Analytics and improvement

The Analytics area turns stored conversations and feedback into four views:

• Overview shows sessions, visitors, messages, leads, escalation and resolution rates, feedback, and response latency.
• Topics groups common asks and compares their volume, helpfulness, confidence, escalation, and lead rate.
• Quality brings together thumbs feedback, low-confidence and no-source signals, sentiment, latency, and end-of-chat ratings.
• Traffic breaks activity down by page, device, browser, location, day, or hour.

Use date and comparison controls consistently. A percentage based on a handful of rated chats is a clue, not a verdict.

A weekly improvement loop

1. Open Quality and inspect negative, low-confidence, and no-source conversations.
2. Check Topics for a repeated question with weak results or rising volume.
3. Read the underlying conversations; do not diagnose from a chart alone.
4. Fix the authoritative website page, file, or FAQ.
5. Retest several phrasings in AI Demo.
6. Watch the same topic in the next comparable period.

Read the signals carefully

• A low-confidence answer may be appropriately cautious, not necessarily wrong.
• A lead or escalation can be a successful outcome when the question requires a person.
• Negative feedback is valuable but sparse; read the transcript and cited source before changing the prompt.
• Traffic differences can reflect audience intent. Compare like-for-like pages and periods.

Prefer small, traceable changes. If you rewrite a source and the system prompt at the same time, it becomes difficult to know which change affected the result.