Test an AloAi Voice Agent before it takes live calls, and review its calls once it does. You can test as a real or mock contact, replay saved test runs, preview the voicemail message, and see which version of the agent each test ran against. After the agent goes live, call latency details and the Agent Activity dashboard show how it performs. This article covers each of these tools.
Placing a test call
Click Test your Agent on the agent screen, then click the Call icon and start talking to the agent.
Action functions, such as Send SMS, do not run during a test.
Testing as a contact
When no contact is selected, the test runs anonymously, so contact-based dynamic variables do not resolve. To test the way a live call works, use the Test as contact picker. The agent loads that contact’s fields and resolves contact-based dynamic variables as it does on a live call. Agent-level dynamic variables also resolve during the test.
You can test as:
A real contact. Search your Aloware contacts by name, phone, or email, and select one. Testing as a real contact does not send messages to that contact.
A mock test contact. Five mock test contacts are available. Each one has test values for every contact field used in dynamic variables, so no variable is left blank.
The test panel shows which contact the test is running as. You can change or clear the selection at any time.
Reviewing test runs
Every test you start from the Test modal is saved automatically. Saved runs appear in the History tab of the Test modal, newest first, with a badge showing how many runs are saved.
Each run shows:
The contact the test ran as, or “Anonymous test” when no contact was selected.
The date and time, the duration, and the number of turns.
An outcome, set by analysis of the transcript when the run ends.
The outcomes are Contained, Partially contained, Business Rule Transfer, Agent Escalation, Confusion Transfer, Early Abandon, Hangup, and Timeout. Filter the list to All or to a single outcome.
Expand a run to read the full transcript, with each turn labeled as the agent or the user. You can also play back the run’s audio to hear tone, pacing, and pronunciation. A run that ends in an error keeps whatever was captured before it stopped, including partial audio and a partial transcript.
Click Run again to repeat the test with the same contact using the agent’s current configuration. Each repeat creates a new run, so you can compare attempts before and after a change.
Saved runs are visible to every user with access to the agent. Test runs are marked as tests, so they do not appear in customer-facing call history, analytics, or billing.
[VERIFY: QA shows an “Error” outcome that is not in the list above. Confirm the Test modal and History tab labels.]
Previewing the voicemail message
For outbound agents, you can hear the voicemail message the agent leaves without placing a call.
Open the Test modal and select the Test Audio tab.
Click Test voicemail.
The preview plays in the agent’s current voice using its voicemail setting, including the message text and any pronunciation or speech settings. To change the result, edit the voicemail message or the voice and click Test voicemail again. The next preview reflects the change without reloading the page.
The preview places no call and uses no minutes. It does not create a call record or a test run, so you can preview as often as you need.
The Test voicemail button depends on the agent’s setup:
On inbound agents, the button does not appear.
If voicemail is off or no message is set, the button is unavailable. Its tooltip reads “Configure your voicemail in Advanced settings.”
When voicemail is on and a message is set, the button is available.
To set up voicemail, see [link: Call settings and advanced call handling].
[VERIFY: Confirm whether an unsaved edit is picked up, and what plays when voicemail is set to Prompt.]
Agent version history
Each time you save a change to the agent’s instructions or configuration, a version is recorded with what changed and when. Use version history to read earlier versions and to see which version each test call ran against, so you can tell whether a change in behavior came from an edit.
To browse versions, open History for the agent and open the versions list. Versions are listed newest first, and the version in use is marked. Selecting a version shows its configuration at the time it was saved, including the AI model, voice, transcription settings, language, welcome and goodbye messages, and instructions.
Versions are read-only. Opening an earlier version does not change the agent, and a version cannot be restored. To reuse earlier wording, copy it from the older version and paste it into the editor.
Filtering test calls by version
Each past test call names the version it ran against. Select a version to show only test calls from that version. Clear the filter to see all test calls again.
The version filter works with the outcome filter. When both are set, the list shows only test calls that match both. If none match, the list says so.
[VERIFY: Confirm where History and the versions list sit in the UI, and their labels.]
Reviewing call latency
Each voice agent call records how long the agent took to respond, so you can find slow responses and see what caused them. Open a call from call history to see its Latency section.
The section has one row for each stage of the response. Only stages used on that call appear. For example, the Knowledge Base row appears only if the agent searched its knowledge base.
Row | What it measures |
E2E (end-to-end) | The time from when the caller stops talking to when the agent starts talking. This is the delay the caller hears, so start here. |
Transcription | The time to convert the caller’s speech to text. |
LLM | The time the AI model takes to produce the first part of its reply. |
TTS (text-to-speech) | The time to turn the reply into audio. |
Knowledge Base | The time to search the agent’s knowledge base. |
Speech-to-speech | The response time of a speech-to-speech (realtime) model, when the agent uses one. |
Each row is measured on every turn of the conversation and shows:
P50: the median response time.
P90 and P99: the slower responses. P90 means 90% of turns were this fast or faster.
Min and Max: the fastest and slowest turns.
Count: the number of turns measured.
A high E2E value means the caller waited. The other rows show why, for example a slow AI model or a slow knowledge base search.
[VERIFY: BLO-30 has no release version and was built in the standalone dashboard. Confirm the Latency section appears in the Aloware call detail view, and its row labels.]
Reviewing agent activity
The Agent Activity dashboard reports how your voice agents perform. Use the Voicemail filter to separate calls where the agent reached voicemail from the rest:
Yes shows calls where the agent reached voicemail and acted on it.
No shows all other calls. This is not the same as “a person answered.” Calls that reached an automated phone menu, or a machine the agent greeted without acting on, also appear here.
How the filter works:
It applies to every chart and summary card, including Total Calls, Average Duration, and Total Cost.
It works with your other dashboard filters.
Yes and No together make up all calls.
It stays applied when you move between dashboard panels or refresh.
Clearing it returns every view to all calls.
Call history has the same Voicemail filter. Filtering Agent Activity and call history the same way for the same period returns matching results.
For example, if an agent’s average call duration looks low, set Voicemail to Yes to see how many calls ended at voicemail, then switch to No. If the average rises, the low figure came from voicemail volume, not short conversations.
[VERIFY: The agent-activity-tab setting is off by default. Confirm which accounts see Agent Activity and how to describe its availability.]