The Complexity of Using Voice Recognition Technology to Measure Customer Satisfaction
As more people order things online, many may have the experience of calling a company's call center. Since a call center directly handles customer inquiries, the quality of customer care often influences a caller's impression of the company. Therefore, for companies and banks, the importance of call center representative training is increasing, in order to improve call quality. Furthermore, companies and banks are expected to extract customer needs from such inquiries.
Traditionally, the quality of call center customer care was assessed only for the entire call center through surveys. Detailed reviewing of individual conversations was therefore difficult. Although individual calls could be recorded and transcribed using voice recognition technology for the purpose of understanding customers' emotions, doing so accurately would be difficult because actual conversations do not necessarily have to be grammatically correct and the recording may include background noise.
Furthermore, people could be perceived as being satisfied or dissatisfied depending on their accent or intonation, even though the words they say are exactly the same. Then, in order to accurately understand customers' emotions, is there any way to analyze not only the words themselves but also how these words are delivered?
Quantification and Machine Learning of Cheerfulness of Voice and Satisfaction Level Allow Automatic Call Quality Assessment
Fujitsu Laboratories has developed technology to automatically identify when people feel satisfied or dissatisfied during a conversation based on their delivery.
In general, a “cheerful voice” is a high-pitch voice, and the pitch undergoes characteristic changes at the beginning and end of a short passage. Paying close attention to these changes, Fujitsu Laboratories quantified the cheerfulness of a voice with high accuracy by analyzing its pitch, changes in pitch and also by identifying unique pitch changes over multiple words. It also successfully quantified speaker satisfaction level during a conversation using the quantified cheerfulness of voice based on the fact that cheerfulness of voice and feelings of satisfaction are strongly correlated. Machine learning of this quantified data, call center interaction surveys, as well as satisfaction and dissatisfaction thresholds will lead to automatic identification of how satisfied the speaker is feeling during a conversation.
For Achievement of High Customer Satisfaction in Various Business Fields
In field trials, this technology pinpointed where callers were satisfied or unsatisfied during a conversation with a call center representative. It could accomplish this with approximately 70% accuracy compared to a human assessment *.
This will allow call center representatives in training to more easily understand which parts of their calls were good or bad. They will also be able to improve their calling skills efficiently. Also, research has confirmed that the more objective assessment results become, the more call center representatives accept the results. Furthermore, assessment criteria can be customized for each site based on the call assessment results.
Fujitsu Laboratories will apply this technology to bank teller conversations and interactive voice response services to flexibly handle customer complaints. In marketing analysis, it will identify features that customers are highly satisfied with. As a first step, this technology will be commercialized at the end of FY2016 as a Fujitsu Limited and Fujitsu FSAS call center service.
During FY2017, this technology will be made into an API ** by the Fujitsu AI solution Zinrai Platform Service. Using this API, Fujitsu aims to provide services that can achieve high customer satisfaction at many places such as banks, retailers, and medical institutions in FY2018.
*: Conducted at three Fujitsu Limited and Fujitsu FSAS call centers.
**: Application Programming Interface