GenAI Is Revolutionizing Conversation Analytics

View this article on the publisher’s website

The conversation analytics IT sector is strong and picking up momentum, due in large part to generative artificial intelligence (genAI)’s contributions. These solutions, first known as speech or text analytics and later as interaction analytics, have changed much more than their name. Initially used to identify why customers reached out to organizations, more recent iterations of interaction analytics applications proved highly effective at capturing, identifying, and delivering actionable intelligence about the customer and employee experience. But when today’s conversation analytics solutions are enhanced with genAI, they uncover deeper insights, operationalize findings, and generate improvements to the customer and employee experience (CX/EX).

GenAI is the missing link in many existing and emerging contact center solutions, particularly those based on conversation analytics technology, because it can improve both the applications’ implementation and ongoing performance. This applies to historical and real-time conversation analytics as well as related applications built on its technology, including transcription, analytics-enabled quality management (AQM), real-time guidance (RTG), next best action, real-time coaching, automated post-interaction summarization, and more. GenAI also reduces the development time for these solutions, as well as the effort required to test them.

Making Conversation Analytics Smarter

As conversation analytics technology and solutions have matured over the past 20 years, they have increasingly excelled at delivering meaningful data on customer effort, sentiment, and satisfaction, as well as helping to change and improve agents’ roles. However, these applications were limited by the transcription engine and language models used to make sense out of the inputs.

GenAI-based solutions leverage large language models (LLMs) consisting of millions (or more) records of relevant data to enrich findings from voice and digital utterances. They more accurately identify customer intents, enhance sentiment analysis and emotion detection, and correlate customer and employee behaviors with business outcomes. On a real-time basis, genAI supplies agents with additional insights, improves guidance and next best actions, and delivers the right customer and historical information, knowledge articles, procedures, reminders, and more to enable agents to provide a great customer experience. But these are just some obvious ways in which genAI is advancing the use and benefits of conversation analytics for contact centers.

It Starts with Transcription

Transcription is key to conversation analytics and its related applications, including those increasingly referred to as agent assist capabilities (e.g., real-time guidance, next best actions, and automated post-interaction summarization), and it is an essential contributor to the AQM process. But transcription engine accuracy varies, particularly when it comes to correctly identifying entities and other company- or vertical-specific terms and phrases; this is where genAI comes in. When genAI is supported by an LLM that is customized for the industry—or better yet the enterprise—it significantly improves transcription accuracy and boosts performance of each solution leveraging its outputs.

AQM Is the Way to Go

Contact center quality management has essentially been done the same way for the past 40 years. Although the process has been enabled by technology, the core activities—building evaluations, scoring agents, performing calibration, and conducting coaching—are done manually by supervisors or quality management specialists. Due to expense, companies have cut back on the number of resources involved in the process, even when the volume of inquiries increases. By 2022, most contact centers were evaluating a statistically irrelevant 1 percent to 2 percent of interactions. While the evaluators generally find agent improvement opportunities, it is not a representative sample, nor is it “fair” for agents. Additionally, many contact centers evaluate only calls, as they have not expanded their QM process to digital channels due to limited resources.

This is where conversation analytics comes in: It can automate the QM process and review up to 100 percent of customer interactions across channels and at a much lower cost. AQM has proven most effective at evaluating conversation openings, the verification process, the delivery of required disclosures, and conversation closings. While AQM is not perfect, it does a good job of identifying relevant and important company and agent trends; evaluating agent soft skills and empathy; and assessing customer emotion and sentiment. The addition of genAI increases AQM’s benefits with its ability to score agent behavior-based categories and generate evaluation forms and coaching content for the supervisor/QM resource, greatly reducing the time they spend on these activities. The expected payback for AQM solutions is 12 to 18 months, which can be much shorter depending on how the findings are applied.

Making It Real

Real-time guidance can make a positive impact on contact centers in many areas including agent onboarding, training, and engagement; the customer experience and brand loyalty; first-contact resolution; and productivity. The purpose of RTG is to give agents the needed knowledge, procedures, and guidance to enable them to spend their time helping customers instead of looking for information. Another form of RTG geared toward sales and collections environments is next best action, which provides targeted prompts for new sales, upselling, cross-selling, retention, collections, or other purposes.

RTG (including next best action) is a good idea on its own, but when it’s augmented with genAI it can more quickly and accurately understand what is being discussed to optimize the delivery of guidance to agents. Just as important, genAI speeds up the implementation process, as it reduces the need to identify all the ways in which the same issues can be communicated by customers and agents. Due to the proven benefits, the payback from RTG is expected to be 3 to 18 months.

Wrapping It Up

Until recently, the post-interaction wrap-up process was one of the most disliked activities required of agents. In most contact centers, agents are required to draft a summary of each customer conversation and post it to the system of record. This information becomes part of the customer’s account history and provides insights into recent conversations or activities when other employees interact with them. Agents typically also need to enter a wrap-up code to capture the intent of each inquiry so managers know about trending issues.

Both tasks are good ideas, but they generally don’t work as planned. Agents may input a “favorite” wrap-up code or use a catch-all category if customers discuss multiple points. Agent-generated summaries frequently omit salient points since they rush to complete them and move on to the next interaction, and they can also include unfamiliar acronyms or simply not be well-written. And since agents don’t have time to read full transcripts, that’s not a realistic alternative.

With genAI, transcripts of voice and digital conversations can be summarized and returned for agent review in a matter of seconds. The automated summaries can be modified by the agent, if needed, and submitted to a CRM system, or the application may post it directly. More advanced solutions can also identify follow-up actions or agent commitments from the conversation and schedule tasks or initiate automated workflows to complete the activities on the agent’s behalf. This is an ideal use of genAI with a payback of three to six months as it reduces the post-interaction summary and wrap up process by at least 50 percent, substantially reducing interaction average handle time while enhancing both the CX and EX.

A Cautionary Tale

GenAI is exciting and has captured the attention of business and IT leaders alike due to its benefits, but it comes with risks that should not be ignored. GenAI is reliant on foundation models, such as LLMs, and most data repositories—e.g., the internet, a company’s internal knowledge bases, or a curated dataset from a CX vendor—might contain some form of bias or information that is right for one organization and wrong for another. Building an LLM is expensive and time-consuming because it must be targeted, tagged, curated, and maintained; and accessing third-party LLMs can be pricey. Companies need guardrails to reduce the risk of hallucinations and inappropriate answers being created by their genAI solutions, as well as trained resources to manage these applications to maximize benefits and minimize risks, which can be difficult and expensive.

Final Thoughts

Conversation analytics has become an essential stand-alone application, as well as an important underlying technology building block. The introduction of genAI has greatly enhanced many aspects of these solutions, making them more valuable and useful for contact centers and enterprises. GenAI fills in many of the gaps in understanding and improves transcription, AQM, real-time guidance/next best action, and post-interaction summarization accuracy, positioning these solutions to resolve long-standing challenges for contact centers and their employees. But this is just the start of the contributions that genAI is making throughout the CX industry. In under 18 months, genAI has transformed many contact center activities, but as good as these solutions are now, their contributions and capabilities are quickly improving due to AI’s inherent intelligence.