Live people dropped
How many real answers were marked as machines and hung up on? These are leads you paid to reach.
AMDY COMPAREFour ways to detect
Four ways to detect
an answering machine.
Built-in timing rules, transcription, carrier-bundled detection and acoustic AI each make different trade-offs. Here’s how they line up for an outbound call center.
| APPROACH | Stock timing AMD | Transcript AMD | Carrier-bundled AMD | AMDY acoustic AI |
|---|---|---|---|---|
| How it decides | Silence and word timing | Speech-to-text, then keyword rules | The carrier’s detector, on its own calls | The sound of the answer, plus tone checks |
| Works with any carrier | Yes | Usually | No | Yes |
| Fits an existing ViciDial dialplan | Yes | Varies | Rarely | Yes |
| Typical time to decide | Often 2–5 s | Needs enough words to transcribe | Varies by product | About 125 ms to start |
| Hesitant “Hello? … Hello?” | Often marked machine | Usually handled | Varies | Handled |
| Recorded “Hello?” robocalls | No | Hard — the words are human | Varies | Recognised as recordings |
| Fallback if service is down | Built in | Varies | Tied to the carrier | Native AMD takes over |
| Cost model | Free | Per transcribed minute | Per call, on top of minutes | Per detection, free to start |
“Varies” means it depends on the specific product and configuration. Check each vendor’s current documentation.
HOW TO CHOOSEMeasure both errors.
Measure both errors.
On your own calls.
A single accuracy number can hide the mistake that hurts you most. Ask every vendor the same questions.
Machines sent to agents
How many greetings and carrier messages still reached the floor? That is wasted agent time.
Time to decide
Every second of analysis is a second of silence for the person who answered — and a reason to hang up.
READY FOR A BETTER FIRST HELLO?
Talk to AMDY Run a side-by-side on one campaign.
Keep your current AMD on the rest of the floor and compare outcomes with real labels.