Voicemail Detection

On outbound calls, ~40% of calls go to voicemail. Without detection, your agent talks to an answering machine for 60+ seconds before noticing — burning credits. With voicemail detection enabled, the bot classifies the first few seconds of audio and either:

• Hangs up quickly, OR
• Leaves a pre-configured voicemail message

When to Use

Example 1: Silent Hang-Up on Voicemail

Goal: your agent makes outbound lead-qualification calls. If it reaches voicemail, hang up silently (don't leave a message — you'll retry later).

Step 1 — Enable in advanced settings

Step 2 — What happens on a voicemail call

[Call connects]
[Voicemail plays: "Hi, you've reached Priya. I can't come to the phone right now..."]
[Bot waits 2 seconds, GPT-4o-mini classifies audio as "voicemail"]
[Bot silently hangs up — call ends in ~3 seconds, costs minimal credits]

Step 3 — What happens on a human call

[Call connects]
[Human says: "Hello?"]
[Bot waits 2 seconds, GPT-4o-mini classifies audio as "conversation"]
[Bot proceeds normally with its welcome message]

Example 2: Leave a Voicemail Message

Goal: collections reminder — if you reach voicemail, leave a polite note asking them to call back.

Settings

What the voicemail hears

Voicemail beep
Bot: "Hi, this is an automated reminder from BlueBank. Your monthly EMI payment of ₹5,000 was due on the 15th. Please call us back at 1-800-123-4567 or visit our app to pay. Thank you."
[Bot hangs up after speaking]

Tips for voicemail messages

• Keep under 30 seconds — most voicemails cut off at 30-60s
• Include a callback number — voicemail recordings don't support tools
• Say the company name first — caller might skip otherwise
• No markdown — TTS reads **bold** as "asterisk bold asterisk"

Example 3: Appointment Reminders

Goal: confirm dental appointment 24 hours in advance. Try to get a live confirmation, but if voicemail, leave the reminder.

Settings

Outcomes

How It Works

1. On call connect, the bot listens for 1-3 seconds of audio
2. Audio is transcribed by Deepgram STT
3. The transcribed text is sent to gpt-4o-mini with a classifier prompt
4. Classifier returns "voicemail" or "conversation"
5. If "voicemail":
   • Plays the configured message (if set)
   • Waits for the message to finish speaking (max 15s)
   • Sends SIP BYE to end the call
6. If "conversation":
   • Proceeds with the agent's normal welcome message + pipeline

Classification typically adds 1-2 seconds of silence at the start of human calls. Worth it for the 40%+ credit savings on voicemail calls.

Cost & Credit Savings

Per-call cost comparison (typical 90-second TTS-heavy call):

At 1000 outbound calls/day with 40% voicemail rate:

• Without: 1000 × (0.4 × ₹8 + 0.6 × ₹12) = ₹10,400/day
• With: 1000 × (0.4 × ₹0.50 + 0.6 × ₹12.50) = ₹7,700/day
• Savings: ₹2,700/day = ~₹81k/month

Prerequisites

• OpenAI API key must be configured (used for GPT-4o-mini classification, separate from your agent's LLM)
• Outbound phone channel — voicemail detection only runs when channel == "phone"

Troubleshooting

Configuration Reference

Only applies when the call channel is phone and outbound.