Objective of AMD Detection

Wolkvox's AMD (Answering Machine Detection) functionality is activated in automatic campaigns, both predictive and progressive, with the main objective of promptly detecting whether the person answering a call is a human or a machine (voicemail).

This early detection optimizes user contact, improves campaign efficiency, and ensures that detected humans are correctly classified and routed to the appropriate service flow, whether for sales, collections, or other services.

The central purpose of AMD is:

• Avoid unnecessary assignment of agents to voicemails.
• Improve the human user experience by efficiently routing them to an agent or a Virtual Agent.
• Optimize the operational resources of automatic campaigns.

1. Detection of Default Voicemails:

The first scenario corresponds to default voicemails, which are standard messages from telephone operators. These voicemails are detected within the first three seconds of the conversation. It applies to operators in any country, regardless of the provider.

This system uses continuous AI training, which is updated periodically to ensure:

• Changes in existing messages.
• New versions of voicemails.
• New providers that are integrated into the telephone ecosystem of Wolkvox clients.

This process ensures precise and updated detection in the face of the constant evolution of operators and their messaging platforms.

2. Detection of Custom Voicemails:

The second scenario corresponds to custom voicemails, where a person records their own message, for example:
"Welcome to Juan Pérez's voicemail, please leave your message after the tone."

In this case, detection is performed by counting words: The system analyzes the number of words detected in the first three seconds. If the message exceeds a certain threshold (for example, more than five words), it is considered not to be a standard human greeting. Under this condition, the call is classified as a voicemail and receives the same treatment as a default voicemail. This approach allows identifying non-standard messages without relying exclusively on pre-recorded patterns.

3. Detection by Silence in the Call:

There is an additional scenario where the call is answered, but no sound or voice is detected from the client. In these cases, the system waits up to 15 seconds, monitoring if any sound, word, or message is produced. If no vocal activity is detected during this period, the call is classified as a voicemail or machine. If the user starts speaking within that time, the call is identified as human and routed according to the established treatment. This mechanism allows handling situations where the user takes time to respond or where the line is left open without interaction.

Post-Detection Treatment

Once the classification is made, the system applies the corresponding treatment:

• Detected humans: are routed to a specialized team.
• Machines or voicemails: the configured flow for voicemails is applied, avoiding unnecessary agent intervention.
• Virtual Agents, as part of automated flows.
• Automatic campaigns with human agents, routing calls to specific groups of advisors.