Answering machine detection

More than 80% on average. Accuracy depends on the region and the carrier: each has its own prerecorded messages and voice assistants.

About 80% of calls with silence after the answer are identified correctly. Accuracy improves as the model is updated under your support plan.

Any detection system carries that risk, so the check is tuned for each campaign. Where a customer is especially valuable, it's loosened: better for an agent to hear an extra answering machine than to lose a person.

The check takes the first few seconds after the answer. We tune the acceptable pause during setup – it's a balance between accuracy and the customer's wait.

Besides pauses and greeting length, the model takes the voice into account. It tells virtual receptionists and voice assistants apart from a human on the line.

The call ends with a flag, and the number goes into the campaign's next attempt – for example, at a different time of day.

Instead of dropping the call, the platform can wait for the beep and play your message – for example, a request to call back or an appointment reminder. Under the TCPA, prerecorded messages generally require the person's prior consent, so use this only for contacts who agreed to it.

Dropped calls are flagged in the report and recorded, and you spot-check them by listening during setup and after launch.

Outbound call recordings automatically build up as training examples, along with errors flagged by agents. The model is retrained and tested on your calls before it goes live.

It does. Through the API, your system learns who answered the call and then applies its own rules.

When the platform calls back someone whose call was missed, an answering machine won't reach an agent, and the attempt is repeated under your callback rules.

Missed call callback →

English is supported: the model recognizes English-language voicemail greetings and carrier messages, and it also checks calls to international numbers. A pilot campaign shows the accuracy on your list.

When the platform dials one number per available agent, the agent is connected as soon as someone picks up – without detection, they'd be listening to answering machines themselves. With detection, only people get through.

It can. Detection is installed along with the other modules – in our data center or on your premises.

Take the number of outbound calls per day, the share of machine answers from the report, and the average time it takes an agent to recognize a machine by ear. Multiply them to get the agent hours you're currently wasting, then add the carrier minutes spent on those same calls. A pilot campaign will show these numbers on your own list.

There are no licenses. Our own model detects answering machines on the server, so there's no per-minute speech recognition fee. Setup, configuration, and model training are billed in support hours, and carrier minutes are paid to your carrier.

Pricing →