The dashboard is organized into three main tabs:
- Bot Performance: allows you to analyze interaction volume, the assistant’s response paths, and key metrics related to retention, evasive responses, handoffs, and RAG performance.
- GenAI Knowledge: allows you to evaluate knowledge base usage, identify the percentage of interactions answered, and detect opportunities to improve documents, intentions, and response quality.
- Conversations: allows you to analyze sessions by category, intention, channel, and resolution status, as well as review individual conversations and their details.
Tip: To get a complete view of the assistant’s performance, follow this workflow:1) Start with Bot Performance to identify interaction volume, how interactions are distributed across response paths, and the main trends.
2) Continue with GenAI Knowledge to identify the documents and intentions involved, analyze unanswered responses, and detect the most frequent types of issues.
3) Finish with Conversations to review specific cases, understand what happened, analyze resolution status, and define improvement actions.
How can you correlate the main metrics?It’s important not to interpret any metric in isolation. Here are some key relationships to consider when analyzing dashboard data:
- Sentiment is not the same as resolution: sentiment describes how the user expressed themselves, not whether their need was resolved. An upset user may still have their need resolved, while a polite user may be left without an answer.
- Response quality is not the same as resolution: response quality evaluates the AI’s response, while resolution status indicates whether the user’s need was met. An incomplete response may be followed by a handoff to an agent who resolves the issue.
- Transferred does not necessarily mean a failure: some processes are expected to involve a handoff to an agent. To interpret the handoff rate, it’s best to analyze it alongside response quality and evasive responses.
- Evasive responses + conversational friction + poor responses: these metrics help identify improvement opportunities from different perspectives. Evasive responses indicate that the assistant could not respond adequately; friction reflects situations in which the user had to rephrase their question or repeat information; and poor responses point to response quality issues. For example, if the same document or intention has many evasive responses, high conversational friction, and a high percentage of poor responses, its review should be prioritized.
- Unidentified label and unidentified intention: queries that cannot be assigned to a configured category or intention can help uncover new user needs and identify opportunities to create or adjust intentions.
Learn what each metric and chart in the GenAI dashboard means in the walkthrough below.