Escalation to a human agent
Escalation to a human agent is the handover of a conversation from the automated system, chatbot or AI agent, to a person, triggered when the request exceeds what automation can do, touches sensitive topics, or when the user explicitly asks for it. It's the mechanism that makes automation sustainable: the AI absorbs the volume, the agent handles the exceptions that matter.
The recurring triggers: an explicit user request to speak with a person, a question outside the knowledge base, signs of frustration or a complaint, operations that require human authorization. At handover, the agent must find the context ready, with the conversation history and the contact's data, without making the customer repeat anything. In well-designed systems, the AI stays suspended on that chat until the agent returns it to automation.
The governing metrics: escalation rate, i.e. the share of conversations passed to a human, time to pick up, outcomes after the handover. Too high a rate signals insufficient training; too low a rate can hide customers trapped in the automation — an experience Meta's policies ask you to avoid by always providing a route to an agent. After hours, escalation becomes deferred pickup with clear expectations on timing.
In SendApp, the agent can step into any chat by suspending the AI for that conversation with a toggle, and reactivate it once the case is resolved; the team works from the shared inbox.
Frequently asked questions
From theory to practice
Everything you read in this glossary can be done with SendApp: try it free, no credit card required.