Answering machine detection may be a significant factor in determining agent productivity when using predictive dialing, but does have side effects.

Answering machine detection is based on analysis of the characteristics of the speech received when the called party answers the phone.  There are various characteristics which may be analyzed, with accuracy dependent in part on the amount of speech which is analyzed.  There are, for example, distinctly different social norms for the speech patterns with which one answers the telephone vs. the speech patterns of an answering machine greeting.  However, just as when a human listens to the initial speech on the line, it is possible to mistake an answering machine for a live person and vice versa.

Here are some key points to consider when assessing the potential impact of answering machine detection.

• Answering machine detection uses the initial speech of the called party to determine whether to treat the connection as a live call or an answering machine.  This will cause a brief delay between the time at which the called party answers the phone, and the time at which they are connected to an agent.  It's important to understand how long this delay will be.  If there are options to control the length of the delay (i.e. modify the amount of speech analyzed), how will this impact the accuracy of the detection algorithm.  Remember that the called party initially has no incentive to engage in a dialogue.  An excessive delay may dramatically reduce the success which your agents have in engaging your targeted audience in dialogue.
• Consider not only the accuracy of a prospective detection algorithm, but also the impact of inaccurate classifications.
• Consider the answering machine which is inadvertently cut through to a live agent.  The agent will then lose time in disposing of the call and waiting for the next call.  This is the lost time that answering machine detection is designed to avoid.  If a detection algorithm accurately detects answering machines 90% of the time, this benefit is then available for 90% of the calls which are answered by answering machines.
• Consider also the live caller who is inadvertently treated like an answering machine, likely being either hung up on, or in some cases receiving an automated message.  The called party will likely consider this a nuisance call, and may be significantly less approachable in the future.  Furthermore, such events may go undetected if they are lumped together in your reporting with other calls which actually are answered by answering machines.
• Some algorithms are based on analysis of background static, recognizing static which is characteristic of a recorded greeting on an aging answering machine tape.  With the increase in digital answering machines, greetings are increasingly recorded as digital images, allowing them to maintain their original crispness over time, and thereby decreasing the effectiveness of such detection techniques.

Answering machine detection can increase agent productivity be disposing of many of the connected answering machines without any agent involvement.  When an answering machine is cut through to an agent, the agent loses several seconds to disposition the call and then typically an amount of time waiting for the next call.  If your targeted audience frequently uses answering machines, Answering machine detection can provide significant increases in agent talk time.  However, this gain needs to be weighed against any impact the side effects may have on the effectiveness of your agent's dialogues once talking.