What happens when the AI can't answer a question?
RavinaroWhat does the agent do when it doesn't know?
It stops. It tells the caller that a person is coming, and it alerts a named person on your team — not a shared inbox that nobody owns.
The failure that costs money is not the agent saying "I don't know". It is the agent inventing an answer, quoting a price that does not exist, or promising an appointment slot that is already taken. A confident wrong answer is worse than no answer, because the customer acts on it.
Who decides what counts as too hard?
You do, during the build. Escalation rules are part of the setup: any question about refunds, anything from an existing client, anything where the caller sounds unhappy, anything about a service you have not launched yet.
The rules are not fixed once. Most businesses tighten them in the first month, when they hear what actually comes up, and loosen them later as they learn the agent handles more than they expected.
Does a missed answer just disappear?
No. The exchange is transcribed, the escalation is logged, and it shows up in your monthly report. That report is the reason the care plan exists: the same three questions escalating every week is not an agent problem, it is a missing answer, and adding it is a five-minute change.
Can the agent be wrong?
Yes, and pretending otherwise would be dishonest. An AI agent can mishear a name, misclassify an enquiry, or state something inaccurate. We guarantee delivery and response speed, not commercial outcomes.
What reduces it is narrowing the job. An agent asked to answer anything about anything will be wrong more often than one asked to answer twenty known questions and escalate everything else.