Answering-machine detection is one of those features that looks like free money on the demo and turns into a fistfight once you actually run it. The pitch is clean: your dialer listens to the first second or two of every answered call, decides whether it's a person or a voicemail greeting, and only patches a human through to your agent. Everything the machine catches gets a voicemail drop instead of your rep's time.
On paper that reclaims a huge slice of your day. In practice, the same algorithm that's saving you 40 seconds a dial is occasionally hanging up on a live buyer who just said "hello" a little slowly — and that mistake has both a revenue cost and a regulatory one.
Full disclosure: I work for Ready, and our Power Dialer ships AMD plus voicemail drop. So I have skin in this. But the honest version of this feature includes the part where it costs you calls, and if I skip that you'll find out the hard way three weeks in.
How answering-machine detection actually works
AMD isn't magic. When a call connects, the dialer analyzes the audio pattern of the answer. A live human says something short — "Hello?" — then goes quiet, waiting for you to talk. A voicemail greeting is longer, more continuous, and often ends with a beep. The detector listens for those signatures: length of the initial utterance, pauses, the presence of a tone.
That analysis takes time. Most detectors need somewhere between 0.5 and 3 seconds of audio before they'll commit to a verdict. Which produces the two problems that define the whole tradeoff:
- The dead-air problem. While the detector is "listening," your live human is sitting there hearing silence. The longer you let AMD analyze, the more accurate it gets — and the more real people hang up because nobody's talking.
- The guess problem. Shorten that window to reduce dead air and accuracy drops. Now you're misclassifying more people as machines.
You are tuning between two failure modes, and you cannot zero out both at once.
The 40% that's real
Start with why anyone turns this on. On a cold or semi-warm list, a large chunk of answered calls go to voicemail. Depending on list quality and time of day, machine-answer rates of 35–45% are common — and yes, those are rough industry ballparks, not a number I measured on your list specifically.
If 40% of your answered calls are voicemail, AMD lets your agent skip all of them. Instead of your rep listening to a greeting, waiting for the beep, and reciting the same message 200 times a day, the dialer fires a pre-recorded voicemail drop and moves to the next number. Your agent only ever talks to people.
Worked example. Say a rep makes 300 dials a day, 30% connect to something (90 answered calls), and 40% of those are machines (36 voicemails). Without AMD, the rep burns roughly 25–35 seconds per voicemail leaving the message — call it 30 seconds. That's 36 × 30 = 1,080 seconds, about 18 minutes a day, gone. Across a 5-person team that's an hour and a half of paid agent time reclaimed daily. Nobody's arguing that isn't valuable.
The false-positive tax
Here's the part the demo skips. Detectors misfire in both directions, but the expensive direction is when AMD flags a live person as a machine and either drops them or shunts them to voicemail.
Depending on the detector and how aggressively it's tuned, false-positive rates on live humans land somewhere around 8–12% — roughly 1 in 8 to 1 in 12 real people. I wrote a longer breakdown of that number in AMD Hangs Up on 1 in 8 Live Humans, but the short version: those aren't random people. They're people who answered. Answering is the hardest, most valuable thing a lead does on your list. And you're hanging up on some of them.
Run the math against the same 90 answered calls. Say 54 are live humans. A 10% false-positive rate means ~5 live people per rep per day get treated as machines. If your booking rate on a live conversation is, say, 8%, that's 5 × 0.08 = 0.4 booked appointments lost per rep per day. Across a 5-person team over 20 working days, that's ~40 lost bookings a month. Now weigh that against the 18 minutes of saved voicemail time. Which number matters more depends entirely on what a booking is worth to you — and that's the honest answer, not a slogan.
The FCC abandonment problem AMD creates
This is the part that turns a productivity feature into a compliance one. Under FCC telemarketing rules, an abandoned call — one where a live person answers and no agent connects within two seconds — counts against a 3% abandonment cap, measured per campaign over a 30-day period.
AMD false positives feed straight into that number. Every time the detector wrongly decides a live human is a machine and either drops the call or plays a voicemail-drop message at them, that's a person who answered and got no rep. If you're pushing an aggressive line ratio on top of AMD, the two multiply. We covered the cap itself in The 3% Abandonment Rule and how line ratios compound it in Agent Count vs Line Ratio.
The regulators don't care that your detector made an honest mistake. Abandonment is abandonment. So AMD isn't just costing you deals when it misfires — it's spending down a compliance budget you have a legal cap on.
When to leave it on, when to turn it off
There's no universal answer, but there's a decision framework:
| Situation | AMD setting | Why |
|---|---|---|
| Cold list, high machine-answer rate, low deal value per call | On | Time savings dominate; a lost live call is cheap |
| Warm/inbound-nurture list, high deal value | Off or conservative | One live human is worth far more than saved seconds |
| Small dial volume, few reps | Off | Not enough voicemails to justify the false-positive risk |
| High volume, appointment-setting, mid deal value | On, tuned conservative | Longer detection window trades a little dead air for fewer false positives |
| Anything where you're already near the 3% cap | Off | Stop feeding false positives into your abandonment number |
The general rule: the higher the value of a single live conversation, the less you should trust AMD to guess. A wholesaler working a cold county list can afford to lose an occasional answer to save an hour of dial time. A team calling warm, high-intent leads where each booking is worth four figures should be very reluctant to let an algorithm decide who's real.
If you keep it on, tune the detection window toward accuracy and accept a little more dead air — a real person who hears two seconds of silence usually stays on the line; a real person who gets hung up on is gone for good.
What Ready's Power Dialer gives you to manage the tradeoff
The Ready Power Dialer ships AMD alongside voicemail drop, so when detection does correctly catch a machine, your rep never touches it — the drop fires and the queue advances. Call recording lets you audit why a call got classified the way it did, which is how you actually find your false-positive rate instead of guessing at it. Manager monitoring, transfer, barge, and whisper are there for when a live call does connect and you need a closer on the line fast — a separate failure mode we get into in The Transfer-Drop Rate That Kills Booked Revenue.
Plans, for reference:
- Free — $0/mo, 1 agent, 1 number, 500 minutes included, then $0.06/min.
- Pro — $29/agent/mo, up to 3 agents, $0.05/min.
- Team — $69/agent/mo, unlimited agents, $0.0375/min, plus speed-to-lead auto-dial and lead routing.
Minutes bill in 6-second increments, so the dead-air seconds AMD spends analyzing aren't free — another small reason not to run the detection window longer than you need. Full numbers live on the pricing page.
The practical takeaway
AMD is a genuine time-saver and a genuine liability, and which one dominates depends on numbers only you have: your machine-answer rate, your false-positive rate, and what a single live conversation is worth. Before you flip it on across a whole team:
- Pull a week of call recordings and count how often it hung up on a real person.
- Multiply that by your booking rate and your average deal value — that's your AMD cost.
- Weigh it against saved voicemail time, and check whether the false positives are pushing you toward the 3% abandonment cap.
If the math favors it, tune conservative and keep auditing. If your leads are warm and valuable, you might make more money leaving it off and letting your reps hear a few more greetings. Either way, decide with your own numbers, not the demo's.
Want to test it against your real list? Start on the Free Power Dialer tier — 500 minutes, no card — and pull your own false-positive rate from the recordings before you commit a team to it.