Analytics
Analytics tells you whether the agent is getting better, not just busier.
Analytics
Analytics tells you whether the agent is getting better, not just busier.
Every headline number is compared with the period just before it, and charts are grouped by day (or by hour for the last 24 hours) in your own time zone. When something has not been measured yet, Analytics says so instead of showing zero.
Dashboard vs Analytics
Use Dashboard for the quick daily view.
Use Analytics when you need to know whether the agent is doing its job, what it gets wrong, and what to fix next.
Outcomes
The top row answers four questions about the selected period.
| Metric | What it means |
|---|---|
| AI resolution | Ended conversations the AI handled alone, with no teammate involved and no rating of 2 or below. A conversation has ended when it is closed or has had no message for 24 hours. |
| Escalation rate | Conversations that were handed to a teammate at least once, or where a teammate replied. Also shows the median time for a teammate to pick up an escalation. |
| CSAT | The share of ratings that are 4 or 5, with the average rating and how many conversations were rated. |
| Cost per conversation | Chat spend divided by conversations started, plus cost per resolved conversation. Only admins see cost; other roles see conversation volume here instead. |
A falling escalation rate is shown as an improvement, and a rising one as a regression.
Needs Attention
This list links straight to the evidence behind each problem:
- conversations rated 2 or below, with the customer's comment
- questions where the agent searched the knowledge base but found no matching source, most frequent first
- tools whose calls are failing
- knowledge sources that failed to process
- the share of replies that ended in an error
Start here when you want to improve the agent. Most items are fixed by adding or correcting knowledge, or by repairing a tool.
Speed
Speed comes from recorded agent runs for text chat:
- First token is how long a customer waits before the reply starts appearing.
- Full reply is how long the whole turn takes.
Both show the median and the 95th percentile. The p95 is what your slowest one in twenty customers experiences.
Conversations
Conversation volume over time, including quiet days, and the split by channel.
Operations
Collapsed by default. Shows replies and tokens by model, the retrieval cache hit rate among replies that searched the knowledge base, and knowledge base size.
Good Questions Analytics Can Answer
- is the agent resolving more conversations on its own after a change?
- which customer questions does the knowledge base not cover yet?
- are teammates picking up escalations quickly enough?
- did a model or tool change make replies slower?
- what does a resolved conversation cost?
Best Practices
- compare analytics with real conversation review, not in isolation
- watch outcome changes after updating sources, settings, or interceptors
- work through Needs Attention before scaling traffic
- use analytics to prioritize improvements, then verify them in Conversations
Related Docs
Conversations
Review chats, reply as human, and inspect voice sessions.
Feedbacks
Review low-rated chats and comments to find quality issues.
Usage & Costs
Track provider spend across AI, voice, and ingestion.