Building Emma - JustCall’s conversational AI support agent

Emma is an in-product AI support agent that started as a growth experiment to improve AI Voice Agent adoption. Most users relied on docs or demos before experiencing the value. Emma started as an overnight experiment to test whether conversation could drive adoption better than onboarding flows.

Emma increased AI Voice Agent creation by 15.6% and quickly evolved beyond activation. Users relied on Emma for onboarding, troubleshooting, and self-serve demos. Some even cancelled scheduled sales calls because Emma felt faster and easier. Emma conversations also became a source of product insight, helping product managers identify friction, prioritize fixes, and roadmap opportunities through conversation transcripts. Some users even asked to white-label Emma for their own businesses.

AI Voice Agents were powerful and solved real business problems, but users struggled to understand their capabilities or how to set them up. Most users relied on docs or demos to find AI Voice Agents, creating friction before users experienced value.

Instead of another onboarding flow, we explored a conversational entry point for discovery, support, and guidance. The first version of Emma was designed overnight as a lightweight MVP to validate conversational onboarding and support. The MVP unlocked stakeholder buy-in and accelerated investment.

The prototype quickly expanded beyond activation. Users began relying on Emma for onboarding, troubleshooting, and self-serve demos. Every conversation was piped into Slack, turning support interactions into continuous product feedback. Users even cancelled scheduled demos as they found Emma more approachable for self-serve product demos. Over time, users even requested Emma as a white labeled solution to their own businesses.

1. Optimized for speed by building a lightweight 1-night MVP instead of spending weeks refining onboarding concepts upfront.
2. Made Emma highly visible inside the product instead of hiding it behind support flows or documentation surfaces.
3. Treated conversations as a continuous stream of product insight and feedback, not just another support channel
4. Balanced open-ended interactions with guardrails early on to maintain reliability and user trust