It has been more than three and a half years since OpenAI’s ChatGPT sparked a surge in corporate interest in generative artificial intelligence tools designed to reshape work. Even so, Hitachi, which employs nearly 290,000 people globally, has still not rolled out a single enterprise-wide AI tool for all workers across the Japanese conglomerate.

That cautious approach may prove prudent for Hitachi, especially as research has shown that many enterprise AI pilots fail and debates over the cost of AI have intensified. For a company with operations spanning industrial equipment, software, energy, and medical systems, the challenge is not just finding useful AI, but deciding where it can be deployed safely across a sprawling business.

Bala Krishnapillai, senior vice president and chief information officer of Hitachi’s Americas division, says the company’s workforce is already using many AI tools. His internal adoption strategy is organized into three areas. The first covers everyday productivity tools such as Microsoft Copilot and Google Gemini, which are used to summarize emails and meeting notes and for translation. Translation is especially important for the 607 subsidiaries operating across 190 global markets.

Krishnapillai says his IT team works closely with business leaders to evaluate and approve job-specific tools. Those can include AI-enabled content creation for creative professionals or competitive analysis tools that help the sales team. The third focus area is developers and AI coding assistants, where Hitachi works closely with Anthropic.

“From the enterprise AI strategy standpoint, there is no one solution,” says Krishnapillai, who joined Hitachi in 2018 and has served as CIO of the Americas division since April 2025.

Although he encourages AI adoption across the company, Krishnapillai says “consuming of tokens is a big thing. We are putting in some controls.” In some departments where competitiveness is considered critical, such as research and development, there are no AI usage restrictions. More broadly, division managers are responsible for how their teams use AI and for tracking spending.

A major challenge for Krishnapillai has been data complexity. Founded in 1910 as a single mining machinery repair shop, Hitachi now operates businesses spanning rail systems, digital products, industrial machinery, power and renewable energy, and medical systems. Ranked No. 197 on the Fortune Global 500, Hitachi was long known to the general public for making and selling consumer electronics such as televisions and camcorders, a business it has fully exited as it shifted toward digital systems and services, which now account for 27% of revenue.

Hitachi generates $70 billion in annual revenue, a scale built in part through acquisitions. Among its larger recent deals were the $11 billion purchase of the power grids business from Swiss-based tech firm ABB and the $9.6 billion acquisition of U.S. software vendor GlobalLogic.

A century of dealmaking has left data spread across more than 150 customer relationship management systems, including Salesforce, SAP, and Microsoft. When marketing and sales teams worked on a new business proposal, producing an accurate analysis report could take weeks.

“We have massive data stored in our ecosystem,” Krishnapillai says. “It was unmanageable from an IT enterprise standpoint.”

To address that problem, Krishnapillai turned to enterprise software vendor Appian to connect data on top of the legacy infrastructure without migrating everything into a single database. He says the approach has given teams two main benefits: they can create new projects faster and draw sharper insights from Hitachi’s data ecosystem.

“When we create a proposal now, it becomes stronger, more compelling, and very competitive,” Krishnapillai says.

Hitachi and Appian say the new setup has produced a 40% efficiency gain for the sales and marketing team and cut operating costs by 20%.

Matt Calkins, CEO of Appian, says the data fabric will also make it easier for Hitachi to adopt agentic AI. He says autonomous agents can sometimes ask unexpected questions and search for unanticipated data sources across the business to reach their conclusions.

“Everybody’s got scattered data, and everybody needs to inform unpredictable agents,” Calkins says. “They’ll be like humans, exploring the enterprise and making decisions, and so they need to go places that you didn’t already orchestrate and expect.”

That is an optimistic view if everything goes right. Rogue AI agents—a report published this week said models built by Anthropic and OpenAI took unsanctioned actions—highlight the risks of this approach and the need for tight governance.

Krishnapillai says he is focused on establishing security protocols, data protection, and privacy before the company is fully ready to embrace autonomous tasks. He says Hitachi is actively speaking with vendors to find software that can monitor the creation of all AI agents, which would help avoid unnecessary duplication while also encouraging agents that can be used across functions throughout the business.

“We are in the early phase,” Krishnapillai says of Hitachi’s progress on agentic AI. “Our goal is to become autonomous in the future. But we are not there yet.”

John Kell

This story was originally featured on Fortune.com