According to ElevenLabs, its internal agents allow teams to save over 1,200 hours per month in their daily operations. These agents answer more than 5,000 questions monthly, and they also take action, building demos for clients and opening tickets.
These tools were first used to more easily find information in Slack, the company’s professional messaging platform, whether it was user retention data, documentation, or an account’s history. They then evolved into agents capable of taking action: an employee asks a question, reports a bug, or requests a demo, and the agent responds or executes the request.
Sales and On-Demand Demos
The sales teams use two agents. The “GTM bot,” for go-to-market, answers questions about clients, while the “FDE bot,” where FDE stands for Forward Deployed Engineer, i.e., an engineer deployed with clients, handles product questions and demos. Among recent examples, a salesperson asks “Who are our key clients with logo usage rights in the UK, France, and Germany?”, while another question concerns the main use cases for the API.
The FDE bot can also assemble, from a simple message, a functional agent configured for a prospect’s use case. These are complete demos produced on demand.
Support and Engineering: Automating Triage
The “triage bot” handles feature requests and bugs, following a different procedure for each type of issue. Before creating a ticket, it searches for relevant context in Slack, checks for duplicates among existing tickets, and ensures the information is sufficient to diagnose the problem. The ticket is then opened in Linear with the appropriate priority and forwarded to the relevant team via a responsibility assignment document added to the agent’s knowledge base. This work, according to the company, was previously very manual between support and engineering.
The product and product marketing teams ask their data questions in dedicated channels and receive the analysis directly in the thread. An example cited by ElevenLabs asks how the conversion of free users to ElevenCreative paid subscriptions has evolved over the last six months in the United States, and what the main factors are. Employees can then discuss or ask follow-up questions in the same thread. However, the quality of the answers depends on what the agent can consult, hence the careful selection of each agent’s data.
What the Company Gets Out of It
ElevenLabs emphasizes that these agents must be reliable and accurate from the start, otherwise teams stop using them. In a shared channel, errors are more visible, but successes are too, which has fostered adoption.
According to the company, the clearest gain in answer quality does not come from rewriting instructions, but from removing fields. “Schema curation beats prompt engineering,” it writes. Each additional column or field gives the model an opportunity to make a mistake. For Salesforce fields with fixed values, like an opportunity stage or account type, valid values are now fetched at startup and injected into the MCP tool definitions—a protocol that connects models to external tools—so the model can only choose from existing options. Finally, ElevenAgents’ Architect restructures overlapping procedures, and each ticket created in Linear is checked for its fields and destination team.
