More and more, contact centers need to analyze the sentiment of the interaction between the customer and the contact center agents to get an indication of a customer’s potential future behavior. This behavior includes canceling contracts, buying products or services, becoming an NPS-based promoter, etc. Measuring emotions in the right way, therefore, becomes very important. Our research and experience have shown that if you just relied on keywords as a measure for the emotion, it would be misleading.
What is True Emotion Analytics and how is it measured? Xdroid’s VoiceAnalytics uses machine learning and artificial intelligence to identify and measure true emotions. VoiceAnalytics analyses each speech block of a couple of seconds and assigns an emotion value to it whenever an emotion different from neutral is detected. VoiceAnalytics uses a Deep Neural Network to automatically extract feature sets for identifying emotions. The performance of each feature set is then carefully and thoroughly evaluated and yields a set of so-called “speech markers” that are used to qualify each speech block. When an emotion different from neutral is detected, the system puts the emotion into one of four possible categories (happy, disappointed, displeased, uncertain) and assigns a value to it. The larger the value the stronger the emotion. Thresholds can be defined to adapt to cultural differences in the way people express themselves. A weighted sum is taken over all the emotion values of every speech block of the conversation to yield a so-called overall Emotion Index (EI) for the complete call. The EI allows for a ranking of calls from e.g. a high to a low displeased emotion index, for comparing a call’s EI to the average value across all calls of the contact center and individual operator’s score to the group average, etc.
Sentiment analysis typically uses the meaning of words, terms, and expressions to determine the sentiment of interaction and limits itself to qualifying interactions as
either positive, neutral or negative. Often, sentiment detection and emotion detection are used interchangeably. However, emotion analysis goes well beyond just categorizing identifying a positive or negative sentiment to identify true emotions like happiness, unhappiness, uncertainty, disappointment. Certain nuances can be found in interactions and conclusions can be drawn from interactions that are different from the conclusions resulting from sentiment analysis only.
Sentiment analysis of “No thanks, I wouldn’t like to sign up for that this service” might result in a negative sentiment result because of “wouldn’t like” while it might have been expressed with uncertainty. Identifying that uncertainty creates an opportunity to still try and turn the negative of not signing up into a positive by taking away the last bit of doubt through a follow-
up interaction. Sentiment analysis of “Thanks, I really enjoyed your support” might be rated as positive sentiment because of “thanks” and “enjoyed” while it might have been expressed in an unhappy way. Identifying this sarcasm creates an opportunity for taking corrective actions towards that customer that otherwise might turn to competition.
Sentiment analysis of “I’m impressed with the way you fixed that problem, so I have to thank you for that” might be rated as a positive sentiment because of the “impressed” while the “I have to thank you for that“ could have been expressed in a disappointing way because the customer expected more than just having the problem fixed. Maybe it’s appropriate to offer that customer compensation on top of just having solved the problem.
Through the usage of our True Emotion Analytics module customers have found that the vast majority of “Thank you” expressions is rather an obligatory politeness statement than a sincere expression of real gratitude and positive feelings. The number of occurrences when “Thank you” was expressed with negative emotions (sarcasm) is even almost as high as 50% of the number of occurrences when “Thank you” was expressed with a positive emotion (sincere thanks).
Xdroid’s capability of identifying and quantifying happy, displeased, uncertain and disappointing moments in the call goes beyond spotting a (positive or negative) sentiment only. Combined with productivity and speech style analysis and keyword spotting it enables a new and innovative way to boost contact center efficiency.