Why test calls matter
Most production issues show up before scale if ops runs a realistic test call:- missing variables
- wrong voice/language
- poor opening message
- tool misuse
- transfer failure
- bad post-call mapping
- recording playback issues
Minimum test matrix
Run at least:| Test | Expected outcome |
|---|---|
| Normal success | Bot completes objective, ends cleanly, analysis is correct. |
| Customer silent | Re-engagement happens, then RNR/no-customer-message disposition. |
| Customer hangs up | disconnected_by=customer; result stored. |
| Bot ends call | disconnected_by=bot; result stored. |
| Tool path | Tool executes and event telemetry appears. |
| Transfer path | Bot transfers; result.was_transferred=yes and transfer-state fields populate. On failure, result.transfer_failed_reason is set. |
| Callback request | Customer asks to be called later; analysis emits callback_requested + time text (requires the callback double-gate enabled). |
| CRM pre-fetch | Variables render in opening message/prompt. |
| CRM post-call push | CRM receives expected fields. |
| Recording playback | Call detail recording works. |
Communication quality checks
During test calls, listen for behavior that will hurt production conversations:| Check | Expected behavior |
|---|---|
| Opening message | Customer noise should not cut off the first bot turn. |
| Short acknowledgements | Filler like “haan”, “hmm”, “okay”, or “ji” should not derail the bot. |
| Real interruption | Important phrases such as “wrong number”, “already paid”, “agent”, or “settlement” should interrupt correctly. |
| API/tool wait | Bot should avoid long dead air if a tool call takes time. |
| Silence | Re-engagement should sound natural before RNR. |
| Knowledge answer | Bot should use attached knowledge only when appropriate and avoid guessing. |
| Latency | STT, LLM, TTS, tool, or RAG latency should be visible in call events/metrics. |
- transfer requested
- call enters queue
- agent accepts
- hold audio stops
- wrap-up disposition is required
- agent call record links to bot call
Inspect after each test
Open the call detail and check:- transcript
- recording
disconnected_by- call duration
- events
- tool calls
- knowledge search events
- post-call analysis
- QC output
- CRM push/integration logs
Common failures
| Symptom | Likely cause |
|---|---|
Bot says {{FIELD}} aloud | Missing or misspelled variable. |
| No recording | Storage config, upload failure, or call ended before recorder initialized. |
| Post-call analysis missing | Auto disposition skipped analysis or prompt is empty. |
| Transfer tool absent | Transfer/Agent Desk not enabled for the bot/transport. |
| CRM push missing field | Mapper uses wrong variable path. |
| Test call fails before bot speaks | Carrier, LiveKit, config fetch, provider key, or active-hours issue. |