Friday, September 19, 2008

Perhaps the most frequently asked question about traditional voice broadcast campaigns, i.e., excluding VoiceXML, is with regards to answering machine detection (AMD). AMD is somewhat fuzzy because detection is heavily reliant on the response on the other end of the line. This is mainly because once the copper ends it is tough to predict the nature of the conversation, whether it is a live person or a machine. Answering machine detection then becomes an exercise in predicting human behavior. For example, a typical person answers "Hello?" and waits for the other party to respond. On the other hand an answering machine usually has a run-on message "Hi this is Jane Doe, I'm not in, please leave your name , number and message, and I'll get back to you as soon as possible". Our detection algorithms take this and more into consideration when detecting answering machines and typically manage close to 90% accuracy (better than the 80% industry average) . Common reasons for 10% errors:

  1. Call is answered by a live person, but person does not say "Hello"
  2. Call is answered by a live person, but there's either too much background noise or the receiver has bad noise canceling.
  3. Call is answered by answering machine, but the message has long silences or is too short.
  4. Number is a PBX with multiple levels of voice mail prompts.

Hopefully this answers most questions about answering machine detection. As always, please drop us a line if you have any questions.