Proposal for pilot operation of customer inquiry classification automation

Conclusion: Over the next two weeks, we will pilot automatic classification for only one type of delivery inquiry, and the response will be confirmed by a person in charge. 1. Problem to solve Currently, out of an average of 80 inquiries per day, 40 are delivery inquiries. Representatives spend an average of two minutes reading and triaging one inquiry. These figures are based on the work history provided and do not include time to write a response. 2. Compare alternatives Maintain Status Quo: There will be no additional cost, but you will retain 80 minutes per day of sorting delivery inquiries. Improved classification rules: Although simple to implement, queries with different wording may be missed. AI classification demonstration: Can respond to various expressions, but requires inspection for misclassification. Usage fees and setup work time need to be confirmed. 3. Recommended Action Plan First 2 days: Prepare classification criteria and answers using past inquiries with personal information removed. Days 3-7: Compare the AI's classification with the human operator's classification and record the error type. Days 8-10: Compare accuracy and actual work time to decide whether to scale up. 4. Judgment criteria We propose an accuracy of over 95% and a 30% reduction in personnel classification time as pilot goals. Goals are not actual achievements. If personal information is exposed or incorrect automated sending is discovered, we will stop immediately. 5. Request for approval We request a two-week trial targeting only delivery inquiries and designation of an operations person. Budget limits, tools used, and storage policies are confirmed separately before execution.