Agent Performance
See how each agent handles its calls, compare agents and find what to change when one underperforms.
Showing instructions for Xebit VoiceCloud.Showing instructions for Xebit VoiceKit, on your own infrastructure. Switch with the selector at the top. Updated 22 October 2026.
Agent performance tells you how well each agent is doing its job: how many calls it handled, how many reached a proper end, and what those calls achieved. Look at it per agent, because a reception agent and a collections agent succeed in different ways.
Where to find it
- Sign in to the Xebit VoiceCloud dashboard at
voice.xebit.ai. - Sign in to the Xebit VoiceKit admin console on your own host.
- Open the calls list and filter by agent.
Every call that agent placed or answered is now in front of you, with its outcome and captured fields.
What to look at per agent
| Measure | What it tells you |
|---|---|
| Total calls | How much work the agent is carrying |
| Completed calls | How many conversations reached their end |
| Failed to connect | How many calls never connected. A high number usually points at the contact list or the carrier, not the agent. |
| Caller hung up | How many people left before the end. This one is about the agent. |
| Inbound and outbound | The direction split, for agents that do both |
| Call duration | Unusually short calls can mean a weak opening; unusually long ones can mean the agent is looping |
| Outcomes from your fields | What the calls achieved: leads qualified, promises to pay, appointments booked, callbacks requested |
| Handoffs to a person | How often the agent passed the call on, if you capture it as an outcome |
Compare agents
If you run more than one agent on similar work, filter the calls list for each in turn over the same date range and compare:
- which agent completes more of its connected calls;
- which agent loses fewer callers in the first minute;
- which agent produces more of the outcome you care about, such as bookings or qualified leads.
Compare like with like. An agent calling a cold list will always look worse than one calling customers who asked for a callback.
Behaviour signals in Observability
Observability watches agent behaviour as well as infrastructure. For each agent it tracks failed tool calls, off script answers and prompt regressions after a change, alongside response time per turn. If an agent's results dropped right after someone edited its instructions, this is where you will see it. See Observability.
Improving an agent
- Listen to a few calls that went badly and read their transcripts. Find the moment the conversation turned.
- Open the agent's settings and review its instructions. Vague instructions produce vague conversations. See Agent Prompting.
- Change one thing at a time, so you know what made the difference.
- Start a test call and walk through the same situation that went wrong.
- Watch the next day's calls to confirm the change helped.
See Agent Testing for a fuller guide.
Next steps
- Call Analytics: the calls list and its filters.
- Call Recordings: listen to one agent's calls.
- Transcripts: read what the agent said.