, and quietly stops dialing. Volume collapses, and so does the pipeline. Hiring more people just means more people to burn out. And the reps who do push through spend most of their day talking to voicemail instead of talking to buyers.
-
The solution
Sophie dials. All day, every day, at a volume no team could sustain. She pulls the customer's real history before she speaks, so she knows what they bought and when. She handles price objections without cascading down on price. She takes the payment on the call, either on the card already on file or on a new one, and processes the upsell. Then she logs the outcome and moves to the next number. She does not get tired and she does not get rattled.
-
How it was built
VAPI for the call, Claude Sonnet as the reasoning layer, ElevenLabs for the voice, Deepgram for transcription. The interesting part is the tools. She has seven, each one an n8n workflow exposed to her as an MCP tool: look up a customer, query past transactions, charge the card on file, process an upsell, take a new card, update customer details, replace a saved card. They hit the client's e-commerce and order platform directly. That is the difference between an agent that talks and an agent that does. One detail I insisted on: card numbers are entered on the keypad, never spoken, so no card digit ever touches a transcript. Every call comes back through a post call pipeline that analyses the outcome and writes it to Supabase and to the client's sheet.
Watch the wires draw. This is the real architecture.
VAPIClaude SonnetElevenLabsn8n MCP toolsSupabase
WANT ONE OF THESE FOR YOUR BUSINESS?
Schedule a Call
Contact