When an AI voice agent is the wrong answer

Voice agents have crossed the line where a short, well-scoped call is indistinguishable from a competent human one. That makes it tempting to point them at every phone number you own.

They work best where the conversation has a narrow goal and a clear success state: confirming an appointment, collecting a missing detail, qualifying an inbound lead, or rebooking a no-show.

They fail where the caller is upset, where the answer depends on judgement, or where the outcome is irreversible — refunds, cancellations, medical advice, anything legal. In those cases the agent should identify the intent and transfer in under fifteen seconds.

The rule we use: if a wrong answer costs more than the call saved, a human takes it. Everything else is fair game for automation.