ElevenLabs adds trigger-loaded SOPs to voice agents
By Steven Van ·
Alpha "Procedures" in ElevenAgents load playbooks only when a matching situation comes up, keeping agents focused.
ElevenLabs has introduced Procedures in ElevenAgents, a new way to define exactly how conversational AI agents behave in specific situations. Think of Procedures as standard operating procedures (SOPs) for your agents: packaged playbooks that spell out what to do, when to do it, and in what order. Crucially, the instructions load only when a matching situation is triggered, keeping the agent focused rather than drowning it in a giant prompt.
What Procedures Actually Are
A procedure tells your agent how to handle one specific task. Each procedure has two parts: a trigger that describes the situations it applies to, and content that describes what the agent should do in those situations. Taken together, an agent's set of procedures works like an employee handbook — one entry per common scenario. When the conversation hits a situation that matches a trigger, the relevant procedure loads, and the agent follows it. The take an order-ID, check-eligibility, issue-refund flow that ElevenLabs used to demo the feature is a textbook example: the refund playbook only surfaces when a customer actually asks for a refund.
Structured vs. Free-Form
ElevenLabs ships two flavors. Structured procedures are an ordered list of typed steps that the agent runs the same way every time — ideal when the sequence is non-negotiable, such as verifying a caller's identity, escalating a ticket, or taking a payment. Free-form procedures, by contrast, are written as natural-language instructions the agent interprets and adapts to the situation. The split lets teams enforce compliance-critical flows with rigid steps while keeping softer, judgment-heavy interactions flexible. A single agent can mix both, choosing the right playbook for each moment in the conversation.
Why Load-On-Trigger Matters
The headline design choice is that procedure instructions load only when triggered rather than sitting in the base prompt the whole time. That keeps the agent's active context lean and its behavior predictable — it is not juggling refund logic, account-recovery logic, and troubleshooting logic all at once. The practical payoff is better outcomes across a range of goals: higher upsell conversion, faster ticket resolution, and more consistent handling of edge cases. For support teams especially, that means each request type — refunds, identity verification, account recovery, connection troubleshooting — can get its own dedicated, well-tested playbook.
Where This Fits for Creators and Teams
Procedures round out ElevenLabs' push into agentic voice and chat, sitting alongside its visual Workflows editor. If you are building a support line, a booking assistant, or an outbound sales agent, Procedures give you a maintainable middle layer between raw prompts and full custom code: update one playbook without rewriting the whole agent. The feature is currently in Alpha, so expect the tooling and step types to evolve — but the core idea of composable, trigger-scoped SOPs is a strong signal of where production voice agents are heading.
Sources: Procedures — ElevenLabs Docs, Structured Procedures — ElevenLabs Docs, ElevenAgents Overview.