Interaction analysis by Recorder Analytics Rules

Understand how interaction are analyzed so that you can properly set up Recorder Analytics Rules.

Timing of interaction analysis

Recorder Analytics Rules perform analysis either during the interaction (real-time) or after the interaction ends (historical processing).

• Real-time: Analyze live interactions as they are being recorded. You can analyze all live recorded interactions, or select specific live interactions for analysis.

• Historical processing (also known as post-processing): Analyze previously recorded interactions using a campaign. While the campaign can select specific interactions, you must analyze all interactions in the campaign; you cannot analyze a subset of campaign interactions.

Amount of audio configured for analysis

Recorder Analytics Rules process some or all interaction audio, depending on the requirements of the Engine and the task being performed. By default, Identity Authentication and Fraud Detection processes only a section from the start of the interaction. However, enrollment always processes the entire interaction.

• Whole interaction analysis: Analyze all the audio from each interaction when performing identity verification or watch list detection. Analyzing all the audio increases the chances that the system can successfully perform identity verification and watch list detection. However, analyzing all the audio consumes more system resources and can slow server performance.

• Fixed duration: Analyze portions of the audio from each interaction when performing identity verification or watch list detection. You specify the exact size of the audio sample you want to use. Use of this setting optimizes the system performance.

Customer identification and interaction analysis

Analysis of interaction audio begins as soon as audio is received, up to the maximum amount of audio to use that is configured in the Recorder Analytics Rule. As customer metadata become available, the attributes configured in Identification Rules are used and analysis for customer verification continues with the customer voiceprint model (if one is available). The metadata values for customer identification can arrive at any point in the interaction.

Before V15.2 2020R1, for verification to be attempted, the arrival of the unique ID was required during the processing period at the beginning of the interaction. With V15.2 2020R1 or later, the Voice Biometrics Engine builds a model of the customer voice on the interaction during the analysis window. If the unique ID does not arrive during the analysis window, the model is retained until the Maximum Time to Wait for Customer Identification is reached or the call ends. When the engine receives the unique ID, it processes the verification. If a match is detected, the engine triggers the Verified event. However, the engine only triggers a Not verified event after the Maximum Time to Wait for Customer Identification expires.