Honeywell transforms with gen AI

Having spent years transforming into a fully data-driven enterprise, American conglomerate Honeywell is now poised to leverage gen AI across every facet of its business to drive improved productivity, collaboration, and innovation.

The Fortune 500 company, which primarily operates in aerospace, building automation, industrial automation, and energy and sustainability solutions, uses Microsoft 365 Copilot to aggregate and summarize content, GitHub Copilot for code generation and software modernization, and is testing Microsoft 365 Copilot for sales and Finance with Salesforce and SAP integrations.

“This is a disruptive technology that’s fundamentally going to change how we work, live, and play,” says Sheila Jordan, the company’s SVP and chief digital technology officer.

In all, the company now has more than 16 gen AI use cases in production, available to all its 95,000 employees. An AI copilot from Moveworks has also helped Honeywell automate IT help desk requests, reducing the number of tickets humans have to address by 80%. Another virtual assistant, Red, leverages Honeywell’s archive of 350,000 pages of product manuals and more than 50,000 internal articles to answer questions.

Jordan, who came aboard five years ago, explains that Honeywell has been on a massive, years-long transformational journey, from its infrastructure and network, to everything above it. So her role encompasses applications, infrastructure, and all data except for go-to market, which is on the engineering side and is looked after by her colleague Suresh Venkatarayalu, SVP and chief technology and innovation officer. The major thrust of the early part of Jordan’s tenure was getting Honeywell’s house in order in regard to data.

“We had 4,500 applications when I joined and now, we’re down to a little over 1,000,” she says. “We took all those fragment applications and moved them into the strategic platform. We bet hard on the enterprise data warehouse.”

Honeywell uses Snowflake for its enterprise data warehouse (EDW), and Jordan says it holds everything: bookings, billings, backlog, inventory. “We run the company now as a data-driven enterprise,” she says.

Working to strengths

With the EDW in place, the next step on Honeywell’s journey is to streamline end-to-end processes. Jordan wants to facilitate a seamless work experience between, for instance, sales and marketing teams, or engineering and manufacturing teams. A big element of that is reducing or eliminating administrative work and repetitive tasks that don’t require much in the way of skill or judgment but consume a lot of time from employees who have valuable skills.

“Our engineers don’t want to be typists; they want to be system engineers,” she says. “They want to think. They want to look at the code and evaluate what it can and can’t do. We want to take away the typist part, writing the code, but the reality is that reviewing the content, reviewing the code, making sure it makes sense, the value is just enormous.”

She notes that Honeywell is well-positioned to leverage gen AI because of the work it’s done on its data and data strategy. She cautions that her peers should be careful not to jump the gun here. The technology has many exciting applications, but a rock solid data strategy is an essential first step.

“You can’t have a gen AI strategy without a data strategy,” she says. “The good news for us is that while our data isn’t perfect, it’s in such great shape that we really do run the company with data-driven insights. Now we can layer gen AI on top of that and unleash new knowledge, new insights, and new opportunities.”

Putting the pieces in place

Honeywell is bullish enough on the transformational potential of gen AI that Jordan has appointed one of her reports as the program leader for it across the company. The program leader has established an ambassador within each of Honeywell’s functions and strategic business units, and those ambassadors, in turn, meet regularly with their functions and business units about potential use cases for gen AI. Then Jordan’s team meets every few weeks to discuss the potential use cases and prioritize ideas.

“They decide which ones make sense from a business perspective,” Jordan says. “In that same meeting, I have my technical experts who write the code and the large language models.”

The role of the technical experts is to advise the team on the feasibility of implementing the use case from a technical and time perspective, as well as seeing if the idea is even possible to begin with. At this stage, she needs projects that have real business value that can be delivered quickly with measurable results. There’s a temptation to go too big and get bogged down, or start too small with tactical things where the value rarely adds up.

“I need a flywheel effect and momentum,” Jordan says. “I don’t necessarily want to work on something that’s going to take years to implement.”

Anything associated with marketing, content, or project management is probably a good candidate, she says. For instance, project managers spend enormous amounts of time taking notes on actions and following up on those. Gen AI that automates those tasks could make project managers much more productive. Legal presents other exciting possibilities, where a copilot could extract critical data elements from contracts.

To support the company’s gen AI efforts, Jordan’s team has also created a gen AI academy to help train both technical and business employees. For the business employees, the idea, in part, is to help them better articulate their needs and ideas to the ambassadors.

“It’s really important you train the organization on what you should and shouldn’t do,” Jordan says. “We have this management operating system called the accelerator where you can get information and training on everything we’re doing. We use the gen AI academy to make sure we’re communicating with employees and training, and developing them.”

The value metric

As part of the use case identification process, Jordan’s team creates a value hypothesis. What value do they anticipate the project will deliver the business? The team then selects an appropriate technology for the use case and runs a proof-of-concept (POC) for a short time.

“Or, if it’s a large language model, we just go,” she says. “You know it’s not going to be perfect day one, so you iterate.”

Throughout the POC process and after the project is deployed in production, the team regularly tracks the results and compares them against the value hypothesis. Then they showcase those results to the business on a monthly basis.

“If you find it isn’t working, or you’ve got to force adoption or drive it hard, you’ve probably got the wrong use case,” Jordan says.