By Kurt Williams, Co-Founder and Chief Evangelist, InMoment
Historically, “emotion” hasn’t gotten much respect in the business world. Terms like “fuzzy,” “squishy” and “soft” dominated the dialog. My, how things have changed! In 2015, Forrester Research released a study that evaluated the three dimensions of customer experience: effectiveness, ease, and emotion, and declared emotion the most powerful influence on brand loyalty. Earlier this year, analyst and founder of the Temkin Group, Bruce Temkin, put an exclamation mark on the idea, crowning 2016 the “Year of Emotion.”
The customer experience industry has now turned its attention squarely toward determining how to reveal, classify, and take action on emotion. What better way to check in on the progress that’s been made in this area than to attend the “Sentiment Analysis Symposium” hosted by Alta Plana’s Seth Grimes in New York City.
A Melting Pot of EmotionAs one of the world’s most vibrant melting pots, New York was the perfect environment to examine both sentiment and emotion. During the conference I tweeted, “If you don’t love New York City, you’re not doing it right.” Hats off to Seth and the Alta Plana staff for picking the perfect venue to analyze sentiment.
I’ve been a frequent attendee of the Sentiment Symposium; even spoken a few times. Perhaps my favorite part of the Symposium is tracking how the text analytics industry has evolved over the last decade. Sometimes it slowly changes. Sometimes it violently surges ahead. The 2016 Symposium was a landmark example of the latter—especially in the field of customer experience.
The Three Seismic Forces of CXThe last 15 years of the Voice of the Customer industry have been characterized by surveys that asked customers to rate (or score) the performance of key activities by a particular brand. These programs have been driven by a desire to improve overall scores collected and aggregated during those surveys with OSAT, CSI, and NPS being most typical. In 2016, three forces are dramatically changing this traditional view of CX and VoC: customer journeys, modern text analytics, and machine learning. All of these forces were on display at the Sentiment Symposium.
Customer JourneysA focus on the customer journey has taken the CX industry by storm—and with good reason. It isn’t just a set of diagrams, a shiny new process, or a tool for cultural change (though it is all of the above). It is a fundamentally different approach and a paradigm shift for the industry characterized by a desire to redesign the customer experience in a holistic manner. Customer journeys don’t focus on any one touchpoint or activity. They track the entire customer experience from cradle to grave. Rather than focusing on scores, a customer journey approach looks at the entire customer experience, which—like any story—has a beginning, middle, and end. Customer journeys are becoming the orchestration layer for modern customer experience programs. The speakers at the Sentiment Symposium who were dealing with customer sentiment clearly understood that customer emotions should be understood contextually at each moment in the customer story, in addition to understanding the overall journey.
Modern Text Analytics = Ensemble MethodsSentiment analysis is a sub-discipline of the larger field of text analytics. Discovering sentiment within text or audio has always been an important part of text analytics because it provides an extra (and very human) dimension of understanding that goes beyond topical exploration of text. This year at the Symposium, every element of modern text analytics was on display: machine learning, rule bases, lexical analysis of parts of speech, and taxonomies. What was fundamentally different this year, though, was that all of these approaches are finally being combined effectively to create more accurate and precise understanding of unstructured data.
Combining approaches, which is called an “ensemble method,” is clearly delivering the best understanding of customer sentiment. No single method is good enough to deliver the results the industry expects. The debate of whether rule-based lexical approaches are better than statistical machine learning approaches is over: the best results are had by leveraging both.
Machine LearningMachine learning has always been a staple at the Sentiment Symposium, but this year it exploded onto the scene. Machine learning is a type of predictive analytics where an algorithm or set of computer instructions modifies and improves itself based on repeated exposures to stimulus. Different types of machine learning exist for different prediction scenarios, but they all behave similarly. The machine is trained to spot patterns in data typically using sample data that is known to exhibit the pattern. When the machine algorithm has been exposed to enough data to learn the pattern, it can be used in the real world to spot similar patterns in new data it hasn’t seen before. Anyone who has used linear regression has used a very simple form of machine learning, though greater predictive power is possible using more advanced algorithms such as decision trees, support vector machines, and neural networks.
Jason Baldridge, co-founder of People Pattern and former professor of computational linguistics at the University of Texas anchored the conference with a thought-provoking tutorial on using computational linguistics to provide insights from sentiment and emotion. He demonstrated how ensemble machine learning methods can be combined with linguistic rule approaches to create what he calls “Aspect-Based Opinion Mining,” which leverages the strengths and minimizes the weaknesses of each method. He further explored the value of context on understanding emotion using non-textual aspects such as geography, social networks, and contextual imagery.
Social SentimentPerhaps the most typical use case of sentiment analysis is in the processing of social media posts. I’ve often described sentiment analysis as one of the three pillars of social media marketing analytics (the other two being influencer analysis and mention analysis). This year, the sentiment of social media has been analyzed in a more mature fashion. Unfortunately, sentiment analysis is all too frequently based on analyzing the structured scoring captured from posts. Bottlenose’s Adam Blumenfeld pointed out that saying, “Emotions are not contained in a ‘Like’ button… We are all storytellers here.”
What’s Next: Deep LearningMachine learning is becoming a key component of every text analytics offering. It is also the future of emotional understanding of human-generated content. A particularly advanced type of machine learning turned up in a surprising number of presentations at the 2016 Sentiment Symposium. It is obviously where the lion’s share of research resources are going.
Deep learning builds upon machine learning and moves it even closer toward the capability of the human mind. Machine learning already attempts to model human understanding through the use of techniques such as neural networks, which mimic the way the brain's neurons learn. Deep learning builds upon that foundation additively by leveraging rule bases and ensemble methods as inputs into deeply structured neural networks in order to achieve even higher levels of machine cognition.
This type of learning requires massive computing resources that previously have only been available to large institutions like governments and internet giants such as Google. Cloud computing infrastructures have changed the rules making deep learning available to many kinds of businesses.
In the near future, deep learning will unlock highly accurate machine language translation, image and video recognition, and better sentiment engines capable of dealing with the nuances of communicating human emotions.
Turning Emotion into Business ImpactThe point of sentiment and emotion analysis is to drive positive business outcomes, about as far from “fuzzy” as you can get. Traditional VoC programs have had challenges in providing the insights required to make meaningful change both culturally and financially. This challenge has been caused by an overemphasis on scoring, survey fatigue, and the difficulties inherent in modeling emotional brand affinity as a bank of survey questions.
The closing keynote speaker, Forrester analyst, Anjali Lai, highlighted this difficulty and challenged the industry to refocus on business impact. The key to business impact lies in harnessing the emotional power of the customer story. “Emotion directly impacts a company’s bottom line and has the strongest impact on loyalty. Brands that are the fastest growing and have the highest loyalty also have the best emotional intensity with customers,” said Lai. The story of the customer journey—and how they emotionally feel about that story—lies in unstructured data such as customer comments. While most CX practitioners are aware of the power of unstructured data, not very many are using it effectively. Lai added, “Two thirds of CX professionals collect unstructured data, but only half of them actually use it.”
The adoption of customer narratives to understand the journey represents a massive opportunity for businesses to drive impact. Forrester Research found that 40% of customers choose brands because of relevant, resonant experiences, 13% actually pay more for brands that deliver personal experiences, and 24% of customers will recommend brands because they resonate emotionally. Additionally, they found that emotions like surprise, gratitude, and appreciation result in much higher spend, increased repeat business, and dramatic improvements in brand affinity.
While key themes such as advanced analytics, customer journeys, and deep learning permeated the 2016 Sentiment Symposium, the true star of the show was driving business impact by leveraging the emotional power of the customer story. Anjali Lai summed it up best by saying, “Based on [Forrester’s] research, emotion drives loyalty, and loyalty drives business.”