Credit Saison: ChatGPT Enterprise pilot delivered 954% ROI
Credit Saison ran a ChatGPT Enterprise pilot with 315 employees from late June to early August 2025. Reported ROI was 954% against a 500% target. A month after the pilot, the company rolled out the product to all employees.
On October 30, 2026, OpenAI published a case study on Credit Saison, a Japanese company with a credit card business (about 31.16 million cardholders) and financing and international operations. It concerns a ChatGPT Enterprise pilot (the corporate version of the chatbot with administration and data controls) that ran from late June to early August 2025.
The pilot involved 315 people — just over 8.5% of roughly 3,700 employees in Japan. The group included volunteers, people nominated by the project team, all executives including the president, and representatives from every department. According to the company, the pilot delivered 954% ROI and savings of 14 hours per employee per month, or about 168 hours per year. The original targets were more modest: 500% ROI and 7.5 hours of savings per person per month.
What sets this apart from a typical corporate deployment is the speed of decision-making and the breadth of participants. The pilot lasted about a month and a half, and the decision to deploy company-wide was made a month after it ended. In September 2025, ChatGPT Enterprise was given to all employees. According to OpenAI's account, the pilot group itself provided the impetus: executives and rank-and-file participants together turned the results into an argument for scaling. The motivation was practical: Credit Saison has many employees in generalist roles, and the company wanted to understand how AI would affect day-to-day work. Chief Digital and Technology Officer Kazutoshi Ono framed the approach this way: "We won't know until we actually start using it."
In practice, the case shows how companies measure AI impact: not just by the number of licenses, but by hours saved and return on investment relative to a pre-set benchmark. Separately, the materials mention time savings for the engineering team through Codex (a code generation tool), but the specific figure of hours per month is not given in the source data — there is no data.
What remains unknown: exactly how Credit Saison calculated ROI — which costs and benefits were included in the calculation — is not disclosed in the note. There is also no data on the cost of deploying to all employees or on whether the 14 hours per month in savings persisted after going beyond the pilot group.