The Answering Machine Detection module in 3CLogic determines whether a call has been connected to a ‘Live’ person or an answering machine. Since the Answering Machine Detection takes place on the agents PC rather than on the central server the call is quickly transferred to the agent if a ‘Live’ person is detected. In case of central server after the server detects ‘Live’ person, the server has to identify ‘available’ agent and then attempt to transfer the call.
Why Answering Machine Detection?
Answering Machine Detection is a critical module for any Call Center solution as it helps to save valuable agent time by increasing Average Talk Time (ATT) of agents and broadcasting messages on answering machines. ATT is defined as the time an agent spends speaking to ‘Live’ leads in one hour. Answering Machine Detection can be used by the Dialer to leave pre-recorded message on the answering machine automatically.
3CLogic Answering Machine Detection
Answering Machine Detection Configuration
Answering Machine Detection is configured at campaign level allowing the call center administrator to manage Answering Machine Detection at individual campaign level rather than at call center level. The ability to configure Answering Machine Detection at campaign level gives more flexibility and control for the call center administrator. To enable/disable Answering Machine Detection, log into the management portal, Go to the Campaigns tab, Click on “Manage” , Go to the “Advanced” tab under manage. There is a drop down available to enable/disable Answering Machine Detection for the campaign
The Answering Machine Detection module runs only with the following Dialer Modes:
1. Progressive Dialer
2. Predictive Dialer
If the connected call is detected as an answering machine when the the Dialer is running in the ‘Predictive Dialer’ or Progressive Dialer mode, the call is immediately disconnected and the next call dialed if the broadcast on answering machines is disabled. If broadcast message on answering machine is enabled, “Play recorded message on Answering Machine”” under “Broadcast” for the campaign (Campaigns-> Manage-> Broadcast), the Dialer would wait for a period of silence and start playing the message file (.wav) and disconnects the call. As soon as the ‘Broadcast’ starts the dialer places the next call for the agent while a message is being left on the Answering Machine.
Note: If the power dialer is being used, only the call color is changed as an indication to the agent that according to this system this seems like an answering machine. The system does disconnect the call or leave a message on Answering machine.
Answering Machine Detection Settings
There are three predefined settings to configure Answering Machine Detection:
The conservative setting implies that “The system will become more conservative in determining Answering Machines”. This means there is a chance that some Answering Machines will go through as humans and the agent will have to determine that the placed call is answering machine and disconnect the call manually.
The conservative mode is suited for campaigns where the number of leads is less or connecting with a ‘Live’ lead is important. The average talk time for agent will be less but very few if any ‘Live’ leads are misinterpreted as answering machines e.g. A 5 Star Hotel needs to call its premium members and let them know that tomorrow is the last day to renew their membership before it expires. Such a campaign where the numbers of leads is less or contacting a lead is extremely important and you want very less chance of a human being detected as an Answering Machine, the mode of “Conservative” would be recommended. To summarize Answering Machine Detection in conservative mode:
1. More conservative in detecting Answering Machines
2. Chances of Answering Machines going through as ‘Live’ person to Agents increases
This is the default setting and the optimal setting for Answering Machine Detection. This gives the most accuracy and a good balance between false positives (‘Live’ Persons are detected as Answering Machine Detection) and false negatives (Answering Machine Detection are detected are ‘Live’ person). This is recommended mode for running Answering Machines and fits most of the campaigns being run. This mode is recommended for Sale Campaigns, Political Campaigns, Surveys, and Polls etc. To Summarize
1. Provides a good balance between false positives and false negatives.
2. This is the default mode and the recommended mode for most campaigns.
In aggressive setting, “The system will be very aggressive in determining answering machines and will use the shortest possible time to do so”. The aggressive setting increases the chance of ‘Live’ persons being detected as answering machines. This is suited for campaigns where there are a large number of leads, limited time and the some leads can be skipped (can be tried later again, if time permits).
This setting is recommended for campaigns where trying to connect to more leads is important than connecting to all ‘Live’ leads e.g. “A political campaign to broadcast messages to ‘Live’ persons with a lead list of over a million leads to inform them of a meeting scheduled for the next day”. To summarize:
1. Aggressive in determining answering machines.
2. Chances of humans being detected as answering machines increases.
Answering Machine Detection Parameters
The following parameters are used by the Answering Machine Detection module of 3CLogic. The parameters are:
Max Analysis Time
This parameter specifies the maximum time Answering Machine Detection module will run trying to detect an answering machine. The time starts after a call is connected. If the module is unable to make a decision within this time period the default is to determine the call as a human. The most effective time to detect Answering Machine Detection is the small window between 3500ms to 4500ms after the call is connected. Making the max analysis time too low or too high results in ‘Live’ person all the time. This parameter has no contribution to the modes of Answering Machine Detection (Conservative, Moderate, and Aggressive). This specifies the maximum time for analysis in which the module makes the decision. This value is best set at 4000ms and in future this would be made static for all the three modes.
Initial Silence Time
This is the time when a call connects to the first appearance of voice signal. Typically answering machines have a wind up time and hence take some time to respond after a call is connected unlike a human who generally says a ‘hello’ as soon as call is ‘connected’. If there is no voice signal between the call connect and the value as specified under “initial silence time” the call is determined as an answering machine. The effect of increasing this value is to make the Answering Machine Detection module more conservative in detecting answering machines.
Thus, the module might let a few answering machines pass to agents as ‘Live’ persons. As you keep increasing this parameter there is a higher chance that you would get voice signal within that time frame and as we get voice signal, it’s a human according to the system.
Max Silence after Greeting
This is the time a human waits after the initial greeting. Typically humans say “Hello” and then wait for a response from the other end. Answering machines typically do not wait for a response and continue to speak (Thank you for calling the residence of ..) .
Increasing this parameter will make the Answering Machine Detection module more aggressive in detecting answering machines. This will happen because if this parameter is increased, the second “hello” from the human who was waiting for a response might come in and since we did not have a silence time less than “Max Silence after greeting”, this would be determined as an answering machine. Applying the same logic, decreasing the parameter will make the Answering Machine Detection module conservative in determining answering machines.
Max words to Analyze
This parameter specifies the number of words the module used to determine if the connected call is an answering machine. Typically humans would say a “Hello” or “Good Morning” and wait. Answering machines would speak a longer sentence and hence more words within max analysis time. Reducing the number of words makes the module more aggressive in determining answering machines. On the flipside increasing the number of words makes the answering machine module more conservative.
1. If packet loss is too high we cannot accurately determine if placed call is an answering machine, hence would always go to the default which is ‘Live’ person
2. If there is too much of noise in the channel the module falls back to the default that is ‘Live’ person
3. If the module cannot determine within max analysis time and there is uncertainty we fall back to the default that is ‘Live’ person
4. If there is any delay in the media there is a high chance of detecting humans as answering machines or vice-versa.
5. The module tries to detect machine tones as answering machines which include Fax numbers.
6. Answering Machine Detection is accurate from 80-85%