Generative AI: Automating Quality Management and Fueling Conversation Analytics

Introduction

Rising call volumes, increasingly complex calls and maintaining a qualified agent staff; these are three of the most complex challenges facing contact centers today. Their business objectives, however, are more ambitious than they’ve ever been.

Contact centers are in the revenue protection business. As a manager, you’re focused primarily on identifying negative interactions, elevating the customer experience and preventing customer churn. It’s essential that you have a complete understanding of the content and context of every phone interaction between agents and customers. That said, you probably operate in a business environment that requires you to do more with less.

Focus on Automation

Contact centers continuously monitor and measure customer satisfaction. It’s no secret why; delivering the best customer experience possible is Job #1 for a contact center, and that’s why agent performance is also front-and-center in the minds of managers, supervisors and team leads. 

In many contact centers, agents’ interactions with customers are manually evaluated by supervisors and team leaders. The crucial word here is "manually". Supervisors must listen to calls in order to understand whether or not the agent followed the call script, a review activity that takes virtually the same amount of time as the call itself. 

The result? Contact centers using a manual process typically can’t evaluate more than 1% of their calls, dramatically limiting their understanding of the quality of service they provide to their customers and compromising their ability to identify and address any issues. Leaders are essentially operating blind, making decisions – or not making them – absent the proper insights. There’s an inability to identify training needs, employee turnover increases and the customer experience and the organization’s brand suffer.

Enter automation and AI-fueled conversation analytics - tools and capabilities developed to simplify and streamline daily processes and analyze interactions without the need for manual intervention. Access to AI has been democratized; it's not just available to technologically advanced enterprise-sized companies. Even mid and small-sized contact centers are accessing its possibilities.

Generative AI-Fueled/Automation Conversation Analytics

Conversation Analytics is a technology tool that converts spoken language (an audio signal) into written text (a transcript). The technology is also capable of providing additional outputs useful for analyzing an agent/customer interaction, such as emotion and sentiment detection, along with the detection of other acoustic parameters.

It touches every facet of quality management, using the nuances of speech and silence to provide insight into customer and agent attitudes, agent aptitude, process effectiveness and trending issues. With it, 100% of contact center calls are automatically analyzed for compliance, customer satisfaction, agent effectiveness and much more. That time-heavy review and analysis now automated, Managers and Supervisors are free to devote more of their time to strategic priorities.

Conversation Analytics reaches deep into the Quality Management process. The visibility it provides fuels analysis, which yields insights. Managers can use those insights to evaluate and train agents. They can identify compliance issues, sources of customer dissatisfaction, agent performance issues and process inefficiencies.

Outputs:

• Continuous Speech Recognition - highly accurate speech-to-text capabilities and transcriptions
• Emotion Detection – automatically reads customers' emotions by examining variations in pitch or tone, interruptions, periods of silence and more.
• Acoustic parameters – identified for the overall conversation, as well as for each participant. These values are also available as absolute values (for instance, cross-talk time) and relative values (for instance, the ratio of cross-talk time against overall handling time)
  
  
  • Crosstalk time - participants talking at the same time
    
  • Silence time - nobody is talking
    
  • Number of interruptions - when one participant was interrupted by another
    
  • Speed of speech - number of words per minute
    
  • Speech phrases - predefined speech tags and phrases identified and highlighted on top of the transcription 
  • Gender

A visualization of available outputs provided below:

Conversation Analytics Use Cases

With the help of Conversation Analytics a user can easily leverage the available outputs in the following ways:

• Understanding the content of the call at a glance (when looking at key attributes) without the need to playback the call, reviewers can answer the following questions:
  
  
  • What was the overall mood of the customer? What was their emotional state at the beginning of the call, and how did it change as the call progressed?
        
  • Was the agent moody? Were they rude? Polite? 
        
  • How many seconds was the overall cross-talk time?
        
  • How many times was the customer interrupted by the agent and vice versa?
        
  • Was total silence time suspiciously long (was the agent pretending to work)?
        
  • What was the overall silence time (Was the agent able to help the customer)? 
        
  • Did the agent speak clearly? Do they talk too quickly? 
    
    
• Finding calls based on these attributes:
  
  
  • Find calls based on the available attributes quickly.
      
  • Real-time, full-text search allows for content to be found in any conversation transcription.
    
    
• Visualization of acoustic parameters and transcription during playback of the call:
  
  
  • Speech phrases and the detected emotion are highlighted within the interaction player´s waveform
       
  •  Captions are shown in the player during playback
    
    
• Leverage emotion and acoustic parameters when automatically evaluating conversation´s agents by the system (refer to the section Automated Quality Management for more details!)

Automated Quality Management

As mentioned previously, the manual review process is very time-consuming, and contact center managers are only able to evaluate an extremely small percentage of customer interactions. The ultimate goal for every contact center is to evaluate - or at least to understand - the experience your agents deliver during 100% of their conversations with customers.

Having an automated solution that can analyze all interactions on a regular basis can identify emerging issues, find new opportunities and highlight the outstanding and poor agent performance. Additionally, alerts can provide quick information about both outstanding and problematic conversations.

Automated Quality Management (AQM), sometimes also called Analytics-enabled QM automatically evaluates all conversations and the agents involved based on various parameter categories such as: 

• Metadata captured from CCaaS platform (handling time, attached flags, information about hangup party, etc.)
      
• Metadata analyzed by Automatic Speech Recognition (emotion, acoustic parameters, and transcription; see above for more details)
      
• Post-call assigned metadata (tags)
      
• Metadata assigned by external systems, typically a Survey tool (NPS score) or CRM system (case-related information)
  
  

The logical structure of AQM is shown in the picture below:

Automatic Quality Management Use Cases

With the help of AQM, Eleveo will automatically assign a score to all conversations based on a predefined set of rules set by the user. Typically, contact center managers want to reward agents who correctly follow the call scripts and other internal policies. 

The following table shows examples of how AQM works. However, rules for evaluations are fully configurable by the customer.

An integrated alert system is an important part of the tool, and the system notifies a predefined user or the agent´s supervisor if an automatically assigned score is less/greater than the defined threshold.

More AI-Enabled Capabilities

These next-generation AI capabilities add a new dimension of conceptual intelligence to call reviews and analyses:

• Auto Summarization increases agent productivity 10 percent or more by saving them from manually summarizing one call before taking the next.
• Flagging and Topic Detection alerts to potential customer churn. Wouldn’t you like to know if a customer mentioned a competitor by name or requested an escalation?
• Natural language ChatGPT-like engine allows you to build Auto Quality Management rules that can programatically, conceptually and accurately grade calls at a level mimicking a human’s capabilities.
  

Natural Language Processing
Natural language processing (NLP) refers to the branch of computer science—more specifically, the branch of artificial intelligence or AI—concerned with giving computers the ability to understand text and spoken words in much the same way human beings can.

Speaking of AQM and its new capabilities, the goal is to classify every single transcription according to the customer´s call script. Therefore, every question included as part of the call script (scorecard/questionnaire if preferred) will be automatically answered by the tool. The advantage of using this approach is that it is likely that the AI/Machine Learning tool will find the correct answer if trained using an appropriate dataset. The requirements to provide a data set and subsequently training the tool (goes hand in hand with costs) are potential downsides and have to be considered during the implementation phase for every single customer.

Text classified by Natural Language Processing (actually, the output of the classification) is intended to be used as an input parameter in AQM and will therefore have a direct impact on the conversation evaluation.

Reporting
Contact center managers and supervisors must be able to monitor, measure, follow trends, and work with aggregated data. The AQM and Conversation Analytics results must be available in the form of reports as well as KPIs on the Eleveo Dashboard.

Even Greater Automation
Once the system understands that something went wrong in a conversation it can propose an appropriate follow-up action. Such an action might include the scheduling of a manual review or the scheduling of agent training. On the other hand, the system can also reward outstanding agent performance, perhaps by letting them choose their preferred shifts.

With the help of Conversation Analytics and AQM tools, contact centers will be able to analyze what happened in each and every conversation, identify potential issues almost immediately, and constantly improve workforce performance – all of which are keys to improving customer satisfaction, and that is Job #1 for contact centers.