CIS’s AI search service, Arrow Search, is used to streamline responses to internal inquiries.

【 Summary 】

Since its establishment in 1922, Fuji Baking Group has had the mission of “working hard to create delicious products of high quality and delivering them with sincerity to our customers” and has introduced many hit products loved by Japanese families.
To provide “quality and delicious products,” the Production Department has been consolidating various data and documents representing the background and history of the products into the Notes/Domino system. When revising the price of a product, it is necessary to review each raw material one by one, but since the designations differ from supplier to supplier, it was necessary to search using a variety of keywords. Therefore, knowledge and experience were required just to trace the relevant information, and the cost of searching to gather information on a vast amount of raw materials was an issue. While considering a solution to this problem, we learned that AI can absorb fluctuations in language, and we decided to introduce the system, thinking that if we let AI learn the trends of questions and answers and register various terms, phrases, and expressions as synonyms, we could successfully extract the knowledge of our predecessors while conducting “ambiguous searches.
We told CIS that we wanted to make it easier to provide feedback on search results by rating them on a scale of ○, △, △, ×, etc., and asked CIS to reorganize the system to make it easier to use. As a result, we found that it is relatively easy to improve search accuracy in the fields of general affairs and human resources because they use common Japanese terms. However, we also found that it is effective to prepare a dictionary of synonyms and synonyms for original terms and industry terms related to materials with business partners, for example.
By using this system, the company became aware of the need to search first and found that the number of inquiries was reduced and the time required to respond to individual inquiries was also shortened, leading to a reduction in workload by approximately 20%.