ElevenLabs Resolves 22,000 Monthly Support Tickets via ElevenAgents

ElevenLabs uses ElevenAgents to handle over 22,000 customer support tickets monthly without human intervention for the majority of cases. The AI agent processes text, screenshots, and PDFs, integrating directly with the ticketing platform to look up account data and execute actions.

Performance Metrics

Last month, the agent resolved 65% of tickets end-to-end. The average initial response time dropped to 60 seconds, down from hours previously. Across all interactions, 70% of conversations received positive user ratings.

Implementation Strategy

The build focused on translating existing standard operating procedures (SOPs) into executable workflows using two key features:

• Procedures: Support playbooks for refunds, billing, and voice clone questions were ported into repeatable agent workflows. Structured Procedures ensure deterministic steps for critical actions like refunds.
• Webhook tools: These allow the agent to dynamically retrieve user information, such as subscription tier and credit usage, enabling end-to-end resolution.

Quality Assurance and Iteration

To maintain high quality, the team employed several mechanisms:

• Channel behavior settings: Differentiated responses for email (concise) versus voice (interactive).
• Simulation tests: The agent was tested against roughly 50 frequent scenarios. Tests are re-run after configuration changes to prevent regressions.
• Shadow mode: The agent was evaluated on production traffic within the ticketing system before full deployment.

Continuous Improvement with Architect

The support team uses ElevenAgents Architect, a natural language agent builder, to diagnose issues and improve performance. Architect analyzes escalated tickets to identify root causes, such as knowledge base gaps or missing tools, and proposes changes. Before shipping updates, Architect drafts new simulation tests to ensure new procedures succeed and existing functionality remains intact.