/
CASE STUDIES
/
Market Research
/
Convo
Talking to Your Customers at Scale: An AI Voice Agent for Live Market Research
/
CLIENT
Convo
Visit site
/
TIMELINE
3 months
From static surveys to real conversations
A market research company wanted to move beyond static surveys and run real-time, conversational interviews at scale. After an initial PoC on VAPI, we rebuilt the voice stack on LiveKit and AWS, creating an AI voice agent that can talk naturally to participants, capture transcripts, and produce study-wide insights.
Scalable, natural conversations
Early PoC on VAPI too limited and expensive to scale
Need for natural, low-latency conversations that feel like real interviews
Desire for richer study-wide analysis, not just per-call transcripts
How can we 10x our lead generation and make the process more seamless?
/
SOLUTIONS
Natural voice conversations & emotional intelligence
Migrated from VAPI PoC to a custom LiveKit-based architecture on AWS
Built a pipeline combining text-to-speech, speech-to-text, LLM, and turn detection models
Added real-time and post-interview transcript analysis to surface patterns and themes
Designed the system to be configurable per study (scripts, personas, questions, tone)
/
RESULTS
Talking to customers at scale
Enabled scalable, natural voice interviews at a fraction of the original PoC cost
Gave researchers a way to “talk to customers at scale” without hiring more moderators
Delivered study-wide insights and summaries directly from the interview transcripts
600ms
Voice-to-voice Performance
100+
Concurrent calls per server instance
VAPI to LiveKit
from POC to production-ready LiveKit instance in less than 3 weeks
Excellent
Rizki Anugrah
Product Manager of Blockhaus
