ElevenLabs Uses Its Own AI Agents to Run 230 Customer Interviews in 24 Hours.

ElevenLabs has started using its own voice AI technology to solve a classic product research problem: how to gather deep customer insights at scale. The company announced on LinkedIn that it deployed ElevenLabs Agents to conduct more than 230 customer interviews for its ElevenReader app in under 24 hours, achieving an 85% success rate.

The move highlights a growing trend among AI companies: turning their own tools inward to replace manual workflows. ElevenLabs framed the experiment as bridging the gap between traditional research methods. Live customer interviews deliver nuanced insights but don't scale beyond a handful of conversations per week. Surveys scale easily but lose the conversational depth that reveals why users behave the way they do.

ElevenLabs Agents are conversational AI systems designed to handle voice interactions in real time. The company used these agents to conduct 10-minute phone calls with ElevenReader users, asking open-ended questions and following up based on responses. According to the post, 85% of calls stayed on-topic and were classified as successful. The agents maintained an average call duration of 10 minutes, long enough to surface meaningful feedback without exhausting participants.

The company claims it shipped insights to production the next day, a turnaround time that would be difficult with traditional interview synthesis. The post points to a detailed build explanation in the comments, though the exact technical implementation was not disclosed in the announcement.

This internal use case follows ElevenLabs' recent enterprise deployments, including partnerships with Deutsche Telekom and integrations with customer support platforms at scale.

The experiment demonstrates a practical application of conversational AI beyond customer service. Product teams at smaller companies often face a choice: invest weeks in manual interviews or settle for survey data that misses context. If voice agents can reliably extract qualitative insights at survey-level scale, it changes the economics of user research.

ElevenLabs is not alone in exploring this territory. Tools like Voiceflow and Intercom have introduced conversational interfaces for support and engagement, but ElevenLabs' focus on voice fidelity and real-time interaction positions it differently. The company raised a $500 million Series D in February at an $11 billion valuation, backed by demand for its voice cloning and agent capabilities.

The 85% success rate is notable but leaves questions. ElevenLabs did not specify how it defined "on-topic" or what happened in the 15% of calls that failed. Call quality, participant dropout, and agent hallucination remain common challenges in conversational AI deployments. The company also did not disclose whether participants knew they were speaking with an AI, a detail that affects both ethical considerations and response quality.

If the approach proves reliable, it could accelerate product iteration cycles. Teams could run hundreds of interviews after every feature release, identify friction points in near real-time, and adjust roadmaps without waiting for quarterly research sprints. The cost per conversation would be lower than hiring a research firm, and the data would be structured enough to feed directly into product analytics tools.

For now, ElevenLabs has shared the high-level results but has not released the agent template or detailed methodology. The company promised a build walkthrough in the comments, which could provide a blueprint for teams looking to replicate the workflow using ElevenLabs Agents or similar platforms.

The move also signals a shift in how AI companies validate their own products. Rather than relying solely on usage metrics or support tickets, ElevenLabs is using conversational agents to directly ask users what works and what doesn't. That feedback loop, if it scales, could become a competitive advantage in fast-moving markets where user expectations change faster than traditional research can track.