Skip to main content

Understanding AMD (Answering Machine Detection) and Voicemail Tuning

How AMD works, when to tune the Analysis Time setting, the "connecting at end of voicemail" pattern, and when to escalate to support.

AMD (Answering Machine Detection) decides whether an outbound call was picked up by a human or a voicemail. When AMD works, the dialer can drop a pre-recorded voicemail message or skip the call entirely. When AMD doesn't work, you may see agents connected to a voicemail that's mid-greeting or already saying "goodbye." This article explains the setting that controls AMD timing, common patterns we see, and when to contact support.


How AMD Works

During the first few seconds of an outbound call, the dialer's telephony layer listens to the audio coming back from the called party. Based on the cadence, length, and shape of the audio, it classifies the call as one of:

  • Human — connect the agent immediately.

  • Machine — drop a voicemail message (if configured), or hang up.

  • Unknown — fall back to default behavior (usually: treat as human).

The window the dialer listens for is controlled by the Analysis Time setting.


The Analysis Time Setting

Default: 4000 ms (4 seconds). [Confirm with PM: exact default in current build]

What it does: Tells the dialer how long to listen before deciding human vs. machine.

When to increase it: If agents are frequently connecting to the end of a voicemail greeting (catching only "...please leave a message after the beep" or "goodbye"), increase Analysis Time to 5500 ms as a first step. A longer window gives AMD more audio to classify on, and delays the agent connect until after the greeting on machines that end fast.

⚠️ Tuning Analysis Time is a tradeoff. A longer window means humans wait longer before the agent comes on — which can hurt pickup-to-connect rates. Test changes on one campaign first.


Premium AMD (Upgrade)

Premium AMD is an enhanced detection model with better accuracy on tricky voicemails. It's available as an upgrade


Troubleshooting steps:

  1. Confirm the which browser the agent is using (we recommend using a chromium based browser).

  2. Increase Analysis Time from 4000 ms to 5500 ms on the affected campaign.

  3. Run 50–100 calls and check whether the pattern reduces.

  4. If you're on Premium AMD, ask support to temporarily disable Premium AMD on your agency to rule out the connectAfterAMD race.

  5. If the pattern persists, escalate to support with specific call IDs.


When to Escalate to Support

Contact support if:

  • Tuning Analysis Time made no measurable difference.

  • You're on Premium AMD and seeing dropped calls at the connect moment.

What to include in the ticket: 3–5 specific calls, the times of the calls, the campaign, and the agent who took them.


📌 Tip

Don't tune Analysis Time agency-wide as a first move. Pick one campaign, change the setting, run a meaningful call volume (50–100 calls minimum), then compare connect-to-conversation rates before rolling the change wider.

Did this answer your question?