Question

You are designing a customer-facing agent in Copilot Studio for a utility company. Business stakeholders report that when customers phrase questions in ways the agent does not anticipate, the agent responds with abrupt 'I did not understand' messages and ends the conversation, frustrating users. You must configure the agent so unrecognized inputs are handled gracefully—attempting to reason over connected knowledge sources before escalating to a live agent. Which configuration approach should you implement?

Accepted Answer

Customize the system Fallback topic to invoke generative answers over the configured knowledge sources, then route to human handoff only if no grounded answer is found

Answer

Create additional trigger phrases on every existing topic so all possible user utterances are matched to a defined topic

Answer

Disable the Fallback topic entirely so the agent silently ignores unrecognized inputs and waits for the user to rephrase

Answer

Lower the trigger recognition threshold to zero so the closest-matching topic always fires regardless of confidence