Answering machine detection (AMD)

How it works

How AMD works

The system listens to the audio right after the call is answered. A person usually says a short "Hello?" and waits. An answering machine talks longer without pauses and often ends with a beep. Classic answering machine detection measures exactly that: how long the first phrase lasts, how much silence there is, and how much speech in total. Newer systems add speech recognition and machine learning models, which lets them spot voice assistants too.

• Initial silence – how long until the first speech
• Greeting length – a person says 1-2 words, a machine speaks for several seconds
• Pause after the greeting – a person waits for a reply
• Beep – the tone after which voicemail starts recording

Limits

Accuracy and trade-offs

Answering machine detection always trades one error for another. A strict setting catches more machines but sometimes hangs up on real people. A loose one connects more people but lets more machines through to agents. On top of that, the analysis takes 1-3 seconds, and the person on the line waits for an agent the whole time. In the US, telemarketing rules also limit silent and abandoned calls, so check your settings with your compliance team.

Practice

Setting it up well

• Tune the thresholds on your own contact list and check the results against recordings
• Track the share of calls marked "human," "machine," and "unknown"
• Send unknown calls to an agent instead of dropping them
• Where it makes sense, leave a short recorded message on voicemail
• Pair answering machine detection with power or predictive dialing – that is where it pays off most

In CONTACT.WORK

How CONTACT.WORK uses it

In CONTACT.WORK, answering machines are detected in the auto dialer, in progressive and predictive modes. The system analyzes the first seconds after the answer, before an agent joins, and tells a live voice apart from carrier voicemail, virtual receptionists, voice assistants, and silence. These calls end automatically and are flagged in the campaign results (the API and webhooks can pass them straight to your CRM), and the detection models are fine-tuned on your calls.