Deflection details page in Assistant analytics

View detailed insights into how your assistants resolve user queries, including deflection outcomes and the types of deflection methods used.

The Deflection Details page provides a granular view of how your assistants handle user queries and the methods used to resolve them. Access this page by navigating to Assistant Designer > Analytics > Self Solved Performance > Deflection Details.

Figure 1. Deflection details page in Assistant analytics

The visualizations on the Deflection Details page help you with the following.

• Understand the distribution of deflection outcomes to assess how effectively your assistants resolve user queries.
• Identify the deflection methods that contribute most to successful resolutions.
• Drill down into individual conversations to analyze deflection patterns and identify areas for improvement.

You can filter the dashboard data using the following options:

• Date: View metrics for a specific date range.
• Assistants: View metrics for specific assistants.
• Channels: View metrics by communication channel, for example, Teams, Slack.

Deflection Outcome Distribution

This area of the dashboard shows the distribution of outcomes for deflection attempts. The chart displays breakdown of Resolved, Not Resolved, Response Provided, No Response Provided outcomes across conversations, helping you understand how often deflection attempts lead to successful resolution.

Figure 2. Deflection Outcome Distribution

The following are the deflection outcomes:

• Resolved: conversations where at least one AI asset, for example, AI agent, skill, topic was executed or a positive feedback was provided by the user
• Response Provided: conversations involving small talk or where the assistant provided a synthesized response with no further action from the user
• No Response Provided: conversations where no response was provided by the assistant
• Not Resolved: conversations where negative feedback was provided by the user or the query was escalated to a live agent

For more information on how deflection events are logged and deflection outcomes are attributed, see KB2822316.

Deflection Types Offered

This area of the dashboard shows the distribution of AI assets offered as deflection methods to users. For example, catalog item, synthesized response, knowledge article, and so on. Use this chart to help you understand which deflection methods are being employed and evaluate their relative usage.

Figure 3. Deflection Types Offered

The following are the deflection types offered by the assistant:

• Agentic: Resolutions handled by AI agents.
• Synthesized: Responses generated by synthesizing information from multiple sources.
• Catalog: Resolutions that directed users to catalog items.
• Topic: Resolutions handled by predefined conversation topics.
• KB: Resolutions that provided knowledge base articles.
• Doc QnA: Resolutions from document-based question and answer.
• Web Search: Resolutions that used web search results.
• Others: Other deflection methods not categorized above.

Use this visualization to help you understand which deflection methods are most effective and optimize your assistant configuration accordingly.

Conversation details

This area of the dashboard displays a list of individual conversations based on the drill downs you select in either of the charts. When you select a conversation, its details appear in a panel on the right side of the page. Collapse the panel when you need more space to browse the conversation list.

Figure 4. Conversation details section in Deflection details page

The conversations list contains the following details:

• Assistant: The assistant that handled the conversation.
• State: The deflection outcome of the conversation.
• Conversation: A unique identifier for the conversation. Select the link to view the full conversation details, including the transcript.
• Deflection type: The deflection method used to resolve the query.
• Created: The date and time when the conversation was created.

Use the Export button to download the conversation data for further analysis.