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:
TestExpected outcome
Normal successBot completes objective, ends cleanly, analysis is correct.
Customer silentRe-engagement happens, then RNR/no-customer-message disposition.
Customer hangs updisconnected_by=customer; result stored.
Bot ends calldisconnected_by=bot; result stored.
Tool pathTool executes and event telemetry appears.
Transfer pathBot transfers; result.was_transferred=yes and transfer-state fields populate. On failure, result.transfer_failed_reason is set.
Callback requestCustomer asks to be called later; analysis emits callback_requested + time text (requires the callback double-gate enabled).
CRM pre-fetchVariables render in opening message/prompt.
CRM post-call pushCRM receives expected fields.
Recording playbackCall detail recording works.

Communication quality checks

During test calls, listen for behavior that will hurt production conversations:
CheckExpected behavior
Opening messageCustomer noise should not cut off the first bot turn.
Short acknowledgementsFiller like “haan”, “hmm”, “okay”, or “ji” should not derail the bot.
Real interruptionImportant phrases such as “wrong number”, “already paid”, “agent”, or “settlement” should interrupt correctly.
API/tool waitBot should avoid long dead air if a tool call takes time.
SilenceRe-engagement should sound natural before RNR.
Knowledge answerBot should use attached knowledge only when appropriate and avoid guessing.
LatencySTT, LLM, TTS, tool, or RAG latency should be visible in call events/metrics.
For Agent Desk bots, also test:
  • 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

SymptomLikely cause
Bot says {{FIELD}} aloudMissing or misspelled variable.
No recordingStorage config, upload failure, or call ended before recorder initialized.
Post-call analysis missingAuto disposition skipped analysis or prompt is empty.
Transfer tool absentTransfer/Agent Desk not enabled for the bot/transport.
CRM push missing fieldMapper uses wrong variable path.
Test call fails before bot speaksCarrier, LiveKit, config fetch, provider key, or active-hours issue.