VM Hunter vs AMDY: how to run a fair AMD test
If you are choosing between VM Hunter and AMDY (amdy.io) for VICIdial, the honest answer is to test both on your own calls. Here is what VM Hunter does, what to look for, and a test protocol that gives you a real number in a day.
Published accuracy figures, ours included, come from the vendor's own traffic and scoring method, so they do not transfer to your list, your carrier or your scripts. What transfers is a side-by-side run on the same calls. This page keeps to facts about VM Hunter and to a method you can apply to any detector.
What VM Hunter does
| VM Hunter | |
|---|---|
| Decides from | A transcript of the first two seconds, plus tone and silence analysis; listens up to three seconds if the callee is still talking |
| Time to verdict | About 2 seconds after answer on a normal call |
| Verdicts | HUMAN, MACHINE, with a reason in AMDCAUSE: voicemail wording, beep, static, SIT tone (disconnected), silence, no audio, call screening |
| Call screening | iPhone, Google and call-blocker prompts labelled separately; can be routed to an intro and an agent |
| Evidence | Transcript, reason and audio sample for every call in the portal; CSV export |
| Integrations | VICIdial and Asterisk (EAGI), FreeSWITCH (Lua), Twilio (bridge, beta), WebSocket API; clients open source under MIT |
| Pricing | Free plan with 5,000 calls a month; paid from $100 a month; Starter is 3M calls at 50 CPS and 550 channels |
What to look for in any AI AMD
- How short answers are handled. "Hello?" followed by a pause is the most common live answer and the one timing-based detection gets wrong. Ask what the detector does when the transcript is one word.
- What happens at the cutoff. A greeting that is still running at two seconds is usually a recording. Does the detector keep listening, or decide on the first two seconds alone?
- Call screening and carrier prompts. On some lists these are a larger share than live answers. Are they labelled, or lumped in with voicemail?
- Calls with no audio. Trunks drop media on a surprising share of calls. A detector that labels these separately shows you a carrier problem; one that does not hides it as silence.
- Evidence. Can you hear what the detector heard for any verdict you doubt?
- Fail-safe. What does the client do when the service is unreachable, and how long does it wait?
A one-day test protocol
- Create two campaigns from the same list, split at random. Point one at each detector's routing extension. Same agents, same hours, same script.
- Keep recordings on for both.
- Next morning, pull 200 calls from each at random, listen to the first three seconds, and mark what was actually there: person, voicemail, carrier prompt, screening, silence.
- Score each detector on those 200: people dropped, voicemails connected, screening handled. Then look at agent talk time and abandon rate for the day as the business check.
- Repeat on a different list type if you run several; results vary more by list than by detector.
For the VM Hunter side, the call logs already hold the transcript and audio for every call, filterable by cause, which makes the listening pass quick. Setup is in the VICIdial guide. If you share the results, we will tell you plainly where we stand and what we can close.
Questions
Is VM Hunter an AMDY alternative?
Can I run both on the same VICIdial?
How should I score the test?
What does switching cost?
Run the test
The free plan's 5,000 calls cover a one-day side-by-side on most lists. No card needed.