The following case study takes a look at how ESJ co. validated and then utilized emotion detection API to help Japanese contact centers gain previously-unavailable insight into their agent-client interactions, which effectively boosted customer satisfaction scores, reduced agent attrition and shortened call handle time.
The Story of CENTRIC and ESJ
Since its inception in 2009, CENTRIC Co. Ltd. has been a leading professional contact center services provider in Japan, offering solutions ranging from consultancy, through outsourced contact center operations, to e-commerce and employee retention. In its relentless pursuit after improving KPIs, CENTRIC became aware of a major blind-spot impairing contact centers’ ability to improve their performance stats – that blind-spot being the gap between customers’ expressed reactions and their actual level of satisfaction.
The team at CENTRIC realized that this blind-spot could only be resolved by finding a way to truly understand customers. CENTRIC launched an in-depth research initiative, comparing available sentiment analysis technologies.
Among those were solutions centered around expressed content analysis via Speech-To-Text, expressed intonation or “Voice Gesture” analysis, solutions utilizing CTI-related and CRM-related customer sentiment profiling – and of course, Emotion Logic Layered Voice Analysis.
The intensive research concluded that Emotion Logic LVA™ was by far the most accurate and informative comparedto other available technologies.
CENTRIC then doubled-down on its research efforts, partnering with Kumamoto University to further validate LVA’s accuracy.
Following the successful research and realizing the wide array of challenges that can be effectively addressed using Layered Voice Analysis, CENTRIC launched Emotional Signature Japan (ESJ Co.) in 2019 – a subsidiary company dedicated to the emotion analysis sphere.
ESJ was tasked with exploring the new possibilities presented by emerging emotion detection technologies with emphasis on utilizing emotion detection to optimize performance and augment the new generation of business processes. With CENTRIC’s blessing, ESJ chose Emotion Logic as its emotion detection technology provider, and has since become a strong proponent, developer, and distributor of LVA-based solutions in Japan.
Is it really working? Validating the technology
Genuine emotion detection is very hard to validate. Validation must be based on masses of real-life audio
data, that is also validated by other means. Simulated scenarios cannot be used, since participants in
simulated scenarios, by definition, will not experience the same emotional states as participants in real-life
situations.
Motivated to develop a disruptive and truly effective service, ESJ’s team picked up CENTRIC’S ongoing
research with the University of Kumamoto. The researchers analyzed a staggering 3M recordings of
customer-agent phone calls. These amounted to 1.6 billion data samples from over 500K cases,
meticulously analyzed over a period of three years. The research included a comparative analysis of LVA-detected agent churn risk, with perceived performance deterioration, self-reported dissatisfaction, and actual turnover. With over 1000 agents analyzed, LVA’s accurately predicted 80% of the outcomes. These high accuracy rates were also achieved during blind-testing calls, where call outcomes were not known to the researchers before analyses were conducted. The comprehensive research earned ESJ credibility with the Japanese government, universities, as well as with private institutions. During the research years, it was found that competing technologies utilizing AI. models trained on emotional expressions voiced by actors, produce systems that can detect acted emotions reasonably well, but fail in real life deployments –particularly in the Japanese culture. With the technology effectively validated and gaining credibility with academia, as well as with the Japanese government and private institutions – it was time for ESJ to introduce emotion detection insights into a large telecom company. The fact that LVA™ was both language-independent and intonation-agnostic, made it immediately field-ready without language or culture-related barriers.
“[In Japan] We are raised to interact respectfully, even when we’re angry or disappointed. As children, we expressed raw emotions, but social norms dictated restraint and decorum, transforming us into `suppression experts,’ which is why it’s so hard to identify the real emotions coming from our customers.”
Mutsumi Sugino, ESJ
The Challenge
The challenge of constantly improving service levels in the contact center space is shared across this global industry. Contact centers invest ample efforts in reviewing calls as a means to assess customer satisfaction, or indications of customer churn.
Similarly, contact centers also try to gauge and prevent agent attrition as they struggle to keep top-performing agents from quitting. ESJ found these challenges to be nearly unassailable in Japanese contact centers. Politeness and nondisclosure of personal feelings are among the main
characteristics of the Japanese culture. This cultural trait is clearly reflected in contact centers – customers rarely voice their dissatisfaction or provide discernable indications regarding their emotional reactions during calls. Similarly, agents, by training as well as by cultural inclination, are extremely polite and emotionally reserved, making it difficult for managers to identify agents who experience stress, anger, sadness and other burnout-inducing emotions.
This information deficit regarding actual customer satisfaction stats, as well as agents’ emotional aptitude and well-being, has been a major issue for Japanese contact centers.
Primarily because of discrepancies between reported customer satisfaction stats and actual customer churn rates. To put it plainly, satisfaction levels looked just fine – yet customers were being lost.
The Results
ESJ was commissioned by one of the top three telecom companies in Japan. The goal was to apply LVA-based analysis to 1,000 phone calls handled by the highest-ranking agents, where the customers were
determined to be “satisfied” by QA specialists. Scanning the calls for unspoken and unexpressed emotions, the gap between what customers and agents said and their actual feelings became
exceedingly clear. Layered Voice Analysis revealed that 70% of the calls featured indications of high levels of stress and aggression experienced by the customers, also indicating where in the calls these emotional states were most conspicuous.
Curiously, ESJ discovered that many of the customers seem to have experienced stress and aggression peaks right towards the end of calls. Investigating this phenomenon, it was found that agents were using a scripted sales pitch followed by a satisfaction questionnaire at the end of the call. This very process correlated perfectly as the cause of the customers’ churn- inducing yet unexpressed stress and aggression. Once the telecom company realized that their process was affecting clients negatively, they shortened the script by 30-60 seconds and eliminated the sales pitch. Consequently, while call handle time was reduced by over 5%, genuine customer satisfaction scores increased (client could not provide increase rates). Agent recruitment, training and retention incurs significant costs to any corporation. High agent attrition and dropout rates are a major concern, both from the operational and the financial standpoints. Upon understanding how LVA can reveal true feelings, the telecom company applied it to better understand their teams.
For the first time, management was able to understand their polite and emotionally-reserved employees
better, accurately gauging their level of contentment and emotional wellbeing during calls. Aware of the
times where agents experienced high stress, and even zeroing-in on the causes of such stress, they were
able to decrease churn & turnover rates by 7.2% during the first year following ESJ’s analysis, resulting in
significant cost savings on training and recruitment
“The gap between what we expect and what we actually see, was a huge challenge to begin with because companies just kind of neglected to accept the fact that their customers may be in stress. Once they had a better understanding about how we should use the solution as a new indicative information to better the operation it actually made a lot of gross changes”.
Mitsumi Sugino, ESJ
The Future
Following this first project, ESJ has successfully introduced Layered Voice Analysis into multiple contact centers, reproducing the positive outcomes, and slowly changing the paradigm regarding understanding genuine emotions in the contact center sphere.
Today, ESJ is leading numerous projects throughout Japan, serving a versatile clientele ranging between major contact centers, insurance companies, HR recruiting firms, system engineering companies and consultancy groups. The accumulated field experience and ever-expanding data, along with the ongoing research collaboration with the university of Kumamoto, encouraged ESJ to develop its own LVA-based platform using emotion detection API. In 2020, following an R&D process, ESJ has announced the launch of ESAS, an emotion detection solution based on Emotion Logic technology tuned for the Japanese contact centers’ space.
