Analytics and Performance
Understand your AI agent's performance with containment rate, reaction scores, and knowledge coverage metrics.
The Analytics page gives you quantitative insight into how well your AI agent is performing, where it is falling short, and how your knowledge base coverage is improving over time.
Opening Analytics
Click Analytics in the left sidebar.
Time window
Use the days selector to change the look-back window. Available options: 7, 14, 30 (default), and 90 days.
Tier 1 metrics (all plans)
These metrics are available on every plan.
Containment Rate
The percentage of sessions resolved entirely by the AI without escalating to a human agent. A well-tuned knowledge base should achieve 70 %+. If this is low, check the KB Gaps report and ingest more content.
Escalation Rate
The inverse of containment rate — the percentage of sessions that were handed off to a human agent. Target: below 30 %.
Total Sessions
The number of unique conversation sessions started in the selected period.
Total Queries
The total number of individual messages sent by customers across all sessions.
KB Coverage Score
The percentage of queries for which the AI found at least one matching wiki entry. A low coverage score means your knowledge base is missing content — use Ingest to add it.
Average Response Latency
The median time (in milliseconds) between receiving a query and returning an answer.
Tier 2 metrics (Pro plan and above)
These additional metrics are unlocked on the Pro plan.
Helpful Rate (reactions.helpfulRate)
The percentage of sessions where the customer rated the AI response as helpful (thumbs up). Calculated from reactions.helpful / reactions.total. Target: 75 %+.
Flagged for Review Rate
The percentage of interactions flagged by the AI or by customers for human review. A high rate may indicate content gaps or AI uncertainty.
Average Human Handle Time (avgHumanHandleTimeMs)
The average time in milliseconds between an escalation being claimed and being resolved. Monitor this to track your support team's response efficiency.
Reaction types
Customers can rate any AI response with one of these reactions:
- thumbs_up — helpful
- thumbs_down — not helpful
- flag — flagged for review
- copy — copied the response
KB Gaps report
The KB Gaps tab (accessible from the Analytics page) shows which customer queries had no matching wiki entry. Use this list to prioritise what to ingest next.
Acting on your analytics
| Metric is low | What to do |
|---|---|
| Containment rate | Ingest more content; review KB gaps |
| KB coverage score | Add missing documentation via Ingest or Sources |
| Helpful rate | Edit wiki entries with low confidence scores; check Review Queue |
| High escalation rate | Lower the escalation threshold in Settings or add more wiki content |