I have a specific edge case question about voicemail behavior:
Scenario: My agent is configured to leave a voicemail message (not hang up). The call reaches an answering machine, the agent starts leaving the message, and while the agent is actively speaking the voicemail, the human picks up the phone.
Question: What happens in this case? Does the agent:
Immediately stop the voicemail message and revert to normal conversation flow?
Continue the voicemail message to completion?
Something else?
I checked the docs on Handle Voicemail and the blog post on interruption handling, but I couldn’t find an explicit answer for this specific mid-voicemail pickup scenario.
There is no call ID. I have my agent set to leave a voicemail upon no human answer. I am just trying to figure out if someone picked up while the voice agent was mid voicemail if it would revert back to conversation flow, carry on with voicemail, or something else?
Good edge case, and the docs genuinely don’t cover it. The Handle Voicemail page explains hang-up vs leave-a-message, but nothing about a human picking up mid-message. So this is my read, not documented.
What I’d expect: detection runs the whole first 3 minutes, not just once, and once the agent is talking, a human speaking should register as a normal interruption and pull it back into conversation. So probably it stops and switches rather than finishing the script. I can’t promise that though, and the exact timing isn’t written down anywhere.
If it matters, just test it. Agent in leave-a-message mode, call a phone, let it start, pick up and talk, then check the transcript and disconnection reason. Worth pinging Retell support too, they’ll know the intended behaviour for an edge case like this.
Great question and not something I have run into because most of my clients do not use voicemail machines that you can pickup! I assume that once the voicemail is detected that the agent stays in the voicemail mode and then will hang up. Have you tested this, as this would be great to learn from!
Hey @lukascarlton When the call hits an answering machine and your agent starts speaking the voicemail (because Voicemail Response is set to Leave a message if reaching voicemail, on either the Prompt or Static Sentence tab), that message is dispatched through the same TTS turn pipeline as any other agent response. It is not a locked or non-interruptible playback, so standard barge-in rules apply.
If a human picks up while the agent is mid-voicemail, two things can happen depending on whether they speak loudly enough to trigger barge-in, which is governed by the agent’s Interruption Sensitivity setting (slider from “blocks interruption” at 0 to “stops immediately” at 1, default 1):
If the human speaks above the interruption threshold, the voicemail TTS is aborted immediately, the internal voicemail state is cleared, and the agent re-enters normal conversation flow. The LLM continues from the full transcript so far, so it sees the partial voicemail it already spoke plus whatever the human just said, and it responds in context.
If the human stays silent (for example, they picked up but didn’t say anything), the agent keeps playing the voicemail to completion and then ends the call as a voicemail drop, because nothing crossed the interruption threshold.
Notes:
This interruptible behavior is specific to Leave a message if reaching voicemail. The other two voicemail actions, namely Hang up if reaching voicemail and Treat as Successful Transfer and Bridge Call, short-circuit the call immediately on voicemail detection and do not go through the interruptible TTS path.
If you find that real humans who pick up mid-message are not being detected, raising Interruption Sensitivity (closer to 1) makes the agent more responsive to a quiet “hello”; lowering it reduces false barge-ins from answering-machine beeps or background noise. The default of 1 is the most aggressive barge-in setting.