"Is my bot handling user requests successfully?"
To answer that, you need to visit the Conversations page, starting with the Most Frequent Topics metric. This metric tracks Kudos and Issues by analyzing the volume of kudos (expressions of satisfaction), neutral reactions, and issues. These issues break down into Incidents (security risks), Resolution Accuracy Issues, Dialog Quality Issues, Containment Quality Issues, Bounced (the chat popped up, but the user did not react), and Unclassified.
- non-functional services (e.g., users complaining about the inability to view account statements),
- poorly designed or missing dialogue flows (e.g., some users have multiple accounts, and the flow doesn’t include a path for this),
- topics that require better communication (e.g., user don’t understand why their payment order did not go through).
- Etc.
This leads to the language analysis table, showing the objects appearing with send and negative sentiment. If you want to exclude utterances dealing with euro payments, for instance, you can further refine the filter by using the regex !eur in the Expanded Objects column to exclude “euro,” “euros,” etc. from the results.
Below the table of actions and objects, there is a list of conversations filtered based on the criteria. You can open and review these individually to pinpoint their specific problems.
Related
- Conversations
- Topics, actions and objects
- Regex cheat sheet
- User utterances