Your Agents Talk to Humans. Not Voicemails.
ViciStack's neural AMD engine classifies calls in under 200ms with 98.3% accuracy. VICIdial's built-in AMD manages 60-70%. The difference is thousands of dollars per day in recovered agent productivity.
98.3%
AMD Accuracy
<200ms
Classification time
Week 1
Typical ROI
The Problem: VICIdial's AMD Is Destroying Your Productivity
VICIdial's built-in AMD (amd_type) uses a simple audio detection algorithm: it measures the duration of the initial speech burst after connection. If the speech is short (like "Hello?"), it's classified as human. If it's long (like "Hi, you've reached the Johnson residence..."), it's classified as a machine.
This approach has a 30-40% failure rate in practice. Here's what that costs you: with 50 agents running 30 dials/hour, and each AMD failure wasting 15-20 seconds of agent time, you're burning approximately 750-1,000 minutes of productive agent time every single hour. At $18/agent/hour loaded cost, that's $225-$300 per hour — $1,800-$2,400 per 8-hour shift — wasted on nothing but your agents listening to voicemail greetings.
The AMD failures also distort your adaptive_dl_level. VICIdial's adaptive algorithm can't distinguish between "no agent available" drops and AMD failures — so it throttles your dial rate to compensate for a problem it doesn't understand. Your entire campaign performance degrades in a cascade that starts with one bad AMD setting.
False positives (routing live humans to voicemail) are even worse. Every live human routed to your amd_agent_route as an "AM" status is a lead you burned. They heard silence. They hung up. They're done. They may file a TCPA complaint.
How ViciStack AMD Works
01
Audio capture
ViciStack captures the first 2 seconds of call audio via a lightweight Asterisk AGI hook. No changes to your VICIdial configuration required.
02
Neural classification
Our neural network — trained on 50M+ call recordings — analyzes acoustic patterns, speech rhythm, and background noise to classify the call in <200ms.
03
Routing decision
Clean calls route to your agents instantly. Confirmed machines route to your voicemail message or amd_send_to_vmx action. No agent time wasted.
04
Continuous learning
Agent dispositions feed back into the model. When agents mark AMD errors, the model retrains on your specific call data.
Technical Requirements
Related Modules
Start with AMD — See ROI in Week 1
The free audit includes your current AMD accuracy baseline so you know exactly what you're starting with.
Start My Free Audit →