Copilot Studio Agent Evaluation: Single Response vs. Conversation
In Microsoft Copilot Studio, evaluating your agent's performance before deployment is crucial. The Evaluation feature offers two data types: Single Response and Conversation (preview). Single Response treats each interaction independently, suitable for testing knowledge base queries and factual accuracy. Conversation evaluation, however, maintains context through multi-turn dialogues, ideal for workflows requiring information continuity. For instance, in an HR assistant scenario, Single Response failed to correctly book sick leave due to lack of context, whereas Conversation successfully completed the task by keeping track of user inputs across turns. Choosing the appropriate evaluation type ensures effective testing of agents for their intended tasks.
Article 5w
Login now to access my digest by 365.Training
Learn how my digest works
Features
- Articles, blogs, podcasts, training, and videos
- Quick read TL;DRs for each item
- Advanced filtering to prioritize what you care about
- Quick views to isolate what you are looking for right now
- Save your favorite items
- Share your favorites
- Snooze items you want to revisit when you have more time