Agent Testing
Test calls, what to check and how to fix what you hear before going live.
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.
Test every agent before it talks to a real customer. A few test calls catch most problems: a wrong greeting, a missed question, a transfer that never happens. They take minutes and save you from a customer finding the problem first.
What testing catches
- A greeting that is wrong or unclear
- Information that should have been captured and was not
- A handoff that does not happen when it should
- Replies in the wrong language
- A booking that is not confirmed or not written to your system
- A workflow action that fails
How to make a test call
From the agent
- Open the agent in the dashboard at
voice.xebit.ai. - Open the agent in the Xebit VoiceKit admin console.
- Start a test call and enter your own phone number.
- The agent calls you over your carrier. Talk to it as a real customer would.
Call the agent's number
For an inbound agent, call the number on your SIP trunk that is routed to it, from your own phone, and go through the conversation.
What to test
| Check | What good looks like |
|---|---|
| Greeting | The right opening line, natural and clear about who is calling |
| Understanding | Copes with casual speech, pauses and a mix of languages |
| Interruptions | Stops and listens when you cut in |
| Questions | Asks for everything you configured |
| Booking | Checks availability, reads the slot back and confirms it |
| Handoff | Hands over the moment a trigger is hit, such as asking for a manager |
| Language | Answers in each language your callers use |
| Out of hours | Behaves as configured outside its hours |
| Closing | Ends the call politely and naturally |
What to listen for
- Long pauses before the agent answers
- Wrong or confusing answers
- Skipped questions
- A stiff, scripted tone
- Handoffs at the wrong moment, or not at all
Review the record afterwards
Open the test call's details and check the transcript, the summary, the outcome and every captured field. Check your CRM or endpoint received what it should. If a workflow action ran, check the event log for its response.
If something is wrong
- Open the agent's instructions.
- Find the part that caused the problem and rewrite it more clearly, or add a line for the situation it missed.
- Save and test again.
Most problems come from instructions that are vague or that never mention a situation. A small change in wording usually fixes it.
Test more than once
Run several test calls before going live, each with a different scenario:
| Scenario | What to check |
|---|---|
| An ordinary call | Greeting, questions, captured fields, closing |
| A booking | Availability check, read back, booking written to your system |
| A handoff trigger | The call reaches a person at the right moment, with context |
| An unexpected question | How the agent handles something it was not told about |
| Another language | Accuracy and consistency in that language |
Test calls are real calls over your carrier, so your carrier charges for them as usual.
Test calls are real calls over your carrier, so your carrier charges for them as usual, and they use capacity on your deployment like any other call.
Next step
When the agent passes, route live calls to it or test it in a small campaign.