More and more companies are promoting DX in their contact centers to address issues such as human resource shortages, operational efficiency improvement, cost reduction, and UX improvement. As a manufacturer of computer peripherals, Buffalo Inc. has always been proactive in adopting new technologies to better connect with its customers. The following presents the company’s efforts to realize highly accurate customer self-solving by utilizing natural language AI search in collaboration with BELLSYSTEM24.

As a manufacturer of thousands of products, Buffalo Inc. (hereinafter referred to as “Buffalo”) has received numerous inquiries and has always worked to improve its customer support. The first measure they took to reduce phone inquiries was the distribution of FAQ videos. The number of phone inquiries dropped from 100,000 per month to 70,000 per month, and they perceived a positive response. However, the number eventually leveled off, so they implemented measures to redirect customers to LINE chat, and then the number of phone inquiries dropped to about 50,000 per month. They have also implemented various other measures, including the creation of new social networking channels. However, while the number of phone inquiries was being reduced, email inquiries remained untouched. At the same time that they began to consider measures to reduce email inquiries, BELLSYSTEM24 proposed a new solution called a natural language AI search engine. “We have a long relationship with the sales representatives of BELLSYSTEM24, and we always ask them to introduce us to new services, solutions, and technologies. When I heard about the AI search engine, I was intrigued by the fact that it understands natural language and can directly provide the correct answer to a customer’s question,” said Mr. Shimada. Natural language refers to the language we speak and write in everyday life and is said to be more ambiguous than programming language. A major difference between scenario-based AI chatbots, which have been common to date, and natural language processing AI is that the former narrows down answers by letting the user choose from multiple options, while the latter can provide answers directly to natural text questions entered in the search window of a web inquiry form.

"To be honest, at first, we were optimistic that it would be easy to implement and would produce immediate results. However, in incorporating this AI search engine into Buffalo’s website, some challenges arose. There was a problem with our company’s website, so I would have negotiated with the web production company, but the sales representative of BELLSYSTEM24 communicated with the web production company and solved the problem. Since they also confirmed and reported back to us, we were able to leave it to them with peace of mind, and we were able to significantly reduce the man-hours required for implementation. All of our various feature requests have also been considered. For example, we were initially told that it would be technically difficult to meet our request that the pop-up window pops up immediately after being filled in so that the customer can see it smoothly, but they persisted in finding a solution even after the implementation, and now they have achieved it,” said Mr. Shimada.

Although the AI search engine was introduced with great expectations, it did not immediately show its power after its implementation. “Immediately after implementation, the correct response rate was 40%, which was lower than when AI was introduced in chat. I was surprised as to why the correct response rate was so low, but when I heard the reason, I understood,” Mr. Shimada said. For natural language AI to be able to understand “ambiguity”, it needs to “learn.” So we prepared training data and performed appropriate tuning and maintenance to improve the accuracy of answers. As a result, the correct response rate is now in the 90% range.
“The scenario-based chatbot also shows almost the same correct response rate of 90%, but natural language AI has by far the shortest process to reach a correct answer. Considering this, the natural language AI search engine is better and more efficient,” said Mr. Shimada.
There are two main reasons behind this success.
The first is that BELLSYSTEM24 was able to work together with Buffalo’s web team and the site management company to create an optimal user experience using natural language processing AI. The other is that we were able to tune the AI in-house. By training web staff on the spot, where they usually handle customer service and operations, while using our AI knowledge maintenance system, we were able to perform tuning to “appropriately solve customer inquiries” in a short cycle. In fact, when the operation team tuned the AI at a rate of about once a week, the correct response rate increased to the 90% range in less than a year.

Not only did the correct response rate increase, but the initially expected effect of reducing email inquiries was also significant. While the number of inquiries using natural language AI search engines doubled, the number of email inquiries per day dropped to less than half. ""After the implementation, we saw an increase in responses by the AI search engine and a decrease in email inquiries. Not only has our goal of reducing email inquiries and facilitating customer self-solving been achieved, but the high correct response rate has also enabled customers to get answers faster. Since everything is completed with AI, we can respond 24 hours a day, 365 days a year, which we believe has greatly improved the customer experience,” said Mr. Shimada.

“After using the natural language AI search engine, we are very satisfied with its accuracy as well. We are also considering future expansion, such as using this AI for other solutions,” said Mr. Shimada about the future prospects of Buffalo, which has acquired new functions. Through the advanced Buffalo and BELLSYSTEM24 initiatives, customer support will continue to evolve in the future.