Answering Machine Detection is the feature nobody notices working correctly and everybody blames when it fails. It sits between your dialer and your rep, listening to the first second or two of every answered call, deciding: human or voicemail? Route to a rep, or drop a voicemail and move on. When it guesses right, your reps only talk to live people and your list burns fast. When it guesses wrong, it hangs up on a human who just said "hello" — and you never even know it happened.
Full disclosure: I work for ReadySMS, and we ship a Power Dialer with AMD and voicemail drop built in. So I have a horse in this race. But the honest version of this story is that AMD is a tradeoff dial, not a free win, and cranking it toward "faster" quietly costs you conversations. Let me show you the math.
What AMD actually does in the first 1,500 milliseconds
When a call connects, the far end produces audio. AMD analyzes it: cadence, silence gaps, the length of the initial greeting, background tone. A human says "Hello?" and pauses, waiting for you. A voicemail plays a longer, uninterrupted greeting with a beep at the end. AMD tries to tell them apart before it commits the call to a rep.
The catch: it has to decide fast. If it waits three full seconds to be sure, the live human has already said "hello... hello?... anyone there?" and hung up in annoyance. If it decides in 800 milliseconds, it's guessing on almost no data. That's the whole game — accuracy versus latency, and you don't get both.
The false-positive problem nobody measures
There are two ways AMD fails:
- False negative: it thinks a voicemail is a human and routes it to a rep. Your rep gets a robot. Annoying, but visible — the rep knows immediately and hangs up. You can measure it.
- False positive: it thinks a human is a voicemail, drops a message (or dead-airs), and disconnects. The live person you paid to reach is gone. This one is invisible. It never shows up as a bad connect because it never became a connect at all.
Industry chatter puts AMD false-positive rates somewhere in the 5–10% range on tuned systems, worse on aggressive settings. Call it roughly 1 in 12 answered live calls getting misclassified as a machine and dropped — and that's a reasonable middle-of-the-road approximation, not a guarantee for your list.
Now sit with that. If you're dialing a warm inbound list where the person literally just filled out a form 90 seconds ago, an 8% false-positive rate means roughly 1 in 12 of your hottest prospects picks up, hears silence or a canned voicemail, and hangs up thinking your company is a robocaller. That's the exact opposite of a speed-to-lead advantage.
The cost per lost connect (worked math)
Let's make it real. Say you're a small sales team running the Power Dialer on our Team plan — $69/agent/mo, unlimited agents, $0.0375/min, dialing a list of 1,000 warm leads.
Assume a 35% answer rate → 350 live answers. Now apply an 8% AMD false-positive rate:
- 350 × 8% = 28 live humans dropped as false machines
If your team books a demo on roughly 1 in 5 live conversations, those 28 lost connections are about 5–6 booked demos you never got. Put a $2,000 average deal value and a 25% close rate against it, and 28 dropped connects is on the order of $2,500–$3,000 in expected pipeline — gone silently, on a single list, because a setting was tuned half a second too aggressive.
The dialer minutes to reach those 28 people cost you almost nothing. The 28 people did.
The latency-vs-accuracy dial, in plain terms
Most dialers expose AMD as a set of thresholds — initial silence, greeting length, decision timeout. Here's the practical version of what each direction buys you:
| Setting direction | What you gain | What you lose |
|---|---|---|
| Aggressive / fast (short timeout) | Reps waste less time on voicemails; list burns faster | More live humans misread as machines and dropped |
| Conservative / slow (long timeout) | Fewer live humans dropped | Reps eat more voicemail connects; the "hello... hello?" delay annoys real humans |
| AMD off entirely | Zero false-positives — every human reaches a rep | Reps hear every voicemail themselves and drop manually |
Notice the conservative setting isn't free either. If AMD waits 2.5 seconds before handing a live call to a rep, the human on the other end sits in dead air, and that delay causes hangups too. There's no setting that makes the tradeoff disappear. There's only the setting that's least wrong for your specific list.
When to turn AMD off completely
Here's the take most dialer vendors won't put in writing: on a warm, low-volume list, turn AMD off.
The whole point of AMD is to protect rep time when you're blasting through thousands of cold numbers and voicemails are frequent. But when you're running a speed-to-lead cadence — a lead came in, a rep is dialing within the first 60 seconds — the economics flip:
- Your answer rate is high because the prospect is expecting contact.
- Every single connection is worth real money.
- The list is small enough that a rep hearing an occasional voicemail costs you seconds, not hours.
In that world, an 8% false-positive rate is pure destruction and AMD's efficiency gain is negligible. Let the rep hear the voicemail and drop it manually with one click. You'll never misclassify a human again.
Rule of thumb:
- Warm / speed-to-lead / high-value list, low volume → AMD off, manual voicemail drop.
- Cold, high-volume list where reps would drown in voicemails → AMD on, tuned conservative, and accept that you're trading some live connects for rep throughput.
Voicemail drop is the safety net either way
Whether AMD is on or off, the thing that makes "hear the voicemail yourself" tolerable is a good voicemail drop. Our Power Dialer lets a rep hit one button to leave a pre-recorded message and immediately advance to the next call — so a manually detected voicemail costs a couple seconds, not the 25 it takes to leave a live message.
That changes the AMD calculus. If dropping a voicemail manually is nearly free, the efficiency argument for aggressive AMD gets a lot weaker. The reason teams crank AMD is usually that manual voicemail handling is slow in their tool — fix that, and you can afford to be conservative or off.
One more thing worth knowing: voicemail drop and ringless voicemail are not the same, and the legal difference matters. If you're leaning on drops, read the voicemail drop vs ringless line before you scale it. And the second-message copy drives your callback rate more than the first — worth tuning.
A practical setup for a mixed operation
If you run both cold prospecting and warm inbound, don't use one AMD setting across everything. Split it:
- Inbound / speed-to-lead queue — AMD off. Reps dial fast, talk to every human, drop voicemail manually when needed. Pair the dial with an instant text fallback so a genuine no-answer still gets a message in 90 seconds.
- Cold outbound queue — AMD on, conservative timeout. Measure your false-negative rate (reps reporting "got a voicemail") and nudge the timeout down only until reps stop complaining — never past it.
- Audit monthly. Pull a sample of AMD-dropped calls and listen. If you hear humans saying "hello" before the drop, your setting is too aggressive. This is the only way to catch false-positives, because they never show up in your metrics on their own.
The takeaway
AMD isn't a checkbox, it's a bet on how much rep time is worth versus how much a live connection is worth — and on a warm list, that bet almost always loses. The default aggressive settings most dialers ship with are tuned for cold-list throughput, not for the high-value speed-to-lead calls where every human matters.
Start by turning AMD off on your warmest queue, lean on fast voicemail drop to keep reps efficient, and audit your dropped calls once a month so the invisible losses stop being invisible. If you want to see how the Power Dialer's connect controls and voicemail drop are set up, the pricing and feature breakdown is here — no pitch, just the knobs and what they do.