Link Leads Blog · October 14, 2026
Answering machine detection decides, in under two seconds, whether a dialer hands a call to an agent or hangs it up — and on an aged list, a wrong call in either direction has a real cost. What AMD actually measures, the two ways it fails, and how its settings sit inside the same abandonment-rate rule that governs the rest of a power dialer setup.
Most dialers don't "listen for a voicemail" the way that sounds — they analyze the first second or two of audio after a call connects for patterns that correlate with a recorded greeting: a longer unbroken stretch of speech before the first pause, a flat energy pattern consistent with a loop, and, on some platforms, a beep at the tail end. A live human answering "Hello?" produces a short burst of speech followed by silence while they wait for a reply — a machine greeting runs on, often straight into a name and a request to leave a message. The dialer scores that pattern against a threshold and routes the call: connect to an agent, play a drop, or hang up, depending on how the campaign is configured.
The two failure modes aren't symmetric. A false positive — a live person misread as a machine — gets disconnected or routed to a voicemail drop while they're still standing there saying "hello, hello?" On an aged list where every connect is already scarce, that's a dial you paid for and a person who picked up, counted as a no-answer in your dispositions when it wasn't one. A false negative — a machine misread as a person — gets connected to a live agent, who now spends several seconds talking to a recording before realizing it and hanging up. That costs agent time, not list value, but it adds up across a shift the same way slow pacing does in the power dialer setup guide.
The Telemarketing Sales Rule caps abandoned calls — answered calls that aren't connected to a live agent within a set number of seconds — at 3% of answered calls measured over a 30-day calling campaign, a limit covered in full in the manual dialing vs. autodialer guide. AMD's own analysis window sits inside that clock: the one to two seconds a dialer spends deciding "human or machine" is dead air from the lead's side, and on a predictive dialer running multiple lines, that same window is also when a second connected call can go unanswered if every agent is already on another line. A detection window set too long doesn't just annoy a live answerer — it pushes more calls past the connection-time threshold that counts toward the 3% cap, on top of whatever abandonment the pacing ratio alone already produces.
Three settings do most of the work on platforms that expose AMD configuration at all:
Detection delay or analysis window. How long the dialer listens before deciding. Shorter cuts dead air and false positives from agents hanging up on real people, but raises false negatives — agents connected to machines more often. Most campaigns land somewhere in the 1–1.5 second range rather than running it at a platform's maximum.
Beep or tone detection. Where supported, waiting for the characteristic beep at the end of a voicemail greeting before deciding is more accurate than pattern analysis on the greeting alone, at the cost of a longer window — a real tradeoff against the point above, not a free upgrade.
Post-detection action. What happens once a call is classified as a machine: hang up silently, or play a prerecorded drop. A prerecorded voicemail drop played to a number the dialer has classified as a wireless line runs into the same prior-express-written-consent requirement covered in the ringless voicemail guide — an aged list without that consent on file should hang up silently on a detected machine, not auto-drop a message.
A call an agent spends five seconds on before realizing it's a machine shouldn't get buried in a generic "no answer" or "voicemail" bucket — tag it separately, the same way the disposition code guide recommends splitting "worked" into something more useful. A rising rate of agent-reported false negatives over a week is the signal to loosen the detection window; a rising rate of agents hearing a live "hello" cut off mid-sentence is the signal to tighten it. Without a code for it, both signals disappear into the same undifferentiated pile and nobody notices the setting needs attention until connect rate has already dropped for a reason nobody can name.
A fresh, high-intent list can absorb a slightly aggressive AMD setting because volume and urgency carry the campaign regardless. An aged list doesn't have that cushion — connect rate is already the scarce resource the whole campaign is built around, so a setting that silently drops five or ten real answers out of every hundred calls is a direct hit to the number that determines whether the batch was worth buying. Checking AMD settings belongs in the same first-session setup pass as pacing ratio and calling hours, not something tuned only after a problem shows up in the numbers.
The abandonment-rate cap and the consent requirement for prerecorded voicemail drops to wireless numbers come from the Telemarketing Sales Rule and the TCPA; some states layer their own telemarketing rules on top. This is a practical guide to how AMD settings work and interact with those rules, not legal advice — confirm your specific dialer configuration with counsel familiar with telemarketing compliance before scaling a campaign.
AMD settings only matter once there's a list behind them worth dialing carefully. Link Leads sells aged SMS and email lead lists at a flat $0.012 per lead, minimum order 5,000 leads ($60), filterable by state and age band (18–64) so a campaign can be sized to exactly the states a dialer setup is tuned for. Build one in the order builder, or pull a free 100-row sample first to test a dialer's AMD settings before committing to a full batch.