Stock Asterisk AMD is free and built in. AMDY is an acoustic AI classifier you add in five minutes. Here's exactly where each wins — no spin.
First, the honest part: stock AMD is not useless
Let's start by giving default AMD its due, because pretending it has no merit would be dishonest and you'd stop reading. Stock Asterisk AMD — the app_amd application configured in amd.conf — has three real advantages. It is free: no line item, no contract, no per-call charge. It is built in: it ships with Asterisk and Vicidial, so there is no extra dependency to install, secure, or keep running. And for some campaigns it is genuinely good enough.
If you run low-stakes, low-volume outbound — a list where a missed connect costs you nothing, where you are not racing an abandon-rate ceiling, and where caller-ID reputation isn't on the line — the heuristic does the job. It listens for the timing pattern of a long, uninterrupted greeting (the hallmark of voicemail) versus a short utterance followed by silence (the hallmark of a person), sets AMDSTATUS to HUMAN, MACHINE, or NOTSURE, and hands that flag to your dialer. For a lot of operators that has been the whole story for fifteen years.
The case for AMDY isn't that stock AMD is broken. It's that "good enough" has a cost, and that cost is invisible until you measure it. Here is the full picture, line by line.
The head-to-head
| STOCK ASTERISK AMD | AMDY | |
|---|---|---|
| Method | Timing & silence heuristic | AI on the acoustic signature |
| Accuracy | 70–85% | 99% |
| Speed | Waits out the greeting window | <200 ms |
| Fast / silent answers | Misread | Handled |
| Carrier false-answers | Missed | Classified |
| Honeypot / spam-traps | Invisible | Detected |
| Output buckets | Human / Machine / NotSure | Human, voicemail, FAS, honeypot, fax, silence |
| Tuning | Constant re-tuning | None |
| Analytics | A flag | 23 reports + per-detection log |
| Carrier lock-in | n/a | Telco-agnostic — keep your carrier |
| Cost | Free, but hidden costs | $0.00010–$0.00025 per detection; 50K/mo free |
A table flattens nuance, so the rest of this article expands the rows that actually move money: accuracy and dropped humans, carrier false-answers, and the hidden cost of "free."
Accuracy and the live humans you never hear from
The single biggest difference is the rows that read 70–85% versus 99%. Stock AMD is a heuristic: it infers a machine from timing and silence, not from what the audio actually is. That works until a real person breaks the pattern — and people break it constantly. Someone who answers "Hello? Who's this?" with no pause looks, on a stopwatch, exactly like the front of a voicemail greeting. Someone who picks up and says nothing for a beat trips the silence logic. The result is that heuristic AMD drops roughly 5–15% of live humans as machines.
That number is the quiet one, because a dropped human never shows up in a report as an error. The call just ends. You don't hear the prospect who was about to say yes; you see a slightly lower connect rate and assume the list was tired. AMDY classifies the acoustic signature of the answer audio — the sound of how the line was answered, not a stopwatch on the greeting — so a fast greeting or a silent pickup is just another shape it recognizes. It reaches a decision in under 200 ms, which is why it doesn't need to "wait out" the greeting window the way the heuristic does.
Carrier false-answers: the row stock AMD can't see
Across roughly 2.3 billion answered outbound calls a month, only about 12.5% are a live human. About 73% are machines, and about 14% are carrier false-answers (FAS) — the network signals an answer when no person ever picked up. FAS is the row where the two tools genuinely diverge in kind, not just degree.
Stock AMD has no concept of FAS. Its world has three buckets — HUMAN, MACHINE, NOTSURE — and a carrier false-answer doesn't fit any of them cleanly, so it usually gets passed to an agent as a "connect" that turns out to be dead air. At 14% of answered calls, that is a large, recurring tax on agent time that the heuristic literally cannot label. AMDY classifies FAS as its own bucket, alongside human, voicemail, honeypot/spam-trap, fax, and silence. It also detects honeypots and spam-traps — numbers planted to catch dialers — which are invisible to stock AMD and which, left unhandled, quietly burn your caller-ID reputation.
The hidden cost of "free"
"Free" is the strongest argument for stock AMD and the most misleading. The license cost is zero; the operational cost is not. Every dropped live human is a lead you paid to generate and then hung up on. Every FAS connect is paid agent time spent listening to silence. Every campaign that needs the heuristic re-tuned — nudging initial_silence, greeting, after_greeting_silence in amd.conf to chase a moving target — is engineering time and a window where the numbers are wrong. None of that appears on an invoice, which is exactly why it's easy to ignore.
AMDY has a visible price — a fraction of a cent per detection — precisely so the trade is honest. The model is monthly base plus included detections plus flat per-detection overage: Sandbox is $0 for 50,000 detections a month with no card; Starter is $79 for 500K then $0.00025 each; Growth is $299 for 5M then $0.00015; Scale is $999 for 25M then $0.00010. To put that in context, carrier and CPaaS AMD add-ons run roughly $0.004 to $0.0075 per call — ten to seventy times more, at lower accuracy. The honest comparison is not "free vs paid." It's "hidden cost vs a known fraction of a cent."
What you give up — and what you don't
The fair worry with adding any tool is what it changes. With AMDY the answer is: very little. You don't remove stock AMD — you disable app_amd for the campaign and let AMDY make the call. You keep your carrier and your dialer; it's telco-agnostic. Install is one bash command, about five minutes, native on Vicidial, Asterisk, FreeSWITCH, and Issabel. And because every detection is logged, you trade a single AMD flag for 23 analytics reports plus a queryable per-detection log — so for the first time the human-vs-machine numbers are something you can audit instead of assume.
So which should you run?
Here is the honest call. If you run a few low-volume campaigns, your margins aren't tight, and you genuinely don't care about a handful of dropped leads, stock AMD is fine — keep it, save the money, move on. But if dropped live prospects cost you real revenue, if abandon-rate math is squeezing you, or if caller-ID reputation matters to your operation, the gap between 70–85% and 99% is not a rounding error — it's the difference between connecting with the people you paid to reach and hanging up on them. And you don't have to take any of this on faith: you can test AMDY free on your own traffic, against your own lists, and read the numbers yourself.
Originally published at https://app.amdy.io/blog/default-amd-vs-amdy
Top comments (0)