Users occasionally ask questions outside the agent’s scope. The business wants a polite message, optional suggested topics, and an escalation path — not a silent failure. Which system capability should you customize?

System topics include Greeting/Conversation Start, Fallback, Escalate, End of Conversation, and others. Fallback handles cases where intent matching fails. The correct capability is the Fallback topic, or an equivalent system topic that handles unrecognized utterances, because customizing it provides controlled UX for unknown intents with guidance, suggested topics, and transfer options. The Conversation Start topic only is wrong because Conversation Start is for session beginning, not ongoing unknown intents, and fallback can run after the first message. The End of Conversation topic only is wrong because End of Conversation is for closing, not matching failures. Disabling all system topics so unmatched intents raise a raw platform exception to the user is wrong because it harms user experience and supportability. Exam tip: Map unknown intent to Fallback; explicit human request to the Escalate topic; low confidence may also route to escalate depending on configuration.