AI Conversation Analytics Dashboard

Summary

The AI Analytics dashboard shows how the AI handles customer conversations. It provides performance metrics, intent breakdowns, outcome distributions, and a searchable audit table of individual conversations. Use it to monitor AI resolution rates, spot trends, and identify conversations that needed human escalation.

Prerequisites

• You must have the messages.view capability.
• Communication AI must be enabled for your organisation. If it is not enabled, the tab does not appear and navigating to /messages/ai-analytics redirects to /messages.

Steps

Accessing the dashboard

1. Go to Messages from the main navigation.
2. Select the AI Analytics tab in the sub-nav bar.
3. The page opens with the header “AI Analytics” and subtitle “Monitor AI conversation performance and identify improvement areas”.

Selecting a time period

1. In the top-right corner, a period selector shows three options: Last 7 days, Last 30 days, and Last 90 days.
2. Select a period. The default is Last 7 days. All metrics, charts, and the audit table update to reflect the chosen period.

Reading the stat cards

1. Four stat cards appear at the top of the page in a single row (two columns on smaller screens):

   • Resolution Rate - percentage of AI conversations resolved without human intervention. Shows a green icon. A trend indicator compares against the previous equivalent period (e.g. previous 7 days) with an up or down arrow.
   • Escalation Rate - percentage of conversations escalated to a human. Shows an amber icon. For this metric, an increase is shown in red (going up is negative).
   • Avg Response Time - median AI response time. Shows a blue icon. Target is under 5 seconds. Displays as seconds (e.g. “3s”) or minutes (e.g. “1.2m”). Shows “Within target (<5s)” or “Above target (>5s)”.
   • Total Conversations - count of AI-handled conversations in the period. Shows a purple icon. Trend compares with the previous period.

2. If there is no change between periods, the trend reads “No change”.

Reading the intent breakdown chart

1. Below the stat cards, the left chart is titled Intent Breakdown.
2. This is a bar chart showing the distribution of detected customer intents. Each bar represents an intent category (e.g. status inquiry, booking request, quote question) with its count and resolution rate.
3. If there is no intent data for the selected period, a placeholder message reads “No intent data for this period”.

Reading the conversation outcomes chart

1. The right chart is titled Estimated Resolution Success.
2. A tooltip icon explains: “Based on conversation activity patterns. A completed conversation may not always indicate satisfaction.”
3. This is a donut chart showing four outcome categories:
   • Completed (green) - conversation finished successfully
   • Continued (blue) - conversation is still ongoing
   • Dropped off (amber) - customer stopped responding
   • Escalated (red) - conversation was handed to a human
4. If there is no outcome data, a placeholder message reads “No outcome data for this period”.

Using the conversation audit table

1. Below the charts, the Conversation Audit table lists individual AI conversations.
2. The table header shows the total count of matching conversations (e.g. “142 total”).
3. Filter controls appear to the right of the header:
   • Intent filter dropdown - options include All intents, Status Inquiry, Booking Request, Quote Question, General Question, Escalation Request, Continuation, Spam, and Unknown.
   • Outcome filter dropdown - options include All outcomes, Resolved, and Escalated.
   • Date from and Date to date pickers for a custom date range.
4. The table columns are:
   • Date - sortable, shows date and time in “dd MMM yyyy HH:mm” format
   • Customer - customer name
   • Intent - detected intent with underscores replaced by spaces, or a dash if none
   • Confidence - AI confidence score as a percentage, or a dash if not available
   • Outcome - colour-coded badge showing Resolved (green), Escalated (red), or Active (blue)
   • Messages - sortable, total message count in the conversation
5. Select a column header (Date, Confidence, or Messages) to sort. Select again to toggle between ascending and descending order. An arrow icon indicates the current sort direction.
6. Select any row to navigate directly to that conversation in the messaging inbox at /messages?conversation={id}.

Paginating the audit table

1. If there are more than 20 results, pagination controls appear below the table.
2. The footer shows “Page X of Y” and Previous / Next buttons.
3. Changing a filter or sort resets pagination back to page 1.

Expected Outcome

• You can see a clear summary of AI conversation performance across chosen time periods.
• Trend indicators show whether metrics are improving or worsening compared to the previous period.
• The intent breakdown reveals what customers are asking about most.
• The outcomes chart shows the proportion of conversations resolved by AI versus escalated.
• The audit table lets you drill into specific conversations to review AI handling.

Troubleshooting

Problem	Cause	Fix
AI Analytics tab does not appear	Communication AI is not enabled for your organisation	Ask an admin to enable AI in communication settings
Page redirects to /messages	AI is disabled or you do not have the messages.view capability	Check with an admin that AI is enabled and your role has the correct permissions
All stat cards show 0 or dashes	No AI conversations occurred in the selected time period	Try selecting a longer period (30 or 90 days)
Charts show placeholder messages	No intent or outcome data available for the period	This is normal if AI has not processed enough conversations. Data will appear as more conversations are handled
Audit table shows “No AI conversations found matching filters”	Active filters exclude all results	Clear the intent and outcome dropdowns to show all conversations, or widen the date range
Clicking an audit row does nothing	The conversation may have been deleted	The row navigates to /messages?conversation={id}. If the conversation no longer exists, the inbox will show an empty thread

Notes

• The estimated resolution success metric is based on conversation activity patterns. A “completed” conversation may not always mean the customer was satisfied.
• The default resolution threshold is 24 hours of inactivity.
• Confidence scores are shown as percentages rounded to the nearest whole number.
• Response time under 1 second displays as “<1s”.