Users of a Copilot Studio agent are reporting that when they ask questions the agent cannot answer, they receive a generic 'I'm sorry, I don't understand' message that doesn't provide any guidance on what the agent can help with. The developer wants to improve this experience. What should the developer modify?
Select an answer to reveal the explanation.
Short Explanation
The Fallback system topic is your safety net message, so customize it to guide confused users instead of leaving them with a cold, unhelpful reply. You’ll list what the agent can help with and offer escalation when needed.
Full Explanation
Copilot Studio includes several system topics that control core agent behaviors. The 'Fallback' system topic fires whenever no other topic's trigger phrases match the user's input. It is fully customizable.
The correct approach is to modify the 'Fallback' system topic to include a more helpful message that lists what the agent can assist with and offers to escalate to a human agent. By editing the 'Fallback' system topic, the developer can replace the generic message with a helpful, branded experience: list what the agent can help with, offer alternative contact options, or trigger an escalation to a live agent via the 'Transfer to Agent' node. This is a high-impact customization that significantly improves user experience for out-of-scope queries.
The custom 'Help' topic with trigger phrases for all possible user confusion patterns is incorrect. Creating a 'Help' topic with trigger phrases doesn't address the fallback scenario — users who fall through to 'Fallback' have already failed to match any topic. The 'Fallback' topic needs to handle unrecognized inputs, and trigger-phrase-based topics cannot cover all possible unrecognized inputs.
Increasing the number of trigger phrases on all existing topics so the agent recognizes more variations and falls back less often is incorrect. Adding more trigger phrases to existing topics reduces the fallback rate but doesn't fix what happens when users do fall back. Both improvements are valuable, but the question specifically asks about improving the fallback experience.
Enabling generative answers on the agent so that it can answer any question using AI without falling back is incorrect as a complete solution. Enabling generative answers may handle some fallback cases, but it requires a configured knowledge source and may still fail for off-topic questions. Modifying the 'Fallback' topic is the correct answer for controlling the experience when the agent genuinely cannot help.