Hitachi’s AI Leader On Adapting Industrial AI In Your Factory

Editor’s note: Chenan Gupta will be a featured speaker at our 2025 Manufacturing Leadership Summit in Cleveland, May 7–8, discussing how you can shape the future of AI-driven innovation in your operations. Please join us > 

Talk to a roomful of manufacturing CEOs about generative AI and you’re likely to hear—as I recently did during one of our Chief Executive Network gatherings—that when it comes to real adaptation and real ROI, they’re hitting a wall.

Sure, their back-office folks—especially in marketing—love it. And it holds promise for projects like creating faster specs or accelerating the bidding process (no small thing to be sure). But they’re having trouble moving past those somewhat obvious entry points to shop floor operations, and real value creation.

If that’s you, says Chetan Gupta, general manager of Hitachi’s Advanced AI Research Center, and leader of the $70.5 billion conglomerate’s Global AI Center of Excellence, you’re hardly alone. Gupta is a leading expert in industrial AI, driving Hitachi’s AI initiatives across research, enterprise and industrial applications and he has plenty to say about the dos and don’ts of the journey.

“What I find is that there is awareness now,” he told Chief Executive in a recent conversation ahead of the Manufacturing Leadership Summit. “What people are asking is: where is the ROI? If I have $100 to invest, what should be my priority be?”

Gupta has a ton of experience helping industrial leaders figure that out through his work within the Hitachi family of companies. I asked him to share what he’s learned, and give those of us who aren’t inside a company with the scale and resources of his firm some tips on how to approach AI more successfully. He did. What follows was edited for length and clarity.

What are people asking you about industrial AI and what’s your sense of where we are with industrial AI right now?

Hitachi is a conglomerate; we have multiple businesses. The electricity grid, a lot of transfer manufacturing, we make rail cars, locomotives. We also make a lot of instrumentation equipment, proton beam therapy. Different businesses are at different points in their AI journey. Some are more sophisticated compared to the others.

In some cases, leadership is trying AI use cases and they’re comfortable with it. In some cases, they’re dipping their toe in the water. It’s all over the place. But Hitachi has been investing in AI for some time. In general, what I find is that there is awareness now. What people are asking is: Where is the ROI? If I have $100 to invest, what should be my priority be?

For us, AI use cases fall into four categories, roughly speaking. One is AI for the enterprise itself, the back-office support processes. Whether you’re a manufacturer or you are Google, you still have to have HR, you have to have finance, you have to have legal. So how can you use AI to make those support processes more efficient? That’s one category I put stuff in.

The other category I put stuff is in for the core processes itself. In the manufacturing value chain, where does AI play a role? That would be the second aspect of AI.

The third and the fourth might not be as relevant because if you think from an industrial standpoint, there are both OEMs and operators. So, for example, in the space of fleets, DTNA (Daimler Truck North America) or Freightliner would be the OEM and something like Penske is an operator. The way they would use AI would be different.

So the other two categories for us become: Can we introduce AI to our production offerings to benefit the operator as well? And the fourth one is: Can we introduce brand-new offerings as services with AI?

That’s how we think systematically about AI.

How do you use a matrix like this to assess the potential for AI in an organization? How have you done that at Hitachi? And I guess another question along this line is: How do you coach the various parts of the business into using AI more effectively, more impactfully?

There is no way to right off the bat say which particular use case would be the most effective. It depends upon the data maturity and organizational maturity of every single entity.

What generative AI has done is made it very easy to use AI, especially for back-office work. So maybe I can use AI as a chatbot for my HR or for my finance or create marketing material. That seems to be the lowest-hanging fruit. That’s where I think the manufacturing companies can start.

But the ROI is not the most there. Most ROI will be where they can transform the manufacturing process itself with AI. That would be things around frontline workers. Can I make my frontline worker, the person who is maintaining my company and operating my factory, much more efficient? That’s where the significant ROI will be.

There are multiple ways to look at it. One is from a process viewpoint. If you look at the manufacturing value chain from design, to engineering, to manufacturing, to maintenance—both of the equipment for manufacturing as well as maintenance of what you manufacture—to support in the end for the product that you’re manufacturing, operations, and then quality, each of these horizontal functions in manufacturing will benefit from AI. You would look at these individual use cases and say, “Can I improve my process for quality improvement, for maintenance or operations for design?”

Another way to look at it is from a human-centric viewpoint, which is: I have frontline workers, I have workers who maintain my equipment, who run my factory, who design my processes or design the parts that I’m going to manufacture. Can I help them be better with AI?

If you want to start thinking of a problem from that standpoint, that’s where the most ROI will be. Where you go in this journey depends on: What’s the pain point for a particular manufacturer? For some manufacturers, supply chain becomes the most important thing. For others, it might be the equipment breaking down. Or quality. So, you could look at each of these horizontal [areas] and then figure out where the most value for you could be.

How do you start to pull together to do that kind of analysis? How do you coach others in the business to do that, to really find the pain points and then think about how AI might be applied there to break friction and change process? How did you guys do it?

I think the way to do that, at least in my experience, is to have three parties in the room. One is the business owner because they understand the commercial pain point. As an AI person, I might say, “Look, I know how to reduce the number of breakdowns of an equipment.” But they might say, “That’s not really my pain point. My manufacturing equipment works really well. My problem is how to do inventory management.” So the business owners need to be in the room because they understand where their pain point is, what the ROI for them is.

The next set of people should be the IT teams of that particular organization. I might want to do inventory management, but I have not created any data on my inventory. That might not be amenable to working with AI right away.

The third is the AI people themselves because once you know the business problem, once you know you have the right data, then they can convert that into a math problem that they can solve with the data. Once you have all three agreeing on something, then I think your chances of success are much, much higher.

If you’re a mid-sized company, this sounds more like a “buy” than a “build” kind of an opportunity. If you don’t have your own in-house AI team, are there more and more vendors out there who are able to work on these discrete problems?

Yes. There are many.

Ok, so help us with this. When we’re analyzing vendors, looking at outside consultants, any tips, any questions we should be thinking about as we start to engage?

First, I’ll put in a plug for Hitachi, but, obviously, you don’t have to go with us all the time.

What you should look at is whether they understand the industrial space or not, whether they have domain experts in the room or not. If you have some of the brightest engineers from Google looking at your problem, if they’ve never worked with shop floor data or with manufacturing data, then the likelihood of success goes down. You should look for expertise, not just in AI, but also the domain that you’re working with. That helps with translating a lot of things.

People in the AI world can be quite abstract in the way they look at the world. Manufacturing lives in the physical world. There should be some domain expertise now to translate one from the other, I would say.

That’s how you should look at the vendor: The right expertise, the right history and whether they have already deployed something in the field. Especially in AI, it’s one thing to do a [proof of concept] and another thing altogether to take a solution all the way to production and then operationalization and deployment. A lot of success and failure depends on that process. Your partners ideally should have dug in all the way to successful deployment of the solution and not just the POCs in the lab.

What’s the key thought that you would give the CEO at a mid-sized manufacturer trying to get through the hype and really put AI to use in their facility in 2025?

I would say two things. The first one, to paraphrase Bill Gates, is that people are overestimating the impact of AI in the short term and underestimating the impact of AI in the long term. What that really means is that your business will get impacted by AI, so you need to really think about it. At the same time, you don’t need to be caught up in every current fad or every current section of the story. Keeping that in mind should help you design a strategy around AI. Take a pause. Work with people who understand this space to see where the most impact of AI can be in your particular business, and then engage with one project and succeed in that. Build on that success, understand the lessons learned and then use the lessons learned in the next POC, in the next project and so on and so forth. You don’t have to boil the ocean. Start small.