At technology and manufacturing company Honeywell, generative AI is everywhere.

“Every function and every strategic business unit is now using gen AI,” Sheila Jordan, the company’s chief digital technology officer, who oversees AI integration internally within the organization, told Fortune. “And the other thing I’m super proud of is that we have it available to all 100,000 employees,”

The company built its own “Honeywell GPT,” which helps employees draft and edit emails, summarize technical documents, translate content, and brainstorm ideas. Employees also use Red, a virtual assistant that serves as a central resource for accessing company information around IT, finance, HR, and the firm’s policy library. Engineers are coding with AI, and the company is reimagining its varied products and services with new generative AI–powered offerings. Overall, the company has 24 generative AI initiatives in production and 12 more on the way, compared with 16 a year ago.

As companies across different business sectors incorporate AI into their operations, an emerging set of best practices reveals a variety of approaches, from decentralized, experimentation-driven cultures to tightly choreographed strategies that can scale across an organization. Honeywell, which ranked at No. 17 in the Fortune AIQ 50 list of Fortune 500 companies with the most “mature” AI capabilities, is a case study in how to excel by taking the latter approach.

Jordan and CTO Suresh Venkatarayalu, who oversees AI product efforts, believe the company’s success in maturing its AI capabilities directly stems from its “six-chapter AI framework.” Along with the organization’s top-down approach to AI, adhering to the framework has allowed them to focus on efforts with immediate impact in order to kick off the flywheel effect.

“What are the use cases? And can I measure and track them?” said Venkatarayalu, describing how the company zeroes in on impact. “In fact, tomorrow we have a meeting with Sheila and the CFO looking at the 2026 road map and to ask me the real question: ‘Could we track it to the P&L?’ And we should track it to the P&L. That’s the way it’s set up.”

The six-point strategy

In the fast-moving world of AI, it can be difficult to prioritize, stay on track, and resist trying to do everything at once. That’s why Honeywell’s leadership created a six-chapter framework in early 2024 to guide the organization’s AI efforts and keep it focused strictly on use cases it believes will truly move the needle.

“We could get distracted by the long, long, long tail and all the noise and all the things people might want to do, but we have a whole program to prioritize those things that are going to move the needle in business value, both on productivity and growth and innovation,” said Jordan, adding that the organization “would have been confused and lost” without the framework and clarity from her and Venkatarayalu about which generative AI capabilities were fit for implementation.

The first chapter of the framework is about the tools—such as Red and Honeywell GPT—designed to assist employees in their everyday workflows. Then there’s chapter two, focused on the use of generative AI for engineering. Chapter three is how the firm “thinks about cognitive automation,” Jordan said, specifically how it’s using different LLMs (large language models) from Azure, Google, AWS, and others for specific use cases. Next, chapter four is all about generative AI in the commercial applications they purchase and use, like Salesforce and other platforms. Chapter five centers on the company’s own products and services. And lastly, chapter six focuses on sales effectiveness.

“I think our chapters will work for any enterprise,” said Venkatarayalu. “It’s productivity, it’s growth, and it’s margins.”

Chasing the flywheel effect

Jordan said the fact that the technology can be applied to so many use cases is one of the biggest challenges to overcome, so it helps to start with ones that have the biggest immediate impact. That way, those early successes can drive the effort forward.

For example, she said early work with GitHub and Copilot were the “first movers” and delivered the value they thought it would, which started the AI efforts off on a strong note.

“If it works, the flywheel takes off. If it doesn’t work, it dies its death, right? So I wanted the flywheel effect where we could do something and show the organization the value of gen AI,” she said.

This means going in with a business case and value proposition in mind, but being open to value coming through in a different way than assumed, she said.

“We could say [the value] was going to be productivity, but in reality, it was a sales effectiveness play. We got a higher conversion from something. So I would just say to stay super open to the business benefits, because they can morph based upon your customer and partner interactions,” Jordan added.

The top-down approach

Another key element to keeping the organization on target and adhering to its AI framework is its top-down approach.

Venkatarayalu pointed to how other companies start with a lot of proof of concepts, letting business units pursue their own strategies and democratizing the AI efforts. But not Honeywell, which he said is “predominantly top-down-driven” when it comes to AI.

“I think this company looks at use cases first, value second,” he said. “And once we believe—along with our CEO and chairman and the business unit leaders—[that a use case will deliver value], we drive that. I think that’s a very different [mindset] than many of my peers.”

Correction: A previous version of this article misstated the number of Honeywell’s business units.

‘Our chapters will work for any enterprise’: Honeywell’s AI chiefs share the strategies that helped the firm mature its AI efforts