Siemens' Chris Stevens on Adaptive Production, AI "Light Bulb Moments," and Why Digital Transformation Hinges on People
Manufacturing Happy Hour
In this episode of Manufacturing Happy Hour, host Chris Luecke talks with Chris Stevens, SVP and GM of the US Automation Business for Siemens Digital Industries. The conversation asks how manufacturers can build operations that flex with demand rather than run on rigid, fixed automation. Stevens argues that the answer combines software-style ways of working, contextualized data, AI, and digital twins, and that the whole effort still depends on bringing people along.
From Taking Humans Out of the Loop to Bringing Them Back
Stevens spent most of his two-decade-plus Siemens career on the software side, in the DISW organization, working on digital twins of both product and production. He moved to the automation group about two years ago. Asked about the most drastic change he has seen, he began with what has stayed the same. Twenty years ago the hot terms were lean manufacturing and continuous improvement. His work then centered on the virtual world: manufacturing bills of materials, bills of process, process optimization, and simulation. The digital twin concept from that era is still in use, and he noted that almost everyone uses the term now.
What has changed, in his account, is the goal of automation. Back then, the focus was on automating a station and taking humans out of the loop. Today the conversation is more about bringing humans back into the loop, and he credits AI with driving that shift. He tied it to the skilled workforce gap and cited studies estimating that 5 to 8 million manufacturing jobs will be available by 2030 that cannot be filled. His response to that gap is what he calls adaptive manufacturing. Systems are automated, but humans stay in the loop to validate when validation is needed, to apply their expertise, or to spread their skills across five stations instead of one.
Stevens also described a cycle of interest in autonomous manufacturing. It was a hot topic three or four years ago, both in automotive and in manufacturing generally, under the banner of "lights out" production. Interest then cooled. He sees it coming back, and pointed to a video he had recently watched about Tesla and Elon Musk's vision for robotics and autonomous manufacturing. He said Siemens is having more and more of these conversations in the semiconductor space. His summary of the arc was lean and continuous improvement focused on automating stations, then adaptive, and now a renewed question of what an autonomous stage might look like.
Moving at the Speed of Software
Luecke asked whether Stevens's software background gave him a different outlook as the industry moves toward software-defined manufacturing. Stevens said customers now understand how software works and want to "move at the speed of software": make a decision, click a button, and go.
He contrasted the engineering and IT world with the shop floor. On the engineering and IT side, companies have automated processes and QA, know how to replicate environments, build in redundancy, handle cybersecurity, and collaborate effectively, especially since COVID. Those practices have not yet translated to manufacturing. His example is PLC code. Today it often lives on one person's laptop and gets carried on a memory stick down to the PLC for an update. In the software-defined model, teams collaborate on the code on a shared platform and validate it through virtual commissioning. Once the code is confirmed good, they push it to the control layer with a button. Stevens said managing shop-floor assets also resonates with customers, including the code, the PLCs, and the various IPCs.
He acknowledged that gaps remain. Certain industries have requirements around high performance, safety, or motion that Siemens is still working to address "as fast as possible." He also observed that IT roles, which were typically absent from the shop floor, are now appearing there, and he read that as a sign that the two worlds are converging. For him personally, bringing those worlds together has been a lot of fun, because he spent years having these conversations outside manufacturing.
AI as a Thought Partner: Copilots Versus Agents
Stevens looks at AI through two lenses, internal and external. Internally, he said Siemens "can't go fast enough." He described a mindset shift that he believes few people have made yet. Most people use AI for small tasks like drafting emails or a white paper. He wants people to treat it as a creative thought leader sitting next to them. Some of his examples: simulating a customer meeting to get feedback on what to do differently, or preparing for a big decision by giving the AI proper context and asking it to pose three interview-style questions that help reach a conclusion.
He calls this kind of use "copilot," which is reactive: you ask, provide context, and get something back. Agents are proactive. Siemens is "placing some big bets" on agents to support staff who work directly with customers, so they have the right information at the right time.
When Luecke asked for a concrete agent example, Stevens joked that in Silicon Valley, if you throw a stone you'll hit a startup building agentic AI. He then described an agent tied to a customer account. The agent scours public information along with internal context such as emails, meetings, and chats. It then surfaces what is happening and suggests next steps. For instance, it might report that company XYZ just announced a greenfield facility with a $360 million investment. It could then suggest packaging material on plant floor layouts or cost estimation for that customer. In effect, the agent comes to you and says there is something to act on. Luecke summarized the distinction as copilots are reactive and agents are proactive.
The Light Bulb Moment: Predictive Maintenance at a CPG Company
Externally, Stevens said AI is "the hottest topic hands down." These conversations happen with presidents, C-level executives, and boards, not only with controls engineers. Leaders see AI as being as transformative as the internet, and he described their mood as "a race to the finish." Siemens tries to anchor these discussions in outcomes and use cases. Stevens said he is surprised how often material he considers fairly elementary produces light bulb moments. His team thought some of its messaging was six months stale and needed updating, but found it still resonates.
His most recent example came from a consumer packaged goods company and involved predictive maintenance. He noted that people understand AI from consumer use, like asking for travel ideas or what to cook with certain ingredients. What they often miss in the industrial context is that a model is only as good as what you feed it.
In this company, 50 machines worldwide were all maintained on the same schedule, whether a machine was 5 or 15 years old. Parts arrived at the same time and maintenance happened whether it was needed or not. That meant downtime plus investment in parts and labor. Stevens said changing this schedule was "taboo." Siemens showed that data could drive the scheduling instead, based on parameters for each machine such as temperature or moisture. Stevens mentioned six parameters in the example. Under that approach, one machine might be serviced on day 10 and another on day 180. When the company ran the ROI, Stevens said, "it just blew their mind," and the discussion led to an opportunity he was working on the day of the recording.
Why Data Contextualization Matters
According to Stevens, the light bulb got brighter once the conversation turned to data. More data makes the model better, but the data also has to be contextualized. Many manufacturers today push large amounts of data to dashboards, which he said causes "dashboard fatigue." A dashboard tells you something is wrong but not why, what happened, or where, and it does not help with root cause.
Contextualization means building relationships in the data. This motor belongs to this robot, which belongs to this line, which belongs to this plant. Those relationships matter when analyzing at the plant, line, and station level. When contextualized data feeds the model, he said, the information gets richer and the savings grow. At that point the light bulb is "so bright you got to put sunglasses on." What he enjoys most is walking a customer from the outcome they want all the way back to a specific machine on their floor and the data it produces, with the digital twin in the middle for validation. He noted that consumer LLMs have been trained on millions of books and much of the internet, and said industry has not yet done the equivalent with its own data. That, in his view, is where the work needs to happen.
Luecke summarized the flow as use case and outcomes, then data, then contextualized data. Stevens agreed and added a fourth step: how hard it is to actually get data off the shop floor. He described a company that bought $2.4 million worth of sensors because it knew it needed data, with a plan to deploy thousands of them. Siemens agreed the sensors were needed but stressed that contextualizing their data was critical.
Edge Versus Cloud
When asked how contextualization is done in practice, Stevens first challenged a claim he attributes to hyperscalers, that everything can be done in the cloud. He said that is not true. For real-time use cases such as visual inspection on the shop floor, cloud latency is unacceptable.
Siemens' answer is its Industrial Edge platform. It connects to the various shop-floor assets, pulls their data into a common platform, and applies technology that maps and contextualizes the data by building the relationships among motors, robots, lines, and plants. From the edge, data can go to a large language model in the cloud when heavy compute is needed, or to a model running at the edge, depending on the use case. Often both are needed. Stevens said this is why Siemens has partnerships with hyperscalers, to support the edge-to-cloud relationship he considers important for AI. Luecke added "contextualization at the edge" to the recipe, and Stevens agreed.
What Adaptive Production Means
Stevens will be discussing adaptive production at SPS Atlanta, and Luecke asked for a simple definition. Stevens contrasted it with the fixed, rigid automation of 20 years ago, where a station performed one motion, such as moving something up and down in 10 seconds, before the product moved on. Adaptive production lets a factory flex with demand as demand, technology, or the assemblies coming down the line change. Stations, especially robotic ones, can handle different configurations and assembly operations.
His simple version: a single line can run 50 different products. Each station reads the incoming product's bill of materials and bill of process, adjusts, performs its operation, and passes the product on.
AGVs, One BOM, and Lot Size of One
To show how this changes both engineering and operations, Stevens used a final assembly scenario in automotive and referred to a Siemens capability he called BOPEX. A vehicle's top hat sits on an AGV that can go to any of three stations. The AGV reads the configuration, identifies which compatible station is open, moves there, and completes the operation. It knows exactly what is needed because it reads the bill of process for that assembly at that station, and then it moves on.
On the engineering side, he said, this lets designers plan from the start with operations and variants in mind. A single master bill of materials can hold options and variants that produce 50 different configurations. That is what leads to "lot size of one," which he described as the modular approach. Customers want their own customized version of a car, a phone, or anything else, with their chosen color and interior, so that it feels like theirs. He said this changes the game in both engineering and production.
When Luecke highlighted "one BOM for 50 configurations" as the core of adaptive production, Stevens added that it also allows virtual validation of every configuration option. Testing then combines virtual validation with the physical testing that is still required. He called it "massive" and "a big deal."
Reshoring: Greenfield Is Straightforward, Brownfield Is Where the Work Is
For a mostly North American audience interested in reshoring and nearshoring, Stevens framed the connection through greenfield versus brownfield sites. Greenfield is the easier case. You start with a blank sheet, apply the standards and technology you want, bring in lessons learned, build and validate the whole digital twin virtually, and then start construction, with no production to disrupt.
Brownfield is harder, and Stevens said Siemens spends about 90% of its time there, though he added it is not as tough as people think. It starts with data, because many companies don't actually know what assets they have. Once they do, a digital twin is built so that new products or programs, automation, or robotics can be introduced virtually. He noted this is far cheaper than doing it physically and doesn't disrupt production. These projects can also consider sustainability and energy optimization. After validation, the discussion turns to implementation, possibly including a software-defined automation layer on top of existing controllers to make changes with minimal disruption.
Stevens said Siemens is in a distinctive position because it covers both the virtual side, with digital twins of product and production that show how product changes affect production, and the physical side. For implementation, the line builder, machine builder, solution partner, and end user sit down together around the digital twin to decide on the approach. That might mean updating the control layer, adding sensors, or adding an edge layer. In his experience, conversations about which products can be made where always come back to creating a production twin, because "there's so much value there."
Digital Transformation Fails on People and Process, Not Technology
Luecke observed that they had spent 36 minutes discussing digital transformation without using the term. He asked how it enables exponential production improvements, referencing recent comments by Siemens executives Del Costy and Barbara Humpton. Stevens called digital transformation an umbrella term, saying there isn't a single company not talking about it. Everything they had discussed falls under it. But he said it is not only about technology. It is also about people and process.
People come first. The skilled workforce has to be included in decisions, especially on usability. He mentioned a call earlier that day where his first question was how the company viewed usability, and it sparked a long discussion. If people don't use the system, he said, the value never arrives, however impressive the technology. That requires proper training and organizational change management.
On process, he listed several questions. Is there alignment up front? Is success defined, including which outcomes count and how you will know you've achieved them? He tied success to adoption. Is validation continuous, for example with scrum-style methods that put mockups in users' hands early? Stevens said many digital transformation projects fail for reasons that have "nothing to do with the technology." They fail on usability and implementation, including poor issue management, communication, and governance. He said Siemens' keynote at SPS would cover the process side extensively.
Closing Thought: Adoption and a Customer-Focused Reorganization
Given the chance to add a final point, Stevens returned to adoption, which he feels is not discussed enough. He said Siemens recently went through a major internal transformation. Before it, the company was, "quite honestly," focused more on Siemens itself and its products than on customers, their industries, and the business problems they are trying to solve. The reorganization aims to understand customers, speak in their language, and match solutions to their actual problems. "Nobody cares about our organization. Nobody cares about our products, features, and functions," he said. What customers care about is what they get.
He ended by describing the current moment as a big transformative wave. Because things are moving so fast, he said, everyone, Siemens and its customers alike, is "trying to figure it out together." Siemens, he said, is positioning itself to be adaptive toward its customers as well.
That's the modular approach. What the customer wants is they want their customized version of the thing for them. Whatever it is, it could be a car, it could be a phone, it could be whatever, right? It's really changing the game both on the engineering and operations from a design perspective and then what I described on the production side.
With use cases for artificial intelligence within the manufacturing industry continuing to crop up, it takes automation leaders like Chris Stevens to help navigate those light bulb moments in AI. Chris is about to share what those light bulb moments look like while defining concepts like adaptive production and sharing the immediate criticality of building a resilient manufacturing ecosystem. Chris, welcome to Manufacturing Happy Hour. And it is Friday afternoon. It's not quite early enough on Friday afternoon yet to have a beer, but that's probably in both of our futures. So, if we were in Detroit, where you're based, or elsewhere, where might we be grabbing a drink this afternoon for this conversation?
That's an easy answer. We are very close to football season, as you know. There's a really cool rooftop bar right by Ford Field. So, and it is a beautiful day here in Detroit. So, we'd be sitting up in the rooftop bar looking at Ford Field, probably talking about the upcoming season. And I would probably tell you I'm a little more nervous this season than last season. So that's probably what we'd be doing, Chris, if we were here together.
Yep. Yep. Before a Lions game, of course. I hope you're not too nervous. I still think they look pretty good. I think you'll be okay.
They got a really tough schedule, so that's why I'm nervous. And, you know, the expectations are high. The expectations are there. And it's hard to beat 15 and 2. So, we'll see though. Very high expectations still. The optimism is still there, but I'm not sure we're going to go 15 and 2. We'll see what happens. [laughter]
Well, I'm originally from St. Louis, Missouri. So, my team left me long ago. So, you know, I'm rooting for your success. It would be cool to see. You're not really underdogs anymore, but people still know you're gunning for that Super Bowl. So, best of luck to you there. So, if we're having this conversation, rooftop bar, you know, I've got a couple "describe it to me as if we were having a drink" type questions throughout this interview, but before we get there, let's get to know you a little bit. I mean, you've had a really long career at Siemens, and I'd love to explore how industry has changed during that time and what you've learned from that. So let's start there. You know, based on your two decade plus career, what is one of the most drastic changes you've seen in the industry, and how have you seen other manufacturers adapt to it?
Yeah, I guess I'll start with, so I spent most of my career on the software side. So on what's called DISW, I just recently came over to the automation group two years ago. So on the software side it was very much focused around digital twin of both product and production. What was consistent and has been consistent is two decades ago we were talking about things like lean manufacturing and continuous improvement, if you remember those terms, right? So those were the hot terms, those were focused on. When I was with Siemens, I would say I spent a lot of my time on the virtual world, so spending a lot of my time on manufacturing bill of materials, a lot of my time on bill of process. What does your process look like? How do we optimize your process? You know, which again goes part and parcel with this kind of lean manufacturing look. Definitely on the simulation side. So a lot of similarities there that transcend even into today, as you know, digital twin is a pretty common term that we're still using. Not just us, but almost everybody uses that term now. I would say, and it was about automation. So when we were using those tools, it was about, you know, how can I take this station and how can I automate it, and how can I take humans out of the loop, right? It was all about that. So that's where we're spending a lot of our time.
Fast forward to today, what's pretty interesting now is the conversation's more: how do I bring humans back in the loop? So, as you know, there's a transformative technology that's becoming a tidal wave right now called AI, and that's really what AI is doing. So yes, of course, the skilled workforce, you know, there's trouble when it comes to filling enough jobs, right? When it comes to skilled workforce,
depending on the study you look at, it's anywhere from 5 to 8 million jobs that are going to be available by 2030 that we're not going to be able to fill on the manufacturing side.
So they're doing two things. One is what's called adaptive manufacturing, right? So how do we, yes, automate systems, but how do we make sure humans are part of that loop so they can validate when we need to validate, right? Or they can bring that expertise when they need to bring that expertise, or they could take their skill set and spread it over five stations, not just one, right? Because that's been the kind of transition.
And then of course is this idea of autonomous, which has been pretty interesting because autonomous was a pretty hot topic, I would say, three or four years ago. Then it cooled down a little bit, not just in automotive but even on the manufacturing side: how am I going to automate? How am I going to get lights out? Then it cooled down a little bit. But now it's starting to come back around, and in fact I just saw a pretty interesting video on Tesla and what Elon's doing with robotics, both from a consumer and a manufacturing standpoint, and his vision around autonomous manufacturing. So I think it's starting to get a little more spark to it again, this idea of autonomous manufacturing. For sure in the semiconductor space we're having more and more conversations there. But that's how we've seen the transition. So initially, a couple decades ago, it was around lean manufacturing, continuous improvement, mostly around automation. How do we automate stations? We transitioned into adaptive, and now what we're starting to see a little bit is this autonomous. Now it's kind of rearing back up and saying how do we get to this autonomous stage potentially, and what does that look like?
So this is a good conversation to have over a drink because I have a few questions that came out of that answer. The first one is, I think, a personal question for you, where your career path makes you probably one of the unique individuals that can answer this. You mentioned you came up on the software side, and now we're really entering an era of what I would say is, excuse me [clears throat], an era of software-defined manufacturing. And do you feel you being early to that, right, where that was kind of your focus, has that given you a different outlook on manufacturing than maybe some of the rest of the industry?
Yeah, it's a good point. What's interesting around what we're calling software-defined automation, or what you just called software-defined manufacturing, is customers know and understand now how software works. And what I mean by that is they want to move at the speed of software. So what they're seeing is: I want to make a decision. I want to click a button. I want to go, right? And so this idea of software-defined automation is starting to resonate extremely well, because if you look at what they did on the engineering side, and again, that's where I spent a lot of my time, with engineering and IT, when I was with DISW, they've automated processes, they've automated QA, they understand environments, they know how to replicate environments, they know how to do redundancy, you know, so there's protection there. They understand cybersecurity as it relates to engineering, as it relates to IT. They certainly understand collaboration; they've taken that to a whole new level, especially with COVID and beyond, right? So they certainly understand all those things.
What hasn't necessarily translated is the manufacturing world, right? So when I talk about all those things, that's heavy enterprise systems, heavy engineering systems, heavy IT, whatever, and there's a whole infrastructure. What they're starting to figure out, and what we're trying to do from an education standpoint, is: guess what? If you take all those concepts and you apply that to manufacturing, you can go much faster, you can be much more efficient.
And so it's starting to resonate extremely well, right? And so this idea of collaboration on PLC code, as opposed to it just being on one person's laptop, taking it literally on a memory stick, going down to the PLC and updating the PLC. They're starting to realize that, hang on, I can collaborate on a platform. I can validate that with something called virtual commissioning. Once I know the code is good, I can hit a button and send that to my control layer and update my control layer and make the changes I need to make in a very quick, very efficient way.
And so, Chris, we're starting to see that conversation resonate extremely well. We know there's still a ways to go on certain requirements. So whether it's in certain industries, whether it's maybe around high performance, or if they're safety or motion, which we're addressing as fast as possible. But what I would tell you is that, oh, and the managing of the assets. So managing the actual code, managing the actual PLC, managing the different IPCs, managing all the different assets that exist on the shop floor also is a very good conversation. So I think all that stuff we learned and we understand on the engineering IT side is now starting to translate to the manufacturing side, and they get it because they have...
But the other thing that's cool is IT is becoming more and more important to that conversation. Typically, what's interesting is when you go to the shop floor, IT is not there, right? And now you're starting to see these roles like IT: I'm the IT manager, supervisor, whatever. So you're starting to see more and more of those roles because they're starting to see those worlds converge. So for me personally it's been a lot of fun, because I spent, like you said, a lot of years having those conversations, just not in manufacturing. So to bring those worlds together is a lot of fun.
Yeah, it's definitely a cool career convergence for you. And we're going to talk about how you can go faster a little later in this interview, but another comment you brought up is one of the trends you're seeing now is how do I bring humans back in the loop, and you mentioned that's what AI is doing. And we could talk all day about AI. In fact, that happens on this podcast somewhat frequently. But how would you describe that? Let's say we're hanging out at the bar. How is artificial intelligence bringing humans back into the loop? I think that would be a great kind of 101 question that the audience is still grappling with, even if they've heard answers to that question before.
Yeah. And there's two different lenses, right? So typically when I talk about AI, there's what we're doing internally, which I don't think we can go fast enough.
So internally, how are we adopting AI, and how is AI that thought leader that sits next to you? Are you looking at it that way? It's much more than just something that helps you put an email together. And I think that's a pretty big mindset shift that not many people have made. I think they're doing small stuff with it, like I said, emails, maybe a white paper, something like those, but they're not quite using it yet where they need to, where I think everyone's going to be using it at some point here in the near future. And that's your creative thought leader. You can do simulations with AI, as an example. So, if you have a customer meeting, I can simulate the customer meeting and it gives you great feedback on what you should do differently. If you have a big decision to make, are you using it? Are you giving it the right context? Are you prompting it? Are you asking it to give you three questions? Give me three questions in an interview style and then help me come to a conclusion on whatever topic it is I'm doing. So, I think we have to get there with it internally.
That's Copilot, so that's more reactive. From a proactive standpoint, it's these agents. So we're placing some big bets on developing some agents that are going to help the workforce, specifically the workforce that's focused on our customers and working with our customers. So, we want to make sure they're very well prepared with the right information at the right time so they can go have the right conversations and provide the most value to our customers. That's extremely important for us. So, we're paying a lot of attention there as well.
So, that's internally. Externally, I would tell you it is the hottest topic, hands down. And I'm not talking with controls engineers or managers; I'm talking with the top floor: presidents, CEOs, managing boards. Everybody wants to... it's a race to the finish. They know it's transformative. It's as transformative as the internet. So if you look at the internet and how the internet gave us access, it's going to transform the way we do business 100%, right, just like the internet did. So it's a race to the finish with them.
And so, working with them, I think the important piece is focusing on outcomes. You know, everyone's educated differently, at different levels, Chris, and that's what I'm finding out. It's interesting to me. I'll think we'll do something that's pretty elementary, and it's tons of light bulb moments. So I'm like, wow, I guess we've got to keep professing and telling our story because it's resonated extremely well. We thought it was six months old and we had to update it, but not really. We're seeing more and more this kind of thirst for knowledge and want to engage. And usually with them, what we try to focus on is the outcomes they're looking for, or use cases, or whatever. So that's usually where we take them.
Yeah. Okay. So you were talking internal first and then external. I'm going to do a quick host recap here, and then I've got some follow-up questions. So your internal examples, I thought you did a really good job describing it in a way I hadn't heard it before, as the thought leader sitting next to you, right? Using it as, I don't know, let's call it a creative consultant, right? You talked about simulating a customer meeting, you talked about getting the right questions to help prepare for a big meeting. That's Copilot.
Agents: can you give us a quick example of an agent? Because agentic AI is another term that is coming up quite a bit, and I think manufacturers are starting to grasp Copilot pretty well. I think agents are still one of the areas where they're like, "Okay, I've heard of agentic AI. I think I know what it is." How would you describe it in that over-a-beer context?
Yeah, there's a joke in Silicon Valley: if you throw a stone, you can hit a startup that's working on some kind of agent or agentic AI for any industry, any function. The way I do it, to make it really simple, Copilot is more reactive. So with Copilot, you're asking it questions, you're giving it context, whatever, and you get something back. Agents are more proactive. So the way you look at an agent, for instance, is we are going to set up agents that will be tied to accounts, and these agents will scour what's happening
with all those accounts. So whatever is out there publicly available or even within your own context, your emails, meetings, whatever, and it's going to present to you what's happening with your account and maybe even make suggestions. So, it's going to say, "Oh, by the way, company XYZ just announced that they're going to build a greenfield facility and invest 360 million. What I would suggest is maybe package this up and talk about how we handle plant floor layouts and how we establish costing," or whatever those things are. And it will actually give you suggestions on what to do next and how to engage with that customer next based on everything that's going on and based on the context that you have within your folders and emails and chats and all that. So that's more of the agent, and we're scouring, going all through that and being more proactive for you and being your assistant, saying, "Hey Chris, I have something for you, go do this." So that's the difference.
That makes a lot of sense. Very, very cool. And I think the way you described it that made the most sense to me: Copilots are reactive. Agents are proactive. And I'm a longtime sales guy. So that sales agent example you gave definitely resonated.
Let's jump to the external side really quickly, because you mentioned when you're going into these meetings you're focusing on outcomes, which makes sense, right? At the executive level that's where I would expect that conversation to go. But you also said there are a lot of elementary moments that are causing the light bulb moments. Can you share what a common light bulb moment is in this day and age around AI? Because based on what you just said, I imagine a lot of listeners are probably still waiting for that light bulb moment as well.
Yeah, I mean it seems like there's always tons of light bulb moments, but probably the most recent one that I could share with you was with a CPG company, and we were talking about predictive maintenance, and what we talked about is the importance. So clearly everyone understands AI from a consumer standpoint. So they're going in, they're using it like I told you, they're doing simple stuff like, "I'm traveling here this weekend. What should I do?" Right? Or, "I have these ingredients. What should I make?" Right. So some cool but simple stuff, right, from consumer.
What they don't realize is, especially in the industrial space, the large language model is only good based on what you feed it. You're going to get certain answers based on what you feed it. And so when we talk about some of the capabilities around predictive maintenance, we can say predictive maintenance, we can go do this. We can predict based on data again. We can predict, you have 50 of these machines globally, we can predict when. Because today what they'll do is they'll schedule all those 50 machines the same way for maintenance, as an example. Whether the machine's 5 years old or 15 doesn't matter. The schedule's the same. The parts show up at the same time. They do the scheduled maintenance whether it needs it or not, which means there's some sort of downtime. Certainly there's an investment in parts. Certainly there's investment in time. Then they execute that maintenance and then they go forward, right? And to mess with that's taboo. Nobody wants to mess with that.
What we showed them, and this was a light bulb moment, what we showed them is based on data, we could do the scheduling for you, and you could do the maintenance when you need it based on these six parameters. It could be temperature, it could be moisture, it could be whatever, right? Whatever those parameters are based on the machine. And so by doing that, you can optimize your schedule, optimize those machines, and do this one on day 10 and this one on day 180 and this one. And when they started doing the ROI on that, it just blew their mind, right? So it was significant, right? Which led us to an opportunity that I was just working on today, actually. So that was the light bulb moment.
So I now, and I get it and whatever. Now we completely turned the brightness up on that light bulb, because it then turned into a data discussion and how important the data is, and the more data you have and the more data you feed it, the better that gets, the better that model gets.
What's also the light bulb that got brighter is the contextualization of the data. So you can't just feed it data. So today they feed dashboards, right? They feed tons and tons of data to a dashboard. It's dashboard fatigue. Dashboard tells you something's wrong. It doesn't tell you why, what happened, where it's at. It doesn't root cause really anything. It just tells you something's wrong. That's why the contextualization of the data is so important. So what we need to do is we need to make sure we create those relationships so it understands that this motor is tied to this robot that's tied to this line that's tied to this plant. That whole contextualization is extremely important, especially if you want to look at the plant level, the line level, to the station level. Those relationships get extremely important. So now when we feed that to that large language model, now that gets even richer, and the more we feed it, the better the data gets and the better your information gets and the more money you save. And that's when the light bulb is extremely bright. So bright you got to put sunglasses on, right?
And so that's the fun part, Chris. I think the fun part is when you walk them from what they want, the outcome, and you walk them all the way back to how that translates to a machine on your shop floor and the data that comes from that machine, and to have that whole traceability. And of course, the twin's right in the middle, right? Because you want to validate using this digital twin. But that whole value stream, when they start to get it, it's really, really cool, because that's when the light bulb goes off. So they really understand it, and again they understand it because they're doing it in their consumer life, and we're translating how important that is from an industrial standpoint. So if you look at the consumer LLM, it's millions and millions and millions of books, internet, everything's been fed to that thing. That's why it's so interesting. That's why you get so many interesting results from it from a consumer standpoint. But on the industrial side we haven't done that yet. And that's where the work's got to happen. So I think that's the fun part. That's the light bulb.
So, I'm going to go over this in the context of a quick recipe. At risk of oversimplifying, tell me if I'm on the right track. When it comes to those light bulb moments, your conversation is really starting around use case and the outcomes associated with that use case, and then that leads to a data discussion, and then that turns into a contextualized data discussion. Is that the right flow, would you say?
Yes. And then how hard it is to get that data from the shop floor, as you know. I mean, you've been doing this for a while, right? So I think they kind of understand that, but because they have so many different assets. Like for instance, we were working with a company that went out and purchased $2.4 million in sensors because they knew they had to collect data. So somehow they came up with a business case. There was a plan to just, you know, launch in and have thousands of sensors. But when we started working with them, we showed them the importance of, yes, you need to go do that, but the contextualization of that is so critical. So yeah, you got it. You're spot on.
Maybe we'll just go into that last part a little bit deeper. How do you contextualize that data? Right? You gave that example of, hey, they were getting a lot of sensors and, yeah, they were getting a lot of data, but it wasn't contextualized. Is there a best practice that you can share in terms of how that gets contextualized then?
Yeah, absolutely. And then the other piece is, depending on your use cases, you can't. So what's interesting is what the hyperscalers are telling everyone, I'll get to your question, is "We could do it in the cloud, we could do it in the cloud, we could do cloud." That's not true, right? If you're doing visual inspection on the shop floor and you need that kind of real time, you can't go to the cloud. You can't afford that latency.
So the platform for us that's extremely important is the Industrial Edge platform. And so the Industrial Edge platform allows us to make all those connections to all the assets I just talked about, pull it into a common platform. Then we apply a technology that helps us do the mapping within the data and the data contextualization. So it helps us build those neural nodes or relationships between all the different pieces I talked about, this motor to this robot to this line to this plant. So we could build that kind of contextualization on that platform, and then we can take that and feed that either to the cloud, a large language model in the cloud, which sometimes makes sense because you need that compute power. Sometimes it makes sense to have that large language model on the edge, depending on what you're doing. And then of course you want both. And that's why we have strong partnerships with some of the different hyperscalers, right? So we can create that edge-to-cloud relationship that's so important when it comes to AI.
So contextualization at the edge was probably the one other thing that I could have added to that recipe we were just talking about.
You got it.
All right.
If you need a sales job, let me know, Chris, because you're doing a fantastic job.
My job is to try to, to an extent, take all the data and contextualize it in podcast form. That's basically what we're doing here on Manufacturing Happy Hour.
Hey, great conversation around AI. I wouldn't be surprised if this comes back into the conversation as we go through the rest of our discussion. But there is another term I came across recently that you've been using a lot: adaptive production. I haven't seen it a ton, so I don't know if I'd call it a buzzword that's being used a ton across industry yet, but it's the perfect opportunity to say, "Hey, we're grabbing a beer." How do you describe adaptive production as if you're having a beer with someone, right? You're going to be talking about this at SPS Atlanta soon. I'd love to learn kind of the 101 about this as well.
Yeah. I think it goes back to, we talked about 20-plus years ago, we were talking about automation, and back then it was very much fixed, rigid automation, right? We apply something, we put that there, and it would go perform something. It would move it up and down in 10 seconds and then move on, and we do the next station.
What adaptive is, is it really allows factories to flex with demand. So as their demand changes, as technology changes, as maybe a different assembly comes through that line, adaptive manufacturing allows certain stations, especially when it comes to robotics, to handle different configurations or different assembly operations as it relates to manufacturing.
So that's, simply put, adaptive is: I have a line that I can run 50 different products down and assemble those different products, and based on the bill of process and the bill of material that's coming that it reads, then that whole line adjusts to what it needs to do for that particular assembly or that particular station. Once it gets to the station, it reads, it performs updates, performs its operation for that station, then moves to the next one. So that's kind of it in a simple form. When we talk about adaptive production, that's what we're talking about.
Well, I like that example. You're flexing with demand. But the way you're describing it, what it makes me think of is this reshapes the way manufacturing is done in a lot of ways. When it comes to adaptive production, as we're talking about, how is the convergence of engineering and operations impacting the way we design and manufacture? Just, how is it changing the state of manufacturing then?
Yeah, well, clearly it's opening up design options, right, from an engineering standpoint. And I'll give you another scenario, which I think paints the picture pretty well, then we'll go back to engineering and operations, so it makes sense.
So think of this idea of AGVs, right? And we call this BOPEX, it's one of our capabilities. We can even talk about car assembly. So if you're doing final assembly for an automotive manufacturer, you have the top hat on a particular AGV. That AGV comes up and it has the option to go to three different stations. Doesn't matter which one. The whole idea is whatever one that's open and obviously can handle that certain configuration. So it reads it. It says, "Okay, I understand this configuration. It's configuration XYZ. It can go to these three stations. This one's open." It shuttles over, goes to that open station, completes the operation. It knows exactly what it needs when it needs it, because it's reading that bill of process for that particular assembly at that particular station. Once it completes it, then it moves to the next one.
And so what's pretty interesting is, again, even in the engineering stage, it's allowing the flexibility, when you're designing these different assemblies, to actually design with the intent of operations and options and variants as it relates to those bills of materials. So now you can have a master bill of materials, and within that bill of materials you can have different options and variants. So you just have one bill of material, but that bill of material can produce 50 different configurations. So now you start talking about this idea of lot size of one, right?
So that's the modular approach. When we're talking to customers, what the customer wants is their customized version of the thing for them, whatever it is. It could be a car, it could be a phone, it could be whatever, right? I want this color, I want this interior, I want this, and it feels like it's theirs. And that's this idea of kind of lot size one. What I just described allows you to do that. It allows you to set up your manufacturing the right way. It allows you to design the right way. But because you have this opportunity for options and variants and all these different configurations, it gives you this idea of you're getting your customized version of whatever that product is. And so that's what it's changing. It's really changed the game both on the engineering and operations from a design perspective, and then what I described on the production side.
I think the line that sticks out most to me there is one BOM for 50 or however many different configurations that are there.
This is like, if I were to summarize the power of adaptive production, that would be what it is. If you've never dove into that space, it's massive, because what you could do is virtual validation on all your different configurations, because we have that solve power now too. So it allows you to create one, which simplifies everything I just described, and allows you to virtually validate all the different configuration options. So from a test and validation standpoint, you have both the virtual and then of course the physical testing that you need to go do to validate that as well. It's massive. It's big. It's a big deal.
Well, speaking of a big deal, let's put this in the context of another topic that comes up on this show. Most of our audience is within the North America region. There's a lot of talk about reshoring, nearshoring, revitalizing
manufacturing. You know, how does adaptive production then connect to revitalizing manufacturing? Right. I think the audience probably can guess where this is going a little bit, but I'd love to hear it in your own words.
Yeah. So I think where you're going a little bit is there's always this discussion around greenfield versus brownfield. Greenfield, obviously, the opportunity for us is you're starting with a blank piece of paper, right? So you can implement the standards you want. You can go all the new technology you think you need. It's greenfield, so you're not producing anything, you're not disrupting manufacturing. So you could do what you need to do there. Take all the lessons learned and apply that to a new plant. Do it all virtually first. Have the whole digital twin done, validated, and then go start your build. That's pretty cut and dry, right? That's why I think a lot of people want to go or try to go or can go, that's where they go.
Brownfield, on the other hand, as it relates to reshoring, that's a little tougher, and that's where we spend probably 90% of our time. And it's not as tough as people think, right? And again, what I would tell you is it starts with the data. So what most people don't know necessarily is what assets they do have. So it starts with the data. We go back to collecting data, so we collect the data so you know what you have, right? Once you understand what you have, then we go back to the digital twin. So now we're still going to go create the digital twin and we're still going to introduce different products or programs down those lines, and if we need to introduce automation, if we need to introduce robotics, we can in a virtual world, which is way less expensive than the physical world, and you're not disrupting production.
So the conversations we're having on the brownfield side is just that. When we're reshoring, it's, okay, let's go look at the twin. Let's go look at what we can do. Let's go look at the stations we have to update. Let's make sure sustainability is in mind. Let's make sure, you know, we're optimizing energy while we're doing this, as an example. And so that's where we're taking it, and that's where we're spending a lot of time with our customers, because we have that capability.
Once we validate that and understand that, then we get into the conversations you and I were talking about earlier, which is how do we go apply that? Can we do something unique like a software-defined automation layer on top of your existing controllers, so we can implement some of these changes with the least disruption to your manufacturing, right? And so then we get into all those conversations.
And again, because we have both sides of the capabilities, it puts us in a pretty unique position, right? Because we have that strong engineering, virtual validation, digital twin of both product and production. So as the product changes, how does it affect production? We have that close tie on the virtual world, and of course we understand the physical world. So when it comes to actually implementing it, that's when we leverage our ecosystem. So we bring in a line builder, we bring in a machine builder, we bring in a solution partner, along with the end user, whoever the end user is, and we all sit at the table and we have a conversation around that digital twin and all figure out what's the right way to go implement this.
And again, we have all the capabilities. It could be just update the control layer. It could be, no, no, no, we need to add sensors here. It could be, no, no, no, we want to put the edge layer here. You know, whatever those options are, that's where we're spending a majority of our time on this whole reshoring: what products can we make where? And again, it starts with that digital twin of production. It always goes there. We have a lot of great conversations, but at the end of the day, you're like, "Okay, let's go create this twin, that makes a whole lot of sense." So it always goes back to that because there's so much value there. There's so much value.
You know, Chris, one of the things that almost surprises me a little bit about this conversation so far, we're 36 minutes in and not once have we used the term digital transformation, but I feel in many ways we've been talking about that the whole time, right? We're talking about ways that you can virtualize your commissioning, virtual validation, you can have one BOM for 50 different configurations. And I think this all leads to my last question. How is all of this allowing for exponential production improvements? I feel like I've seen other executives from Siemens right now, like Del Costy and Barbara Humpton, talking about this lately, and I'd like to hear your perspective and maybe a story that illustrates the promise and realities of what digital transformation can really look like.
Yeah. I mean, you kind of said it, right? Digital transformation is an umbrella term. I mean, there isn't a single company that's not talking about digital transformation. Not one, right? So everybody has some sort of projects as it relates to digital transformation. So I think that's the easy button, right? That's what we go in and say, "Hey, let's talk digital transformation."
I would say when Del and Barbara talk, and rightfully so, they talk about the people as it relates to digital transformation projects. So we talk about this thing as a project. A lot of the capabilities, a lot of technology, a lot of things you and I have already talked about for the last 36 minutes all apply; that's all under that umbrella of digital transformation. The one thing we haven't talked about as an umbrella term, or as it relates to a project, is it's not just about the technology. It's also about the people and the process.
And what's extremely important is, number one, the people. As part of the project, we have to make sure that the skilled workforce comes along for the ride, that they're included in the decisions we're making, especially as it relates to usability. In fact, I was on a call earlier today with a company, and one of my first questions was, how are you guys seeing usability? And it launched into this whole discussion, right? And so that's extremely important, because if they don't use the system, you're not going to get the value. You can have the coolest technology, all the cool things you and I talked about in the last half an hour or whatever, but we have to make sure the people are there and are with us, and we're putting proper training in place, proper organizational change management in place. All those things are extremely important when it comes to digital transformation as a project.
And then, of course, the process. I think the process is extremely important. We spend a lot of time on that, and the process as it relates to implementation. So do we have alignment up front? Do we understand what success looks like up front? What kind of outcomes are we looking at? When do we know we have that? Well, that's when we get adoption, and that's the people coming back into place. And then are we validating throughout? Are we doing interesting things like some sort of scrum technique or whatever, where as soon as we have some sort of mockup we're getting it into users' hands? All those things are extremely important, because there's a lot of digital transformation projects that fail, and it has nothing to do with the technology. It has everything to do with usability, making sure that people like what they have, and then how we implemented it. You know, we didn't do a good job in issue management. We didn't do a good job on communication, right? We didn't do a good job on governance.
These are all things that are extremely important when it comes to digital transformation. So thanks for bringing it up, because it's extremely important. I know when Barbara and Del talk about it, they talk about it in the context of that, and a lot of times mostly it's about people. But that is something that you will see in our keynote at SPS. We do talk about that. We do talk about the process extensively and how important that is.
Well, I'm glad you brought that in here at the end, because my final final question is, we've covered a lot of ground today, a lot of different areas where we could certainly go into more detail. Is there anything you wish I would have asked you, or one other topic you want to add one final thought on before we wrap the conversation?
No, I just think what's important for us, and a major focus for us, is adoption. It's kind of what I just described, and I don't think we talk about it enough. I know we do internally. We talk about how important the customer is and what the customer needs, and making sure we meet those needs as fast as possible. I think that's extremely important. We're setting up our organization to do just that. We personally just went through a big transformation internally, because we were focused more on Siemens, quite honestly, and our products and who we are, and we weren't focused enough on the customer and the industry and the business problems they're trying to solve.
And so that's a huge emphasis for us. It's extremely important we understand the customer. We talk in the language of the customer. We understand their business problems. We make sure that we give the right solutions to solve the right problems. That's our focus. That's everything we're doing today, and that's everything we're going to be doing going forward, right? It's that focus. Nobody cares about our organization. Nobody cares about our products, features, and functions. All they care about is what they get, and that's what they should care about.
So I'm just excited that we are there. Siemens is there. We organized ourselves even more to be focused on that, and we're extremely excited to engage. It's a lot of fun time because of all the things I described. It's very transformative. I mean, there's a big wave coming at us, and it's tons of fun right now because everyone's trying to figure it out. We're actually all trying to figure it out together because it's moving so fast. But we're positioning ourselves to be very adaptive when it comes to our customers as well. So thanks for asking that. I appreciate that.
Oh, of course. Yeah, I think that's a great way to wrap the conversation, right? I like that the things you've been doing externally for your customers, you've been internally implementing as well. You know, we're at the end of today's discussion, but I sincerely hope that when we're down there in Atlanta, we do get to grab that beverage together. So, absolutely. Thanks so much for jumping on the show.
Appreciate it. Thank you. Look forward to seeing you soon.
Article published