Analytics & Performance Metrics

Track your AI agent's performance with real conversation metrics, sentiment analysis, and message feedback insights.

Where Analytics Live Today#

Hay's working analytics are found in two places:

• The Dashboard (/dashboard) - real conversation metrics, sentiment analysis, and live widgets. See Dashboard Overview.
• Feedback Insights (/insights/feedback) - real analysis of message feedback given by your team, with CSV export.

Important: The pages at /analytics, /analytics/reports, and /insights also exist, but they are UI previews populated with demo data - the numbers shown there are not calculated from your account. They are covered at the end of this guide so you can tell them apart from real data.

Only the Dashboard has a sidebar entry. There is no Analytics or Insights entry in the sidebar - Feedback Insights and the preview pages are reached by direct URL.

Real Metrics (Dashboard)#

The Dashboard shows metrics calculated from your actual conversations. All of them accept a date range filter and default to the last 30 days:

Metric Cards

• Active Agents - count of enabled agents, with new agents added this week
• Total Conversations - count of conversations in the selected period
• Resolution Rate - percentage of conversations marked resolved
• Avg Messages per Conversation - average number of messages exchanged

Conversation Activity

A daily count of conversations over time, shown as the Activity chart.

Sentiment Analysis

Hay's AI classifies the sentiment of customer messages as they arrive. The Dashboard shows an overall sentiment score plus the breakdown of positive, neutral, and negative messages for the selected period.

Document Status Overview

Your knowledge base documents grouped by processing status.

Additional Widgets

• Top Performing Agents - agents ranked by conversation count with resolution rate
• Recent Conversations - last 4 conversations with status and last message
• Active Conversations - live widget showing currently active conversations
• Escalations - conversations that have been escalated
• Attention Needed - conversations requiring attention

Not yet implemented: Response time analytics exist as an API endpoint but currently always return zero - the calculation is a placeholder. Any "average response time" figure you see in the product is not real data yet.

For a walkthrough of each Dashboard widget, see Dashboard Overview.

Feedback Insights#

Feedback Insights (/insights/feedback) is the one fully functional analytics page beyond the Dashboard. Its purpose: "Analyze AI response quality through user feedback."

How feedback is collected#

Feedback comes from your team, not from customers. When a dashboard user reviews a conversation, each AI message shows thumbs up / thumbs down controls. Clicking one opens a dialog with:

• Good / Bad buttons
• An optional Comment (optional) field ("What could be improved?")

Once submitted, the message shows "Feedback recorded" with an Edit link to change it later.

Ratings are stored as good, bad, or neutral - there is no star rating and no 1-5 scale. End customers cannot rate messages: the webchat widget has no feedback UI. Every rating is recorded with the reviewer who gave it.

Stats cards#

At the top of the page:

• Total Feedback - total number of ratings recorded
• Positive - count and percentage of good ratings
• Negative - count and percentage of bad ratings
• Success Rate - percentage of ratings that are positive

Filters#

Narrow the feedback list by:

• Rating - All Ratings, Good, Bad, or Neutral
• Source - the channel the rated message came from
• Start Date / End Date

Feedback History#

Each entry in the list shows the rating badge, the source channel, any comment left, a shortened message ID, and the reviewer's email. The list is paginated (20 entries per page). The rated message's text itself is not shown, and the link icon next to each entry does not navigate anywhere yet.

Export CSV#

Click Export CSV to download your feedback as feedback-export-<date>.csv. The export respects the Rating, Start Date, and End Date filters (the Source filter is not applied to the export). This is currently the only working data export in Hay's analytics.

Using Feedback to Improve#

A practical review routine:

1. Check the Success Rate weekly - is it trending up?
2. Filter by Bad ratings and read the comments
3. Look for patterns: a topic the agent keeps getting wrong usually means a missing or outdated document
4. After changing agent instructions or documents, watch whether negative feedback on that topic drops

While testing a new agent, rate its messages liberally - the feedback history becomes your quality log.

Demo / Preview Pages#

The following pages exist in the product but display hardcoded demo data. Do not use them to make decisions.

/analytics - Analytics#

Shows KPI cards (Total Conversations, Resolution Rate, Avg Response Time, Customer Satisfaction), Response Time Distribution, Top Issues, an Agent Performance table, and Channel Performance. All of these figures are demo placeholders, including the star-based satisfaction scores. The timeframe selector, Export, and Refresh buttons do nothing yet. The two chart areas ("Conversation Volume", "Resolution Rate Trend") are empty placeholders. The only real element is the list of channels itself, which reflects your enabled channel plugins - the metrics next to them do not.

/analytics/reports - Custom Reports#

A preview of a report builder (Report Configuration, Metrics Selection, Grouping & Filters, Visualization, Quick Templates, Recent Reports). Generating, saving, scheduling, viewing, and downloading reports are all not implemented - the buttons do nothing yet, and the "Recent Reports" list is demo data.

/insights - Insights#

Shows sample AI-generated improvement suggestions. Accepting or rejecting them only updates the demo list on screen - nothing is saved - and the page itself displays a "Coming soon" notice.

Common Analytics Questions#

Can customers rate my agent's responses?#

Not yet. Rating is an internal review tool: dashboard users rate AI messages with thumbs up/down inside the conversation view. The webchat widget shown to customers has no rating UI.

Where did the "Customer Satisfaction" score come from?#

If you saw it on the /analytics page, it is demo data. Hay's real quality signal is the Success Rate on Feedback Insights, calculated from your team's good/bad ratings.

Can I export analytics data?#

Yes, one export works today: Export CSV on the Feedback Insights page. PDF exports and scheduled reports are not implemented.

What's a "good" resolution rate?#

Depends on your industry and complexity, but generally:

• Simple FAQ-based support: 80-90%
• E-commerce support: 70-85%
• Technical support: 60-75%
• Complex B2B: 50-70%

Brand new agents need time: expect metrics to improve over 2-4 weeks as you add documents, refine playbooks, and adjust agent instructions.

Next Steps#

• Review real metrics on your Dashboard
• Use feedback insights to improve your Agents
• Add Documents for topics with negative feedback
• Monitor your Conversations for escalation patterns