Skip to content
AI Dialer

Glossary · Dialing

What is Answering machine detection?

  • Also called: AMD, answer machine detection, voicemail detection

Reviewed by Sujan ThapaliyaLast updated

Definition

Answering machine detection is the technique a dialler uses to decide, within the first seconds of a connected call, whether a human or a voicemail system answered. Human answers are bridged to an agent; machine answers get a voicemail drop or are hung up.
The three STIR/SHAKEN attestation levels, which underpin how answering machine detection behaves in practice: A asserts both that the carrier knows the customer and that the customer may use the calling number, B asserts the customer only, and C asserts neither. A shared number pool can reach B at best, which is why number ownership keeps recurring in these definitions.
Attestation is the backdrop to most caller-ID and deliverability terms, answering machine detection included.

In practice

How answering machine detection actually works

Detection works on the shape of the audio, not its meaning: a person says a short word and stops, while a greeting runs for several seconds with a consistent cadence and often ends in a beep. Modern systems combine that timing analysis with speech recognition to classify faster and more accurately.

The trade-off is always speed against accuracy. Waiting longer gives a more confident classification but leaves a real person listening to silence, which is the most common reason a live prospect hangs up on an auto-dialled call. Aggressive settings connect faster but hand agents more voicemails.

False positives are the expensive failure. When AMD wrongly classifies a person as a machine, the dialler hangs up on a live prospect, and under US rules that disconnect counts as an abandoned call.

Worked example

A typical target is classification within 1.5 seconds at 90–95% accuracy. At 2,000 calls a day and a 5% error rate, that is 100 misclassified calls, worth measuring rather than assuming.

Answering machine detection: common questions

How accurate is answering machine detection?
Well-tuned detection classifies correctly around 90–95% of the time, usually within 1.5 seconds. The remaining error is not evenly split in cost: a machine misread as a human wastes an agent's few seconds, while a human misread as a machine gets hung up on, and that disconnect counts as an abandoned call.
Why does the caller hear silence before I speak?
That pause is the detector deciding. It is listening for whether the audio has the shape of a person saying two words and stopping, or a greeting running with even cadence. Longer analysis is more accurate and more likely to make a live person hang up, which is the trade-off every setting is balancing.
Can answering machine detection be turned off?
Yes, and on preview or power dialing it often should be. The rep is already on the line and can hear a greeting for themselves, so detection adds a pause without adding information. It earns its place on parallel and predictive dialing, where nobody is listening until a call is bridged.

Sources

  1. Combating Spoofed Robocalls with Caller ID Authentication — Federal Communications CommissionThe STIR/SHAKEN framework, the attestation levels carriers sign calls with, and the mandate requiring providers to authenticate caller ID.
  2. 47 U.S.C. § 227 — Restrictions on the use of telephone equipment — Cornell Legal Information InstituteThe Telephone Consumer Protection Act itself — the consent requirements, calling-hours limits, and private right of action.
  3. ITU-T Recommendation E.164 — The international public telecommunication numbering plan — International Telecommunication UnionThe international number format, the 15-digit maximum, and how country codes and national numbers compose.

See answering machine detection in the product

A parallel dialer places several outbound calls at once for a single rep and connects the first one a human answers, dropping the rest. Because most cold calls go unanswered, dialling three to five lines in parallel produces several times as many live conversations per hour as one-at-a-time dialling.

Explore parallel dialerGet early access