Analyst Perspectives

We are going through an analytics revolution in and around contact centers. There is more data to analyze and better tools to conduct the analysis. One important advancement is assessing customer sentiment and using it to predict (and influence) behavior. Enterprises are using customer experience programs to identify and act upon hidden "moments of leverage" where customers may be inclined or primed to make a purchase. To do that effectively, enterprises need insight into how the customer perceives the relationship: what is said and implied as well as emotions expressed and those not. 

Historically, contact centers collected very basic details of sentiment, mainly through surveys that asked very binary questions about satisfaction. Early use cases for speech analytics were also useful in creating transcriptions that could highlight specific words associated with states of mind. The findings were (and still are) used mostly for agent evaluation, not necessarily to expand the customer's buying footprint. Slowly, the industry is moving towards a deeper use of customer sentiment information. 

There are multiple ways to gather this data. The most basic is through speech processing systems, in which voice data is turned into text by automated speech recognition. As noted, this functionality has been around for decades as a standard part of many agent management tools. Its value is that it provides a snapshot of what might have gone right or wrong during a particular interaction that can be used during agent quality evaluations. The downside is that it does not provide any insight into whether the customer is broadly satisfied with the relationship, nor does it clue you into behavioral intentions. It tells you just enough to know whether an agent needs training or coaching, but unfortunately, its analysis is often delayed long enough to prevent supervisors from interceding to change the customer view of a call gone wrong. 

Another element in the toolkit that has also been in use for many years is acoustic feature analysis. It is the process of using the qualities of the conversation (as opposed to the words themselves) to infer emotion and sentiment. It examines the finer nuances of speech like pitch, tone, intonation, silences, crosstalk and other more subjective components to provide data points that can sometimes be correlated with actual sentiment. Putting these two modes together provides a more comprehensive look at what happens during calls, especially when the analysis software can classify and visualize the findings to highlight patterns. These can be displayed in word clouds, heat maps and other clever visuals to help a manager isolate problems. 

In general, these methods do a good job of identifying small-bore problems, but aren't extensive enough to provide predictive insights, which, in turn, are needed to determine whether someone should be upsold, coddled or pitched something completely new to them. For that, users need more modern systems that use machine learning trained on extensive conversational data to build a model that understands the patterns underlying sentiment. This is what enables predictions based on new text and conversations that the system hasn't seen yet. That's where the current state of the art lies - with artificial intelligence and machine learning-based tools capable of doing broader analysis on the entire corpus of customer communications. Through 2027, automated analytics technologies will become better at determining customer intent and sentiment than customers reporting through surveys. 

Contact centers should be looking at how to expand the range of possible use cases for this technology set. The prevailing view - that speech and text analytics' best use is to assess agent performance - is woefully self-limiting. Instead of focusing on the aftermath of individual interactions, modern sentiment analysis enables the assessment of trends within large groups of customers. It works alongside systems that segment customers into logical groups that share characteristics, which in turn allows marketers to pinpoint offers or campaigns to people with specific interests and feelings. 

Valid use cases begin with center operations, to be sure. They should be using sentiment to discern the mood of individual customers in real time, especially now that real-time AI-based automation can work alongside sentiment to guide the agent through the thicket of customers' emotional responses. Centers need to pivot from using sentiment to assess the aftermath to using it to orchestrate the present and future behavior of both agents and customers. 

AI can produce sentiment assessments in near enough to real time to allow for "instant" saves - that is, automated warnings to supervisors that an interaction has just taken a wrong turn. Even if the interaction is over, the interval that's passed is short enough to allow trained agents to call back and make something right. This is a sharp contrast to finding out weeks later that a person or a policy is creating negative sentiment that has been festering for some time. 

Use cases outside the center are even more exciting. CX is an enterprise-wide endeavor enhanced by shared data and goals. Other departments can use it for product development and assessment, for process improvement, competitive analysis, brand analysis, segmentation and promotions. These analyses are happening anyway, by customer-related teams in marketing and other departments. Adding real-time and historic sentiment data to the mix can tie the contact center more closely to corporate goals and help knit strategic and tactical arms of the business together. 

That said, sentiment analysis does not come without risks and constraints. Any application that maximizes machine learning requires large data sets with accurate categorization to uncover genuine insights. It is also unclear how successful modern systems are at parsing things like ambiguity, sarcasm, irony and local/regional conversational patterns. Users need to understand the different values they get from analyzing moment-by-moment individual activity versus wider trends in larger customer groups. 

Also, when most interactions are handled in digital channels, it becomes important to use sentiment to judge the effectiveness of automated interactions with chat systems, self-service, and feedback surveys. 

Buyers should also be prepared for costs that can run to thousands of dollars per month, including implementation and model training, plus costs for necessary system updates. On the positive side of the ledger, there are indications that implementing sentiment analysis can lead to more efficient call routing, reduced handle time and higher first contact resolution. 

Buyers should address several important questions about current operations and plans: 

• Are you using the full capability set of your current tools? 
• If so, are you using these tools to maintain the status quo (i.e., to assess agent performance) or to grow the revenue and customer base? 
• Have you assessed how other CX stakeholders - especially those in sales and marketing - can maximize sentiment data that can only be collected in the contact center? 

Software providers can use sentiment analysis to boost the value of many different software categories, including marketing automation, customer feedback management, agent performance enhancement, social media monitoring and e-commerce as well as standard interaction handling. Sentiment analysis is not a category unto itself but a powerful tool that enhances the value of the software it is embedded in. It is likely to become a default component of many customer-facing and customer-adjacent processes and systems before too long. 

Regards,

Keith Dawson