Key points
- ElevenLabs says its own ElevenAgents support agent resolves over 22,000 tickets a month, and 65% of tickets last month were closed end to end.
- Average first response time is now 60 seconds, down from hours; 70% of conversations the agent handled were rated positively by users.
- The build relies on Procedures, webhook tools, channel behavior settings, simulation tests (about 50 common scenarios) and shadow mode before launch.
- Human support staff stay in the loop for escalations, policy updates and guarding against overfitting to one ticket type.
Why it matters for marketers: It offers a concrete blueprint for support automation: convert existing SOPs into workflows, test against common scenarios, run in shadow mode first, and keep humans handling escalations and policy.
ElevenLabs published an account of how it runs its own customer support on ElevenAgents, its agent platform. The company says the agent resolves more than 22,000 tickets a month without a person involved. Last month it closed 65% of tickets end to end, cut average first response time to 60 seconds from hours, and earned positive ratings on 70% of the conversations it handled. These figures are self-reported.
The agent handles text, screenshots and PDFs, connects to the company's ticketing platform, looks up account details, takes actions and escalates to staff when needed. According to ElevenLabs, the starting point was the support team's existing standard operating procedures for categories such as refunds and billing changes. These were ported into Procedures, with a structured mode for sensitive steps like refunds so they run the same way each time. Webhook tools let the agent pull data such as subscription tier and credit usage.
For quality control, the team set different behavior for email and voice, ran simulation tests on roughly 50 of the most common scenarios, and used shadow mode to watch the agent on live ticket traffic before going live. Support staff also use Architect, the company's natural language agent builder, to see top ticket categories, diagnose why cases were escalated and draft changes, with new tests added before updates ship.
For marketing and content teams that own or influence customer experience, the lesson is in the sequence: documented playbooks first, tests second, staged rollout third. Teams evaluating similar tools may want to ask vendors how they handle regression testing and production previews, and to keep humans responsible for policy and edge cases.
Sources
This brief is our own summary. Read the full reporting at the source.
