Answering machine detection is one of those dialer features that sounds like free money. Flip it on, and the machine listens to the first second or two of every answered call, decides "human" or "voicemail," and only hands your rep the live humans. Everything else gets a voicemail drop and moves on. No more sitting through four rings and a "you've reached the Hendersons" recording. Your rep talks to people, the software eats the dead air.
That's the pitch, and it's real. But the same feature that saves your reps 30-40 seconds a dial is also quietly hanging up on live buyers it misclassified — and depending on how your dialer is tuned, it might be creating abandoned-call exposure under the TCPA. I want to walk through the actual tradeoff with numbers, because "turn on AMD" is not a free decision.
Full disclosure: I work for Ready. We ship a Power Dialer with voicemail drop and speed-to-lead auto-dial, so I have a horse in the "should you dial" race. I don't have one in the "which AMD setting is right" race — that answer depends entirely on your list and your margins, and I'll show you how to find it.
What AMD actually buys you: the 40 seconds
Here's the time budget on a single answered call without AMD. Your rep hears ringing, the line picks up, and now they're listening — is this a person saying hello, or a voicemail greeting warming up? They wait. If it's voicemail, they wait for the beep, drop or leave a message, and hang up. That's easily 25-45 seconds of a rep's attention spent confirming they're talking to an answering machine.
AMD offloads that. The detector makes the human-or-machine call in roughly 2-4 seconds of audio, and if it's a machine, it fires your voicemail drop automatically while the rep is already on the next dial. So the saved time isn't hypothetical — it's the difference between a rep personally screening every pickup and a rep only ever hearing live humans.
Rough math on a list where 30% of answered calls hit voicemail: if a rep makes 120 dials in an hour and 40 of them answer, ~12 of those are machines. Save 35 seconds on each and that's about 7 minutes an hour handed back to the rep. Over a full day that's real — close to an extra 45 minutes of live talk time.
The false-positive tax nobody prices in
Here's the part the feature page skips. AMD is a probabilistic guess made on 2-4 seconds of audio, and it's wrong in both directions.
- False positive (the expensive one): a live human answers, says "hello" a beat too slowly or with a pause, and the detector tags them as voicemail. Your rep never hears them. They get a voicemail drop instead of a conversation — a buyer who picked up the phone, dropped into your funnel as a callback-maybe.
- False negative (the annoying one): a voicemail greeting gets tagged as human and routed to a rep, who wastes 10 seconds realizing they're talking to a recording.
False negatives cost you seconds. False positives cost you the whole call. And on a warm or inbound-generated list, that live pickup you just misclassified as voicemail was your best contact of the hour.
Detector accuracy varies, but treat it honestly as a lever, not a fact. Aggressive AMD settings — bias toward "machine," decide fast — save the most rep time and misfire on the most live humans. Conservative settings — bias toward "human," listen longer — protect live connects but hand more machines to reps and shave the detection window down to almost nothing.
Working the cost of a dropped live call
Say your live-answer-to-conversation is worth something you can name. Pull it from your own numbers, but as an illustration: if 1 in 20 live conversations books a meeting, and a booked meeting is worth $200 in expected pipeline, then a single misrouted live pickup is worth roughly $10 in expected value the moment AMD drops it to voicemail.
Now run it at scale. A rep makes 800 dials a day, 40% answer = 320 answered calls. If aggressive AMD misclassifies even 3% of live humans as voicemail, and 70% of your answers are live, that's roughly:
320 × 0.70 × 0.03 ≈ 7 live buyers a day silently voicemail-dropped, per rep.
At $10 expected value each, that's ~$70/day/rep of pipeline you're trading for the 45 minutes of saved talk time. Whether that's a good trade depends on whether your reps convert saved time into more booked meetings than the dropped calls would have. On a high-value, low-volume list, it usually isn't. On a huge cold list where connects are cheap and volume is everything, it usually is.
That's the whole decision, and it's list-specific. Run your own version:
| Setting | Rep time saved | Live calls misrouted | Best fit |
|---|---|---|---|
| Aggressive AMD | Highest | Highest | Large cold lists, low per-connect value, throughput-driven |
| Conservative AMD | Moderate | Low | Warm/inbound lists, high per-connect value |
| AMD off | None | Zero | Small high-value lists, every live pickup matters |
The TCPA abandoned-call gray zone
There's a compliance layer here that has nothing to do with efficiency. If your dialer is placing more calls than you have reps to answer — predictive pacing — and AMD guesses "human," the system connects the call before it has a rep free. When no rep is available, the call gets dropped. That's an abandoned call, and the TCPA has specific rules about abandonment rates and the pre-recorded identification message you're supposed to play when it happens.
AMD interacts with this directly. A false negative (machine tagged as human) can trigger a connection attempt that then gets abandoned. Get the tuning and the pacing wrong together and you're generating abandoned calls at a rate that draws regulatory attention — and TCPA statutory damages run $500 to $1,500 per violation, the same exposure band that makes cold texting expensive.
Ready's Power Dialer leans toward power/queue dialing rather than aggressive predictive over-dialing, which structurally limits abandoned-call risk — a rep is pulling the next contact, not the system racing ahead of your staffing. That's the safer default. If you do run predictive with AMD on, you own the abandonment math, and you should be watching your abandoned-call rate the way you watch your connect rate.
Voicemail drop is where the saved time gets recovered — or wasted
The upside of AMD only pays off if the voicemail drops it fires actually do something. A drop that leaves the same generic "hey it's Jordan, call me back" every time gets ignored. We covered this in detail in the voicemail-drop callback post, but the short version: a differentiated second and third message meaningfully outperforms the same message on repeat. AMD hands you a lot of voicemail moments for free — waste them with a lazy drop and you've saved rep time to produce nothing.
Pair the drop with an auto-text. Ready's Power Dialer can fire an SMS on the same event, so a voicemail drop lands alongside a text the contact can actually reply to — which turns a one-way recording into a two-way conversation in your inbox. That's usually where the callback actually happens.
What I'd actually tell a team
- Start conservative, or off. Prove your list's value-per-connect before you let a detector make throwaway decisions about it. On a warm list, off is often the right answer.
- Segment by list value. Run aggressive AMD on the cold, high-volume queue and conservative-or-off on the warm inbound queue. It doesn't have to be one setting.
- Measure the false-positive drop, not just time saved. The time-saved number is easy and flattering. The dropped-live-connect number is the one that costs you. Sample your voicemail drops and check how many were actually live pickups.
- Watch your abandoned-call rate if you're pacing predictively. This is a compliance number, not a productivity one, and staffing is half of it — see the transfer-drop and staffing math.
- Make the voicemail drop and the follow-up text earn their keep. The saved time is worthless if the automated follow-up is generic.
On Ready's Power Dialer, minutes are billed in 6-second increments — Free includes 500 minutes/mo then $0.06/min, Pro is $29/agent/mo at $0.05/min, Team is $69/agent/mo at $0.0375/min with speed-to-lead and manager monitoring. Voicemail drop and auto-text come with the dialer, so you can test conservative AMD, measure the live-drop rate on your own list, and decide the trade with real numbers instead of a feature-page promise. Pricing details are on the product page, and you can start free — the 500 included minutes are enough to run a real sample before you commit to a setting.
AMD isn't good or bad. It's a dial you turn until the time you save is worth more than the live calls you drop — and that break-even lives in your own list, not in anyone's marketing copy.