Answering-machine detection, usually shortened to AMD, is carrier-side classification of who or what picked up, requested on every dial so that the agent is not holding a conversation with a greeting.
On a campaign that did not elect a voicemail message — the default — a machine verdict hangs the call up immediately. Nothing is spoken to the machine, no conversation slot is claimed, and the dialer then takes an explicit retry-or-dispose branch that is written onto the attempt itself. The outcome is recorded rather than disposed silently, which is the bar every disposition on this platform is held to.
Detection is imperfect in both directions, and both directions are treated as compliance events rather than as quality noise. A live person misclassified as a machine hears silence and a hang-up, which from their side is indistinguishable from an abandoned call, so it counts against the same abandonment budget the pacer is governed by, and the classifier’s accuracy is sampled rather than assumed.
Where a campaign has elected to leave an artificial-voice voicemail message under an executed Cold Outreach and Data Provenance Waiver, that message is itself a regulated call. It carries the AI-disclosure opener composed from the same pinned agent version a live conversation discloses from, the internal suppression list is re-read at the instant it is left, and no contact receives more than one such message per campaign. Vocapable records the text it dispatched, that text’s hash and the waiver in force at that moment. What every attempt leaves on the record afterwards is on the outcomes page.