Persol Business Process Design Conducts Demonstration Experiment at Yokosuka City Accounting Division to Visualize Employees' Tacit Knowledge Using Call Recording Data

Persol Business Process Design Co., Ltd. (Headquarters: Minato-ku, Tokyo; President and CEO: Kazuyuki Ichimura; hereinafter "Persol Business Process Design"), a member of the Persol Group which champions the vision "Work, and Smile," conducted a demonstration experiment in Yokosuka City, Kanagawa Prefecture (Mayor: Katsuaki Kamiji) to visualize the tacit knowledge*1 of employees handling inquiries within the city hall. This demonstration aimed to create knowledge data from call recording data and promote the efficiency of internal administrative tasks. *1 Know-how acquired by employees through experience, veteran employees' intuition, judgment points, and undocumented tips and insights regarding work. ■ Background As labor shortages become a social issue, the needs of residents in municipalities nationwide are also diversifying. The challenge is how efficiently operations can be executed and administrative services provided with limited resources. Improving the efficiency of administrative work is particularly urgent, and a growing number of municipalities are tackling this issue by promoting DX, such as utilizing generative AI. In 2023, Yokosuka City became the first municipality in the country to implement ChatGPT, pioneering DX initiatives nationwide. From the beginning, they have been working on learning from existing manuals, such as procedure documents, with the aim of streamlining internal city hall tasks using ChatGPT. Under these circumstances, Persol Business Process Design, which was supporting Yokosuka City's General Affairs Operations Center at the time, possessed successful case studies of integrating tacit knowledge from contact centers into AI. Consequently, Yokosuka City consulted them regarding the use of generative AI, and Persol proposed that it is crucial to use tacit knowledge as training data. Yokosuka City agreed with the approach of visualizing employee tacit knowledge, and discussions began on how to realize this. Therefore, starting with the Accounting Division, where creating a knowledge base was difficult due to the high volume of inquiries and the need for specialized knowledge, a demonstration experiment was conducted. This involved transcribing recorded data of inquiry calls from within the city hall and accumulating it as knowledge data (knowledge creation) to solve the problem. ■ Overview and Results of the Demonstration 1. Collection and Transcription of Call Recording Data, and Generation of Knowledge Drafts Content: Simple call recording devices were connected to employees' phones to collect voice data when employees handled calls (tacit knowledge such as phrasing during inquiries, and the categorization of questions and answers). The recorded audio was transcribed using generative AI. Results: Total recordings: 1,427 Valid recordings for knowledge creation: 870 2. Analysis and Structuring using the KCS*2 Methodology Content: Using the transcribed call data, inquiry responses and knowledge management processes were simulated manually to conduct knowledge mining (new registration, reuse, update). Results: Knowledge created: 499 items (Knowledge creation rate against valid recordings: 57%) Of these, "High-frequency use knowledge (frequently asked content)": 156 items - Visualization of the Accounting Division's phone operations (incoming/outgoing, by department, response time, content, frequency of occurrence) - The inquirer's experience (background of inquiry, question, situation) - The tacit knowledge of the Accounting Division staff (judgment points, categorization, resolution methods) Some of the "High-frequency use knowledge" generated in this experiment was converted into FAQs and integrated into the chatbot deployed for Yokosuka City employees. *2 KCS (Knowledge Centered Service): A mechanism where operators structure customer experiences and interactions (knowledge) obtained in contact centers, reflect and accumulate them in a database as data, and reuse them to improve customer satisfaction and operational efficiency. Demonstration period: October 14, 2025 - February 16, 2026 Call recording data: 1,427 cases Recording period: 2 months (40 days), Target phones: 3 Mined knowledge: 499 items High-frequency use knowledge: 156 items Low-frequency knowledge: 343 items ■ Persol Business Process Design's Strengths in Knowledge Creation In creating knowledge from inquiry responses, the inquiry content and the knowledge do not always match on a 1-to-1 basis. Responders address various patterns of inquiry content from different angles to connect the inquirer with the knowledge they truly seek. The ability to build a database that includes the knowledge possessed by these responders is our strength.