JapanHealthcare and life sciences2024 – ongoing

Alfresa Holdings

Owl-One in-house AI assistant with search over internal documents

Alfresa's Owl-One assistant resolved about 60% of employee questions in its first months

WorkingVendor case studyVendor ↗

Problem

Employees at Alfresa spent about 35,000 hours a year on internal inquiries about regulations and operating manuals.

Approach

Alfresa built Owl-One on Azure OpenAI Service through the SoftBank Generative AI Package and launched it in April 2024. It handles summaries and drafting. It also searches internal documents such as regulations and manuals in PDF, Excel and Word, using retrieval-augmented generation. The target group was about 4,300 employees.

Results

約2,300employees who had used Owl-One by the end-of-August 2024 survey, out of a target group of about 4,300Vendor case study[1]

約1,200daily active users at the same pointVendor case study[1]

約60%share of questions resolved by the assistant's answersVendor case study[1]

約1,550時間estimated yearly reduction in inquiry-handling work, against about 35,000 hours a year beforeVendor case study[1]

Takeaway

Alfresa measured the assistant by the share of questions it resolved and the inquiry hours saved. It also gave a larger projected figure that includes output quality.

Sources

社内データをAIと連携させ、年間約6,000時間の生産性向上に成功
SoftBankOct 2024Vendor publication

Source summaries

[1]SoftBankOct 2024

SoftBank's customer story on Alfresa's Owl-One assistant. It gives usage counts as of late August 2024, a 60% question resolution rate and an estimated 1,550 hours a year saved on inquiries. It also states a total productivity gain of about 6,000 hours a year when output quality is included.

Full details are in the original sources.

Added 2 Nov 2026 · Updated 2 Nov 2026