Credit Saison calculates 954% ROI from a ChatGPT Enterprise pilot with 315 employees

Credit Saison, one of Japan's large credit operators with roughly 31.16 million cardholders, has published a case study with OpenAI about its ChatGPT Enterprise pilot. According to the report, the pilot paid off at 954% (ROI) against an internal target of 500%, and each participant saved about 14 hours a month against a plan of 7.5. A month after the experiment ended, the company decided to roll the service out to all employees, and in September 2025 everyone got access.

This is a vendor case study on OpenAI's website, so it should be read with a grain of salt: the company supplies the numbers itself, without describing its methodology.

How the pilot worked and what it showed

The pilot ran for a little over a month, from late June to early August 2025. It involved 315 people, slightly under 10% of the company's roughly 3,700 employees in Japan. The group was mixed: volunteers, people appointed by the project team, all executives up to and including the president, and representatives from every department. Many Credit Saison employees work in generalist roles, and the company wanted to test whether AI helps with ordinary day-to-day work, not just with specific technical tasks. Kazutoshi Ono, Chief Digital Officer and CTO, summed up the approach this way: "We won't know until we actually use it."

Results for the 315 participants:

• ROI of 954% against a target of 500%;
• 14 hours saved per person per month, about 169 hours a year, against a plan of 7.5 hours;
• 75% of managers and 70% of rank-and-file employees said in a survey that they could not do without ChatGPT Enterprise;
• about 80% of surveyed legal department staff said they had become better at their work, and more than 80% of IT specialists reported greater job satisfaction.

The basic arithmetic: 315 people at 14 hours each come to about 4410 hours a month. The case study does not give the license cost, so it is impossible to check how those hours produced exactly 954%. All we know is that an ROI of 954% means a return of roughly 10.5 times the costs. In addition, the pilot included volunteers and employees picked by the project team, so the sample is clearly more motivated than the average workforce. And the hours saved, judging by the description, were estimated by the employees themselves.

What worked: managers and custom GPTs

The AI center of excellence ran a dedicated Slack channel where employees openly shared successes and failures. OpenAI trained managers and ran hands-on workshops for beginners. The most notable case was a custom GPT that simulates a shareholder meeting in voice mode. Senior executives took an interest, requests for the prompt followed, and that pushed wider use of voice mode.

Ono insisted that simply urging employees to use AI is not enough: managers have to use it regularly themselves. President and COO Katsumi Mizuno built his own GPT for researching companies and made it part of his workday.

Three use cases from the case study:

• Retail sales. A custom GPT takes the audio of the daily shift debrief in shopping malls and compiles the results for each employee, notes, next-day handoffs and priorities on its own. The meeting format does not change, so the barrier to entry is low. The tool now runs at all locations.
• Corporate sales. A GPT for client research identifies business problems and places where Credit Saison's services could help. In one analysis, it named problems that the employees themselves had not fully recognized.
• Internal sales. The AI plays a prospective client, while newcomers rehearse the conversation by voice and get feedback. Experienced colleagues do not have to step away from selling.

Codex, three levels of AX and limitations

The company splits its AI transformation (AX) into three tracks. Employee AX changes how employees work. This is also where, according to Credit Saison, engineers save an average of 850 hours a month with Codex. That is a separate figure for the engineering team, and the case study gives neither the team's size nor how the number was calculated. Customer AX covers the customer experience: the company is building its own AI call center. Corporate AX, meaning a rethink of the org structure and role distribution with AI in mind, is still at an early stage. Credit Saison is also considering ChatGPT Work.

The company separately addresses security. Once AI starts performing tasks autonomously, protecting customer data will become a mandatory requirement, and the approach will have to be even more cautious.

For teams planning similar rollouts, the case study offers three takeaways. First: a pilot should include managers, not just enthusiasts. Second: the most durable use cases are simple and built into an existing ritual, like the shift debrief. Third: define up front how time savings and costs will be calculated. Then figures like 954% can be verified, not just quoted.