An estimated return on investment of 524% emerged among enterprise users who triggered at least 101 monthly queries, showing how direct software utilization translates into operational savings. When Japanese life insurer Sumitomo Life began testing generative models, the deployment faced the common friction point of regulated finance: navigating security demands without stalling software inside closed internal experiments.

Corporate generative AI adoption yields measurable economic returns when corporate leadership embeds the software into operational processes rather than leaving it as an optional staff experiment. Moving past open-ended pilots requires organizations to align feature controls with strict compliance rules, turning informal digital experiments into standard desktop workflows.

Governance and Initial Pilots

Sumitomo Life began using generative AI at work in 2023, initially watching employees run isolated text questions in chat interfaces. In planning a broader rollout to thousands of employees, the insurer weighed output quality, how well ChatGPT fit different kinds of work, and the security requirements of a financial institution. In 2025, about 100 to 150 people, including executives and planning teams, tested ChatGPT Business before the company moved to ChatGPT Enterprise in June 2026.

To meet compliance demands, Sumitomo Life worked directly with OpenAI to assess ChatGPT Enterprise against the insurer’s security standards. That review established how managers control workspace access and permissions, supported by OpenAI’s default policy of excluding business data from model training.

"When senior leaders don’t understand a new technology, it can be hard to get it off the ground," says Kakeru Tateyama, a senior member of Sumitomo Life’s planning department. "Here, our executives and managers were among the first to use it. They saw for themselves what it could do. I think that made it easier for everyone else to start using it."

Sumitomo Life’s president took up the tool personally during the ChatGPT Business pilot, regularly pressing staff to apply it across management meetings and on the internal corporate portal. As Kakeru Tateyama observed, this managerial adoption removed operational inertia, allowing business units to integrate workspace tools into routine analytical work.

Core Operations and Measured Returns

Adoption shifted away from casual chat toward structured operational tasks across multiple departments. Planning teams used ChatGPT Enterprise to research overseas insurers, evaluate domestic competitors, and assemble management proposals, while investment teams used it to draft code for risk analysis and return forecasts. Teams in accounting and finance deployed the model for research and analysis, while others linked custom instructions and files inside dedicated projects.

Measuring this activity across quarterly employee surveys, Sumitomo Life recorded a weekly usage rate exceeding 80% for ChatGPT Enterprise. The overall estimated ROI across the enterprise reached 203%, calculated from employee-reported time savings.

"If we take our existing ways of working for granted, we won’t get the full benefit of AI," says Masakazu Obi, a deputy general manager in the planning department. "The question is whether we can change the processes themselves. As we introduce AI, we’re also working to change those processes over the long term."

The financial yield scaled directly with query frequency: estimated ROI averaged 55% for 8 to 15 uses per month, rose to 231% for 16 to 30 uses, and hit 524% for users recording 101 or more monthly interactions. As the company prepares to expand access to additional staff in the second half of fiscal 2026, the reliance on self-reported time savings leaves open whether these hours convert into realized balance-sheet gains or remain internal process efficiencies.

Self-reported productivity gains provide a directional baseline, but the actual balance-sheet return depends entirely on whether saved analytical hours translate into lower operational headcount or higher corporate output.

What it means

A threshold of 101 monthly prompts turns an enterprise AI license into an actual operational tool rather than shelfware. When teams merely toy with casual chats, the yield barely justifies the seat cost; meaningful financial return appears only after routine research moves entirely into structured workspaces. Enterprise artificial intelligence pays off only when top leadership actively forces the entire operational routine into the model.

Our stories: AI in Corporate Reporting: The Risks of ChatGPT Templates (April 21); Oracle and OpenAI Bring Enterprise Data Into ChatGPT (October 27).

Generative AIAI in BusinessAI in FinanceProductivityOpenAI