Skip to content
All work
Webinars · Voice AI2025

AI-Powered Webinar Platform with Voice Sales Agents

Live and evergreen webinar platform where AI voice agents qualify attendees in real time, book demos and hand warm leads to humans.

LIVE · 412 LISTENING
Webinars · Voice AI
AI-Powered Webinar Platform with Voice Sales Agents
+31%
Qualified lead lift vs. human SDRs
1,200+
Webinars run
680ms
p95 turn latency
2
SDR headcount reallocated
The problem

The team ran 80 webinars per month but only 9% of attendees got a meaningful follow-up. SDRs were drowning.

Vapi prototypes sounded robotic on tricky objections and leaked buyer signals that the CRM never saw.

They needed a measurable lift in booked demos inside 60 days — without adding headcount.

The approach
  1. Rebuilt the webinar runtime on Stream with low-latency branching: the agent joins the call as a first-class participant.

  2. Designed a two-agent voice stack on Vapi: an empathetic qualifier and a booking specialist, with a supervised handover graph.

  3. Wrote an evals harness with 412 real-call transcripts; every deploy ships a regression report to Slack.

  4. Connected intents to the CRM so every buyer signal — price, timeline, tech stack — becomes a structured field, not a note.

Architecture
Rendering architecture diagram…
Outcome
  • 31% lift in qualified leads vs. human SDRs in an A/B over 1,200+ webinars.

  • Meeting-booked rate on attendees went from 9% to 21%.

  • SDR team reallocated to closers, saving the equivalent of two headcount.

  • p95 turn latency under 680ms on live calls.

What I owned end-to-end
  • Voice agent architecture and prompt library
  • Eval harness and release gates
  • CRM integration and signal taxonomy
  • Privacy review with the client DPO
Next up

Agentic Automation SaaS with Crypto Billing