【HEROZ Inc.】 Announcement of progress of “Next Generation Contact Center Project”
HEROZ Co., Ltd. Press release: October 7, 2024 Announcement of the progress of the “Next Generation Contact Center Project” ~Achieving operational efficiency and productivity improvement through collaboration between generative AI and humans~ NTT Marketing Act ProCX, Inc. (Headquarters: Miyakojima-ku, Osaka, President and CEO: Shinjiro Chotoku, hereinafter referred to as NTT Marketing Act ProCX) and U-NEXT HOLDINGS, Inc. (Headquarters: Shinagawa-ku, Tokyo, President and CEO: Yasuhide Uno) ) group companies, USEN NETWORKS Inc. (Headquarters: Shinagawa-ku, Tokyo, President: Kazuki Kanda, hereinafter referred to as USEN NETWORKS) and HEROZ Inc. (Headquarters: Minato-ku, Tokyo, Representative Directors Takahiro Hayashi, Tomohiro Takahashi, On May 30, 2023, HEROZ (hereinafter referred to as HEROZ) launched the “Next Generation Contact Center” project (hereinafter referred to as the “Project”) that utilizes generated AI and has been working on it. We are pleased to announce that we have achieved significant operational improvements in the first step, “Improving productivity” through enhanced operator support. 1. Background and overview of the “Next Generation Contact Center” project With the serious labor shortage becoming a common issue in the contact center industry, there is a need to recruit and train excellent operators and continuously provide services with high customer satisfaction. Aiming to transform the contact center industry to solve this issue, this project focuses on three themes: “improving productivity,” “improving management,” and “channel optimization” using generative AI.May 30, 2023 I’ve been working on it since day one.
Figure 2 Operator support tool introduction image (1) Realization of ATT reduction through response support function The “USEN Hikari plus” contact center operated by USEN NETWORKS receives many inquiries every day. It is not easy to provide correct answers to such a wide variety of inquiries, depending on the details of each customer’s contract, contract period, and services used. Even experienced operators have to check internal documents such as manuals and sometimes check with someone more knowledgeable, such as a supervisor, before responding, which can lead to longer response times and lower customer
satisfaction. there was. Therefore, in this project, we will use AI (ChatGPT 3.5 to 4o model ), we worked to realize a “response support function” that allows operators to quickly refer to the correct answer at any time. We completed the initial construction of the “response support function” in November 2023, but by working to continuously improve accuracy so that operators can use it in real time at the contact center site, the FAQ correct answer rate was initially 15%. Improved average operator response time from to 80%. Approximately 14% improvement I succeeded in doing so. (2) Realization of ACW reduction through response summary function The “USEN Hikari plus” contact center receives many calls every day, and we ensure that the details of the calls are accurately managed and a history of the calls is kept. For example, even if different operators receive a call from the same customer, it is possible to understand the previous response history and provide the optimal response. However, in order to input the response history, the operator had to remember the details of the response, create a summary, and input it into the system, which contributed to the long post-processing time after a customer response. Therefore, in this project, we are working to realize a “response summary function” that automatically converts the content of the response into audio text and extracts the main points of the text and important keywords from the content, which automatically summarizes the content according to the usage scene. Ta. At the beginning of this project, the summary accuracy was centered around general summaries, making it difficult to extract keywords and important items that were relevant to the business. This time, through tuning using the know-how of NTT Marketing Act ProCX, it has become possible to post the response summary results as they are with almost no corrections, which reduces the average post-processing time for operators. Approximately 74% improvement I succeeded in doing so.
Figure 3 Effect verification results (comparison before and after introduction) 3. Points of action (key success factors) The three key success factors that led to the above results are as follows. ■ Knowledge database renovation: In order to improve the accuracy of the generated AI’s answers, it is necessary to input the correct data correctly. Currently, AI does not have the application power that humans can use to make analogies, interpretations, and decisions, so it is important to provide data in a format that AI can understand. Therefore, in this project, various data held by USEN NETWORKS will be converted and corrected into a format that is easy for AI to understand using NTT Marketing Act ProCX and HEROZ (text conversion of original data, normalization of tables, unification of terminology, etc.) In addition to creating a glossary of technical terms, correcting abbreviations, organizing similar data into categories, and formatting documents so that subjects and predicates are clear, we also implemented various improvements such as controlling reference sources using prompts to improve accuracy. It has come true. ■ Evaluation of generated content: We repeatedly evaluated and improved the answers generated by the AI until the answer accuracy was truly usable by USEN Hikari plus operators, and we improved the answer accuracy. Specifically, in addition to visualizing “which in-house materials (in-house knowledge) the generated answer content was generated from,” we created a prompt to self-rate the accuracy of the generated content on a three-level scale, allowing for self-checks. We have introduced a system that allows us to make improvements as we go along. In the end, we were able to visualize the current score by installing a user evaluation check function by operators who are actually users, and worked to improve the score. ■ Expansion of answer areas: As a result of the above efforts, the answer accuracy of the generated AI exceeded 60%, and we aimed to further improve answer accuracy by adding the missing data (in-house knowledge) as new internal materials. Specifically, by analyzing the search trends of users at actual contact centers and adding to the knowledge that was missing, we succeeded in raising response accuracy to 80%. In the future, in addition to improvements through the continuous operation of the above initiatives, we will continue to add new functions and make detailed adjustments according to operations to develop the AI into a true support tool for operators, and expand the areas in which this AI can be used. By expanding this, we will strive to realize a “next-generation contact center.”
Figure 4 Tuning process for increasing the sophistication of knowledge search (Comment from the project representative) ■USEN NETWORKS Executive Vice President Toshimasa Yamada
It is a big step that our efforts to create a “next-generation contact center” have yielded results, and we would like to thank everyone who participated in this project. We will continue to strive to improve our operations in order to deliver the “satisfaction” that our customers need. ■HEROZ Representative Director and CEO Takahiro Hayashi
The collaboration between generated AI and humans has made it possible to operate a contact center with unprecedented sophistication. We will continue to pursue technological innovation and aim to provide services that exceed customer expectations. ■NTT Marketing Act ProCX Representative Director and President Shinjiro Chotoku
I would like to thank you from the bottom of my heart for the results we have achieved thanks to our efforts together. We will continue to strive to make even greater contributions through repeated
verification. 4. Future developments NTT Marketing Act ProCX, HEROZ, and USEN NETWORKS aim to further improve “productivity improvement”. Furthermore, in order to realize a “next-generation contact center,” we will develop, build, and operate solutions that deepen
collaboration between AI and humans for “channel optimization” and “advanced management,” and accelerate this project. We will strive to make this possible.
Figure 5 Future development image