A customer messages your support agent: "This is incredibly frustrating. Why can't you just refund my order?" The order is straightforwardly refund-eligible per policy. The agent has the resolution one tool call away. The customer has not asked for a human.
What's the right behavior?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
Frustration without an explicit request for a human is different from explicit escalation. When the agent has the resolution and the issue is within its capability, the right move is to acknowledge what the customer is feeling, surface the fix, and execute. Escalating on frustration alone trains a system that can't handle anyone who's upset — which is most of the people contacting support. Note the contrast with a customer who explicitly asks for a human: that request takes precedence and is honored immediately.
Frustration is a signal about emotional state, not about whether the agent can solve the problem. If you escalate every frustrated customer, your support system becomes a queue management system for humans.
Silently issuing the refund without acknowledging what the customer said registers as the agent ignoring them — which compounds the frustration even when the underlying issue is resolved. Acknowledge, then act.
A "we understand your frustration" reply that's followed by more questions rather than the available resolution is the worst of both worlds — it acknowledges the feeling without addressing the substance, which is what the customer is actually asking for.