VICIdial Voicemail Drops: The Setup Nobody Explains Properly

#voip #asterisk #sysadmin #devops

Voicemail drop is one of VICIdial's most powerful features and also one of the most misunderstood. Done right, it turns answering machine contacts into marketing touchpoints without consuming agent time. Done wrong, it violates FCC rules and can generate fines that dwarf your campaign budget.

If you're running 25+ agents and not using voicemail drops, you're leaving callbacks on the table. Here's how it actually works.

The Basic Flow

Dialer places the call
Answering Machine Detection (AMD) analyzes the first few seconds of audio
If AMD says MACHINE — a pre-recorded message plays, then hangup. No agent involved.
If AMD says HUMAN — call routes to an available agent as normal.

The entire thing hinges on AMD accuracy. A false positive (human classified as machine) means a live person hears your voicemail and the agent never connects — that's a lost sale. A false negative (machine classified as human) means an agent wastes 15-30 seconds listening to someone's voicemail greeting before manually dispositioning.

Audio File Setup

Your voicemail drop message needs to hit four marks in 20-30 seconds:

Identify the caller (FCC requires this)
State the purpose briefly
Provide a callback number or website
Sound natural, not rushed or robotic

Record as WAV (8kHz, 16-bit, mono):

sox input_message.mp3 -r 8000 -c 1 -b 16 vm_drop_insurance.wav
cp vm_drop_insurance.wav /var/lib/asterisk/sounds/

Each campaign should have its own message. A solar voicemail shouldn't reference insurance.

Dialplan Configuration

The voicemail drop plays through a custom Asterisk dialplan context:

[vm-drop-salescamp]
exten => s,1,Answer()
exten => s,n,Wait(1)
exten => s,n,Playback(vm_drop_insurance_medicare)
exten => s,n,Wait(1)
exten => s,n,Hangup()

The Wait(1) before playback gives the voicemail system time to start recording after its beep. Without it, the first second of your message gets cut off.

Then enable AMD and voicemail drop on the campaign through VICIdial's admin interface: set AMD Send To VMX to Y, AMD Type to AMD, and select your audio file as the Answering Machine Message.

AMD Tuning for Voicemail Drops

The default Asterisk AMD settings in amd.conf work reasonably for standard American voicemail but aren't perfect out of the box.

To reduce false positives (the more dangerous error — you lose live connections):

greeting = 2000
maximum_number_of_words = 5
total_analysis_time = 5500

To reduce false negatives (agents connected to voicemail greetings):

greeting = 1200
maximum_number_of_words = 2
initial_silence = 2000

Target: 85%+ accuracy for machine detection with a false positive rate under 5%.

The Legal Reality

The FCC has ruled that leaving a prerecorded voicemail is a "call" under the TCPA. That means:

Prior express written consent required for prerecorded telemarketing messages to cell phones
National Do Not Call Registry applies
The message must identify the caller and provide a callback number
Safe harbor for dropped live calls is separate from voicemail drop

Before enabling voicemail drops: verify consent exists for all leads, scrub against DNC registries, confirm your message includes proper identification, and configure safe harbor separately for live answer drops.

Message Rotation

Leaving the same voicemail three times sounds robotic and tanks callback rates. Use dialplan logic to rotate between messages:

[vm-drop-rotation]
exten => s,1,Answer()
exten => s,n,Wait(1)
exten => s,n,Set(MSG=${RAND(1,3)})
exten => s,n,GotoIf($["${MSG}" = "1"]?msg1)
exten => s,n,GotoIf($["${MSG}" = "2"]?msg2)
exten => s,n,Goto(msg3)
exten => s,n(msg1),Playback(vm_drop_v1)
exten => s,n,Goto(done)
exten => s,n(msg2),Playback(vm_drop_v2)
exten => s,n,Goto(done)
exten => s,n(msg3),Playback(vm_drop_v3)
exten => s,n,Goto(done)
exten => s,n(done),Wait(1)
exten => s,n,Hangup()