Episode #30: The Future of U.S. Manufacturing: Leveraging AI and Smart Factories with Bryan DeBois
In this episode of U.S. Manufacturing Today powered by Veryable, host Matt Horine speaks with Bryan DeBois, Director of Industrial AI at RoviSys, about how data, artificial intelligence, and smart factories are transforming American manufacturing. Bryan shares his 25-year journey in the industry, the evolution of his company's AI initiatives, and how they are addressing current challenges. Key topics include the role of traditional AI, autonomous AI, and generative AI in modern factories, overcoming resistance to AI adoption, and the potential for AI to drive a reindustrialization and reshoring of American manufacturing. The episode provides valuable insights for leaders looking to incorporate AI and digital transformations into their manufacturing strategies.
Links
- RoviSys Website
- Bryan DeBois on LinkedIn
- Navigating Trump 2.0
- Veryable Is Revitalizing U.S. Manufacturing
- Sign Up on the Veryable Platform
Timestamps
- 00:00 Introduction to U.S. Manufacturing Today
- 00:24 Meet Bryan DeBois: A Leader in Industrial AI
- 01:11 Bryan's Journey into Industrial Automation
- 02:39 The Evolution of Industrial AI at Veis
- 03:03 Understanding Operational Technology (OT)
- 04:12 The Shift to Information Solutions
- 05:31 Pioneering Industrial AI at Veis
- 07:14 The Rise of Smart Factories
- 09:54 Leveraging Data in Modern Manufacturing
- 13:02 Exploring Traditional, Autonomous, and Generative AI
- 17:33 Generative AI in Manufacturing: Potential and Challenges
- 20:02 Generative AI Limitations and Risks
- 20:55 Current AI Use Cases and Challenges
- 21:24 Manufacturing's Cautious Approach to AI
- 21:51 Personal Experiences with AI
- 22:21 AI in Document Comparison
- 23:42 The Future of AI in Manufacturing
- 24:09 Overcoming Resistance to Industrial AI
- 25:14 Realistic Expectations and Flexibility in AI Projects
- 27:36 AI's Role in American Manufacturing Competitiveness
- 33:02 The Vision for the American Smart Factory
- 35:02 Getting Started with Digital Transformation
- 36:35 Conclusion and Contact Information
Episode Transcript
Matt Horine: Welcome back to US Manufacturing today, the podcast powered by Veryable, where we talk with the leaders, innovators, and change makers, shaping the future of American industry, along with providing regular updates on the state of the industry, the changing landscape policies and more.
Today we're diving deep into the future of manufacturing where data, artificial intelligence, and smart factories are transforming the way America makes things. Our guest today is someone at the forefront of that transformation. Bryan DeBois, Director of Industrial AI at RoviSys. Bryan has been with Roy. For 25 years helping shape it into a leading global systems integrator for manufacturing and industrial solutions.
Since 2019, he has led their industrial AI group years ahead of the curve compared to many of today's adopters. And under his leadership, the group has focused on three key areas, which we'll discuss today. Bryan is a recognized thought leader in industrial ai, a frequent keynote speaker, and he is been featured on more than 20 manufacturing related podcasts.
His expertise centers on leveraging ai, machine learning, and advanced [00:01:00] analytics to revolutionize productivity. Empower manufacturers and drive true digital transformation across the industry. Bryan, we're thrilled to have you on. Welcome to the show.
Bryan DeBois: Thanks, Matt. Glad to be here.
Matt Horine: Well, we're really excited to have you on and I always like to start out with a little bit of background and journey and how you got to this point.
You've been with your company for 25 years. What first drew you to the world of industrial automation and manufacturing?
Bryan DeBois: My background, I graduated from the University of Akron with a computer science degree, so. With my background, I really knew nothing about the manufacturing space, and so I was just looking for a job.
I was a programmer looking for a job. Now we're talking here that we're in the late nineties, early two thousands. And so I went to a career fair at the University of Akron and I met this company and was fascinated by everything they had to say, and I started here in 2000 as a co-op and then started full-time in 2002.
Really, it's been that 25 years working in this industry that has had me fall in love with manufacturing. I've now had the privilege of going into [00:02:00] hundreds of plants and seeing how things are made. It's just absolutely fascinating. I love being plugged into that aspect of the world. I've really come to appreciate and respect the role that manufacturing plays.
I really do feel like it is the lifeblood of any economy, and it's such an important aspect of what we as Americans do, as we make things. And yeah, I was just drawn here looking for a job and then fell in love with the industry. And it's funny because I don't really use my degree per se as much anymore.
I have not coded in probably over a decade now. I'm in this consultative role and leading a team that they, yes, they write code, but then they also do the data science and the ML model building, which of course we'll talk about today.
Matt Horine: One thing that really stood out to me when we were talking was if you could share the story of how your company was ahead of the curve on this, the industrial AI group that you put together back in 2019, and why you guys chose to move so early into this space as a way to frame our conversation on digital manufacturing.
Bryan DeBois: So 2019 was the 30 year anniversary of Rvac. [00:03:00] A little background on the company. We are an OT system integrator. Now a lot of people are not familiar with that term ot, so for some of your listeners, OT stands for operational technology right now. This is a term that was coined by Gartner back in, I think, 2006, and the idea was that at the time, everyone knew what it meant.
But they needed a term to refer to the world we've oftentimes referred to as plant floor or manufacturing the world where machines actually make a material impact on the real world, right? So we're not just moving data Now. The machines that we deal with, the assets we deal with actually move things.
They make things, they move product. And so that world is called ot. So as a system integrator in that world, we are building solutions to solve problems for those manufacturing and industrial customers. And those solutions oftentimes will touch lots of different systems. So we started in the 180 9 as a traditional control system integrator, and we're putting in the control systems, which again, for some of your listeners who aren't familiar, those are the fundamental smart things on the plant floor, that open valves, close [00:04:00] valves, heat up reactors, move conveyors.
Move robots nowadays, like those are the control systems. So we started out doing that all through the nineties and then in the early two thousands we started to move into the information solution space. So this is where we're still in the plant floor, but now we're leveraging all of that data that's coming from the plant floor and we're doing interesting things with this.
So this is the world of historians where, this is the world of manufacturing execution systems or MES systems, this is that world. There's a lot of custom software in that world as well. So we spent our, the two thousands, I spent the two thousands doing that type of work for our clients. Then in 2019 we hit our 30 year anniversary and we said, what's next?
And so as an OT system integrator, we felt like we had pretty much mastered the things that we had tried up to that point. And so it was time to push ourselves. It's a great culture here at osis. I've spent the last 25 years of my life here. I love the company. I think we do things really well. And one of the things, one of our fundamental tenants is to push ourselves out of our comfort zone.
So we said it's time to do that again. It's time to lace up our sneakers and see what what's out there in the wilderness. And so [00:05:00] we looked at three different areas. One of those was industrial networking, and so we started a whole division that just does, focuses on networking, cybersecurity, locking down plant floor networks, doing all of that, but in the plant floor, not in the carpeted space at those facilities.
The second area was manufacturing execution system or MES. These are big for those of your listeners. These are big software systems. That tend to live below the ERP, but above the plant floor. We started a whole division that just does MAS implementations, right? And digital transformation around that.
Then the third division is the one that I was fortunate enough to be named the director of, and that's industrial ai. And even in 2019, now you gotta remember, this is three years before Chachi PT is released. Right? So at the time I wondered if we were too soon to the party in 20 19, 20 20. I was walking into meetings with customers and I'm saying, I'm the director of industrial ai.
And they're like, okay, what does that even mean? But what I also found though more often than not, is I found that I would walk into clients and I'd talk, [00:06:00] start talking to them about this, and they would say, oh, we have an AI department. It's small. We have a small data science department. So there were companies that were already starting down this journey, and that actually in a lot of ways, validated this decision that we had made, that it wasn't too early, that it was the right time to start moving there.
Then of course, November of 2022 is the release of chat, GPT. Everyone now wants to talk about ai. And so I'm, my phone started ringing off the hook, and so I'm having those conversations with clients. Now, what's interesting, and we'll talk more about this later, but they come in and they wanna start talking about generative AI because that's what they're familiar with.
Generative AI being the umbrella term for things like LLMs and chat, GPT, and those types of things. That's what they're familiar with and that's what they think AI is. And then in a lot of ways, I have to pivot the conversation to things that we are doing today versus stuff that is more down the road.
But really, yeah, that was, that was the impetus to get things started in 2019. So now we've been doing it for over six years. We actually feel like we are in a good place. A lot of OTSIs are just now starting to dip their toe into this. [00:07:00] And meanwhile, we've stubbed our toes more times than not. And we figured out where, where the pitfalls are and we feel like now we're in a good shape to, to really deliver these.
And we've got real AI models that are running on plant floors in production. We feel like we've got a pretty good handle on things now.
Matt Horine: That's great. I think that pivots us directly into the manufacturing space, which you talked about, you had grown such a passion for earlier in your career and the types of customers and clients that you worked with.
A lot of people who listen to our show or probably in their daily work life, hear the thing or the phrase like a smart factory. And when we talk about smart factories, what does that really mean in practice? It's not some kind of futuristic thing, but it's something that makes all the stuff that we already use in our daily lives.
Bryan DeBois: Yeah, this term smart factory and Industry 4.0, and even digital transformation to a certain extent, there's a lot of hype around them and they, they've started to get morphed into whatever these big consultancies want them to mean and the big vendors want them to mean, and so they start to lose their meaning completely.
For me, I think that in a lot of ways, the story of smart factories [00:08:00] starts in the early two thousands, and we've been really building smart factories for a long time now. This is not a new thing. What we're trying to do is really just make them smarter. So we're really trying to get to smarter factories, and what that looks like is leveraging the large volumes of data that we've been collecting now.
Right? So when I started my career in the early two thousands, we were still teaching people what a historian was and why am I keeping all this data, right? Why? I literally would have clients say, we just throw the data out. Why would we ever keep it? What could we possibly ever do with it? No. Thankfully that was the minority.
Most of our customers bought into the vision, and so they've been collecting this data. Now what do we do with it? I walk into many of these companies and they want to tackle some of these analytic, big analytic and AI problems, and they're afraid they don't have enough data. Typically, what I tell them is that you are probably more data rich than you really think.
You know that you are. We've been collecting all this data. Now what do we do with it? And so it's about building smarter factories. And a big part of that, and [00:09:00] one of the big primary drivers today is this loss of, and I know you're very familiar with this, but this loss of expertise as folks have retired from the workforce, right?
People are calling it a silver tsunami. It's primarily baby boomers who are leaving this workforce. They're taking with them decades of experience about how to run these, this equipment, how to make these products, how to maintain this equipment. And so while ai, I always tell customers AI is part of the story.
It's not the whole story, but AI is part of the story about how we're going to then catch up and fill that gap between that expertise that they've lost and making those novice operators, those newer operators more effective. And part of AI's part of that story.
Matt Horine: It's definitely become part of the story, and you make a lot of really great points, especially around data.
People don't know how much they have and how much is generated, and it's something that really is an asset to a lot of organizations that they didn't know prior to this. How are manufacturers today that you work with using this data differently [00:10:00] compared to maybe a year ago, maybe five years ago? It sounds like they've found it.
How are they using it differently?
Bryan DeBois: Five, seven, even 10 years ago, the story was about, okay, we've been collecting some data and we wanna look at it at this site and we wanna solve this very specific site level problem. Now the story is about we've got all this data, we've got 40, 50 plants in the enterprise.
How can we pull all of this data together and get insights that then stretch across the entire organization? And you've got companies where they're very integrated in that maybe an upstream plant within their enterprise makes a product that then goes to a downstream plant in that same enterprise, in that same company.
For further processing or things like that. So they already have a very interconnected web inside their enterprise. What can they do to find insights that are going to have an impact across that entire enterprise? And so it's about pulling all that data together and finding correlations between all of those things.
Sometimes the problems that, that they want us to [00:11:00] solve are real intricate. I had a, this was a company that makes supplement supplementary like powder that you would put mix with water and that kind of thing. And so they're filling these plastic tubs with this powder and they're filling it tub after tub.
Everything's fine. Then they start filling and they have not switched the SKU or anything else. It's just another batch, same sku, and suddenly the filler is getting jammed up and every time that the filler clogs it's hours and hours of downtime for them to clear that clog, reset the line, and get things moving.
Again, these lines run very fast, so to to ramp back up, up to the top speed, and they're like, we don't understand what's going on. And again, by the time you're reaching for ai, all the easy problems have been tried. All the easy solutions they've tried. And so when you tease it apart, it's okay. It ends up being something along the lines of, and I'm making this up, but something along the lines of when the raw materials from this upstream supplier and the humidity is this, and this filler hasn't been maintained in this long, and you're running this particular product after this other product, like it can be that intricate of a thing that that's where you're gonna have these jams, right?
So it's about finding those [00:12:00] insights, but to be able to build a data set to even be able to do the data science to get to that insight, we're looking at all kinds of data silos across the organization to be able to pull all that that together across the entire enterprise. I've got a customer right now that's got six different ERPs, okay.
But to solve some of these big problems, and that's what they wanna solve, is these big problems with ai. Then we've gotta figure out how to pull all that data from all six ERPs. Get it into a data set that is, is clean, correlated, consistent, where everything's being referred to in the same way in that one data set.
This is where a lot of times the cloud comes into play, right? Like at Love it hate it. Like the reality of it is that to get to that scale. We're typically using the cloud, and so that's where the cloud becomes a core part of all of this is that's the destination in a lot of cases where we're pulling all of this data together.
Matt Horine: You can definitely see where that comes into play because a lot of times, and what we've talked about a lot on the show is that it's only as good as the input, right? If you're not getting good data or clean data or the right kind of data, then the output is [00:13:00] not gonna be beneficial. Pivoting now a little bit into.
More industrial AI and practice ROIs has divided those three categories, those three segments, which you highlighted earlier, traditional ai, autonomous and generative. Could you walk us through how each of these segments is applied on the shop floor and you know, what's the differentiation between the three?
Bryan DeBois: I came up with this a couple years ago because I ha, I was struggling to explain to folks who you know, and I don't blame them, it's my job to know all this. It's not their job, right? So they're just trying to keep the plant running and I figured out that this was the best way for me to explain to them what the different types of AI are.
This allows me then to start to pivot the conversation away from AI that is a little maybe more bleeding edge and maybe not ready for prime time to AI that's been established and that we've used for a long time. So we'll go through each of these categories briefly. So, traditional ai, and again, it's funny to call traditional AI when a lot of these companies are just now starting to dip their toe in ai.
But the reality of it is that the algorithms in this traditional AI bucket have been around for decades and are well [00:14:00] established and are already in use in a lot of products and things on the plant floor. So. Traditional AI that the subcategories there would be like anomaly detection. So that's where we can hook up an ML model.
It will learn what normal looks like, and then it'll be able to tell you when things start to go abnormal. It doesn't need any training at all. You literally just hook it up and within a few weeks or months, it'll be able to start telling you when things are going abnormal. That's anomaly detection. Now, importantly, it can't tell you why.
All I can say is that things look different today than they did yesterday. Go investigate. That's anomaly detection. The second broad category under traditional AI is predictive so that anytime you've heard the word predictive, that falls under this category. So that's gonna be predictive maintenance, predictive quality, predictive set point.
And the idea behind that is that we're taking large volumes of very clean, very correlated data, and we are sending them into an ML model and it's, and we're training it to learn to predict a single value. In the case of predictive quality, what's the final quality of this batch gonna be? And it'll be able to predict that before you send any samples to the quality lab.
In the case of predictive maintenance, how many days until this piece of equipment is gonna go down, right? [00:15:00] That's the goal of predictive maintenance. So that's that whole predictive category. And then the final subcategory under traditional AI is your computer vision. This has been around for a long time, although recent improvements over the last five years have made computer vision even more capable.
And the one thing I typically say with this is that we really should be looking at computer vision as a new source of signals. It's not just about, oh, the computer vision is rejecting or accepting something. Like that's a very rudimentary use of it. We should be looking at. It's much more capable now than it was.
So we should be looking at it as, send me signals and then I'll make decisions based on the signals similar to how a human, if you were paying a human to do that. That's how they would interface. So that's how we wanna look at computer vision. That's traditional ai. That's it. Then we move to the second category.
That's autonomous ai. Now, autonomous AI is different. Autonomous AI is actually able to make human-like decisions. The way it does that is, is it uses a learning algorithm called Deep Reinforcement Learning. And this came out of DeepMind back in 2016, which was the Google spinoff, if you guys [00:16:00] remember, alpha Go and that whole thing.
So deep reinforcement learning is actually able to, it's been proven to understand causal connections. It can build long-term strategy and its goal seeking, so it's always trying to win the game. So that's autonomous ai. We can build decision support systems that can look over the shoulder of a novice operator and say, look, based on the current state of the system, this is the next best move I would make.
And then they, the novice operator makes that move and then it would say, okay, now I would make this move. Now I would make this move and it can make those recommendations. That's autonomous ai. The final category and the one that's getting all the hype right now is generative ai. That's the one that chat, PT LLMs, all the image generation, all the video generation, it's all under that generative AI umbrella.
And while it's very powerful and we already are seeing the impact that it's having on the world, there are some very specific limitations, which we can talk about later to generative AI that don't really make it applicable on the plant floor yet. It's really something we're keeping away from the plant floor for the moment until some of these [00:17:00] things are worked out.
So that's it. Those are the three categories of ai. And that's typically I lead with that explanation with customers so that they get a better feel for what we're doing.
Matt Horine: It's a really good overview and something that kind of separates it into a few different buckets for most listeners. 'cause our exposure to it is through the lens of the consumer, right?
We probably hear about AI on LinkedIn, ai, this or that. Zooming in on generative ai, it's often seen as that what you highlighted, the flashy consumer tech making AI images or maybe using it an LLM to help you in your daily tasks and organization. Where do you see it going and adding value in a plant or factory environment?
And you said we don't see that yet today. Is that a capabilities thing or is that an applications problem, or how do you see that eventually maybe folding in or, and are we close
Bryan DeBois: to it? So where it can have an immediate impact, and I fully support it today, is in two areas within a manufacturing facility.
And both of those are in the carpeted space by the way. The first area would be on the design side. So you're already seeing [00:18:00] vendors, Siemens and Rockwell, and you're, and all the vendors are basically coming out now with co-pilots where to write the code that runs on those controllers that we talked about earlier.
That's called Ladder Logic Code to write, help write that code. It's now you've got a co-pilot and it's fully supported by the vendor and it's gonna, it's gonna spit out this ladder logic and help you write it faster on more like the PLM design. You've got a lot of ai, generative AI that can help you in laying out either the design of the factory or the design of the actual machined part that you're gonna make and all of that.
That's all fine. Use it on the design side all you want. The other primary area where people are using it is for knowledge problems in the carpeted space, right? So. Like any business, you've got large volumes of data. You've got lots and lots of data, and you're lots of written data. Lots of SOPs. Lots of PDFs, and you're just trying to sift through it all and find the answers to all of that great use of generative ai.
Where we wanna be real careful is in Lever, trying to [00:19:00] leverage generative AI on the plant floor itself. The problem is that you've got some very ambitious AI vendors. Many of them came from the IT space, so they don't understand the plant floor at all, who are saying, Hey, you've got a maintenance problem.
I'll tell you how to fi fix your maintenance problem. You gimme all your SOPs, you gimme all of your help desk tickets, you gimme all of your maintenance records, you gimme all your manuals for all of your assets on the plant floor. I'm gonna send it all into this LLM, or now they're calling SLMA small language model, which is supposed to be like super specialized and not have these problems, which it still does.
And I'll train this model, this chat bot, and then you're gonna turn it over to your least experienced maintenance person, and they're gonna, because your most experienced person doesn't need it. So it's gonna be in the hands of your least experienced maintenance person who doesn't have the ability to question the recommendations that it's saying.
And they're gonna say, Hey, this machine's down. How do I fix it? Now, whether or not that LLM knows how to fix it, it's going to give you an answer. 'cause they're coded not to say, I don't know. And so it'll say, I know exactly how to fix that. Here's what [00:20:00] you're gonna do, is you're gonna torque this bolt and you're gonna add this in.
You're gonna do this and you're gonna rub the engine. You could blow up the plane, you could kill somebody, right? So it's not appropriate to use these tools on the plant floor yet. And this is commonly known as hallucinations. This is the primary limitation of generative AI is when it gets over its skis.
When it starts to speak with authority about something it doesn't know anything about when it's writing an email or a marketing blurb. It's just funny when it's on the plant floor, it can have serious consequences. And the AI vendors know this, right? So if, if you're talking to an AI vendor right now, and they're, if you're listening to me right now and your AI vendor is saying something like this, you need to ask them, is this AI product supported in scenarios where it can have an impact on life limb or loss of property?
And when pressed, most of them will, all of them that I've talked to will say, oh yeah, no, it's not supported there. Then okay, you have your answer. Because on the plant floor, everything we do can have a risk of life, limb, and property. I have no doubt that we'll be using generative AI in three to five years on the plant [00:21:00] floor and it'll be safe.
The thing people forget is this is a very new technology. We've not even, next month is the three year anniversary of the release of chat GPT. Let's have some perspective here. People like, let's not try to jam something like that down. And it's funny because manufacturing is typically very risk averse.
For some reason with generative ai, they are just in such a hurry to shove that down on the plant floor. And all I'm just trying to say is let's hit the brakes. We've got two other categories of AI that are proven that have been around for decades. Let's look at what we can, the impact that those can have today.
Let's just keep an eye on generative AI and, and not try to rush into it. It's really weird in a lot of ways.
Matt Horine: No, it's a, it's got a lot of buzzword around it. I think that the consumer facing mindset of it is something, yes, we can use it in our everyday lives. I know people that basically use it as a search engine at this point, which it's replaced a search engine in many ways.
I've been told once before that I've been gaslit by ai. I, I asked a question and asked for a deep analysis and it just kept working on it. [00:22:00] Saying it'll be here soon. And to your point, there wasn't a no, I can't do that. Not even, I can't export that to a spreadsheet for you. It was just, I'm still thinking, here's, do you want the first 15 rows of this analysis?
Sure. And then it couldn't do that. And so it was just, I was told I was being gaslit by ai. I hope I'm not the only one that's fallen victim to it.
Bryan DeBois: It happens to me all the time. I'll give you my own example. So I, I'm using it in very specific cases, and again, I've got the background that I can go and double check it's work and things like that, but.
Particularly for things where you're just trying to even get pointed in the right direction on where to look for something or you're, or very mechanical type of tasks. So this was the task. I had two contracts and one was in word format and one was in PDF. Now they were versions of the same contract.
Now I could go through the whole exercise of figuring out how to export one to this, that and the other and make a comparison, but I'm like, you know what? I bet you. That I was using Microsoft copilot at the time, but any of 'em would've, would've done the same thing. So I'd be like, I bet you that it, I could probably give you both documents and it could probably [00:23:00] do the comparison for me.
And so it did, and it put it this nice little table together of here's the material differences between these two versions of the contract. I go, that's great. Perfect. That's a good use of generative ai. Then it goes, would you like me to create a single red line? That combines the two and shows what's specifically what has been changed.
I go, wow, you can do that. That sounds great. Yeah, do that for me. And same thing, it goes process analyzing Aly, and then it comes back and it's got, it's, I'm done. Here's a, here's the link to the final red line, PDF. Click this link. I go, oh, this is great. Click the PF. Now, as you could already probably guess the PDF was empty.
Because this is what it does is it gets over, its skis, it over promises and under delivers. And again, it's funny, I just rolled my eyes and I went and whatever and went on with my day. But again, back to the plant floor, like we can't have that. You, you can't make this mistakes down there. And we know that because we've been doing this for 36 years, like how high the stakes are on the plant floor.
So that's all I'm telling people is just hold off, hit the brakes a little bit on, on trying to roll this thing out.
Matt Horine: You said something earlier that was of interest. Manufacturers are [00:24:00] usually slower to adoption. On a lot of different things and there seems to be some kind of appetite for this in a different way.
I'm trying to marry those things up. It doesn't make a lot of sense. What's holding some manufacturers back from adopting industrial AI and how can leaders overcome that resistance? So you see a lot of appetite for it, but there's also probably a big space where people have stood off to their detriment, where it's, there's something that works for you.
How do they find that?
Bryan DeBois: So there's a number of things. Some of them are the typical things that hold back any manufacturing project, lack of budget. There's not really a champion within the organization to bear that torch and keep that project moving forward. Political wins within that organization. A lot of these projects fail not because of any technical problem, but because of political problems within the organization.
Leadership changes while you're in the middle of negotiating this whole industrial AI project that they've got major leadership changes, sales, mergers, acquisitions, all the those. Types of things specific to ai. I would say lack of understanding of ai, but a lot of that's starting to go away. I'm out [00:25:00] there.
I do a lot as, as you mentioned in the intro, I speak at a lot of trade shows. I'm doing these podcasts. I'm trying to get the word out on industrial ai, and there's lots more people like me in the space that are trying to be thought leaders. Get people comfortable with AI on the plant floor, and that's part of it too.
I will also say that I think that there's unrealistic expectations on both sides. I think that there's unrealistic expectations in that I'll go into a plant and they make a very specific thing. Let's say I was in a plant that. It makes a very specific fiber in a patented way, and it's very specific and they wanna know the 10 other plants like them that have done this, the very specific thing that you do and no one else does.
No, I don't have 10 examples of textile plants that have done this exact thing that you're trying to solve. So unrealistic expectations about that. Understanding that this is newer technology and in a lot of ways we're solving problems no one's ever tried to solve before. So you have to go into it with that kind of a mindset.
Yes, there's gonna be risk, we're taking some risks, the customer's taking some risk on this, on this journey. We do a good job, I think, of trying to de-risk project [00:26:00] at every step of, of the way as we can, as much as we can, but. There's risk there. I can't guarantee you at the beginning of this project that this AI model is gonna be predictive, that it's gonna do exactly what we set out to do.
Now, in a lot of ways, we can oftentimes pivot partway through the project and say, you know what? We started down this road. You're missing important data. Let's set it up to collect that data. But while that's happening, while that's collecting, we actually discovered this other use case. Let's now go and try to solve that use case instead.
So having that flexibility to be able to pivot is important. Unrealistic expectations there, but then also unrealistic expectations where we still go in and people are like, so is this Skynet? Is this gonna put us all out of work? I'm like, no, it's not. In fact, in, in a lot of ways like it, it can be very smart in a very narrow way, but broadly, these are very dumb.
AI is very dumb, and I can say that in the role that I have, people come and ask me broadly, should I be afraid of ai? And what? I'm always like, no, you don't need to worry about it. Like I. I use it every single day. It is very dumb. We're so far, even the people who [00:27:00] are saying we're close to artificial general intelligence, A GI, we're so far from that in my mind.
Yeah, it's nothing to worry about there, so I deal with a whole spectrum of expectations, so it's fine. It's all part of the job.
Matt Horine: Yeah, no, for sure. It comes with the territory and I'm sure it's a relief to a lot of folks who are listening. That Skynet is not activating. It's one of those things that bridging up the, the Hollywood version of what this is and the real life applications.
Ideally, what we talk about a lot with the American worker is that it's uplifting the worker. It's up tooling them. It's doing this in a way that makes them not only more efficient, but safer, happier in their work, more productive. Leads me into our next segment and something we talk about a lot on this show, the back road trends and Reindustrialization reshoring in general.
Something very specific to that that I think could give America the competitive edge and specifically in manufacturing. Where does AI fit in that strategy? Not only to stay globally competitive to, but to bring that industrial base back and put us on the path to maybe dominating the century and manufacturing as well.
Bryan DeBois: I think it [00:28:00] absolutely can be an American competitive advantage. So I'm an American, so I, I have that bias and I speak from that perspective, right? So I wanna see American manufacturers be successful and I want them to be incredibly productive. And we've got a heritage in, in, in the United States of industry and productivity, and I love that, and I wanna keep that going.
And I think that AI is going to be, it's not the only p. Part of the story, but it's in a very important part of the story of making sure that we can keep up. It's the difference between when you are competing against countries, and that's really what it is. We're competing against these against countries and they can just throw bodies at the problem.
And humans are amazingly innovative and clever and adaptable, and so they can just throw tons of those humans, hundreds of those humans at the problem. We don't have that type of a workforce in America, and that's okay, but what that means then is that we have to supplement with AI and automation. We need to get to that point.
So I think that AI needs to be a fundamental part of that strategy. Now, I'll take it a step further. When I look at how, [00:29:00] frankly, out of date a lot of these factories are, we've got a lot of ground to make up. When people think of manufacturing in factories, they typically, in their mind, if they're not in this world like you and I, they're envisioning a factory with robotics and stuff flying all over the place, and very minimal staff and just this super futuristic thing.
And that does not describe the majority of customers that I work with. In fact, I'll go as far as just, I'll call them out. Pretty much you're describing an Amazon fulfillment center and that's about it. But that doesn't, that's not the vast majority of companies by any stretch of the imagination. It is still highly manual.
It is still very much the quality, the end quality of the product. The throughput of those machines is still very much dependent on the skill of the operator and. That's the world that we live in, and so to be able to make up that gap, we're going to have to leverage ai. There isn't any other technology that I'm aware.
There's nothing in the wings that's just waiting there to revolutionize manufacturing. Ai is it? And we're starting from ground zero. Like I said, these are new [00:30:00] projects. People have never really tried to adapt AI to some of these particular problems that they're trying to solve. So we, we gotta go in, roll up our sleeves, be the innovators that we are in America, and we've gotta leverage AI to make these things happen.
And I'll address one more thing, 'cause this is the most common question I get. Are we putting people out of work? It is the exact opposite. It's the exact opposite. They can't find enough people to do these jobs. To the tune of hundreds of thousands of people shortfall for the U.S. manufacturing workforce in any given year, as those primarily baby boomers retire, the workforce that's coming in, they're just not, and I don't ever wanna say anything disparaging about different generations or anything like that, 'cause I don't believe in that.
But they're just not interested in spending the next decade and a half learning how to make a vinyl extruder work. They just aren't, that's not their thing. So you're dealing with high turnover, low levels of engagement. They may be there for less than a year, and then they're gonna move on to the next thing.
Yes, we've gotta upskill them. Yes, we've gotta get them more engaged. But then we also have to get the equipment [00:31:00] smarter so that really anyone with a couple days of training can get good product out of that, can get good throughput, can get good quality, and those machines are not smart today. So let's work on leveraging AI and automation.
To make those machines smarter and therefore multiply that effect that operator can have.
Matt Horine: You brought something up interesting there that we talked about on our show last week, specifically the textile industry, and our focus on that in the last show was essentially about how it was the victim of globalization, where there was this massive half century swing where we baed almost a hundred percent of our own clothing to, we import 97% of it as it stands today.
And you look at some of the use cases that are going on and why tariffs were put in specifically on the industry, like a scalpel on this industry. Was because the potential for automation and for AI plugging into that where you have these old factory towns where you now have the potential to do from 1990, I think the stat was basically it's one 10th of the work [00:32:00] effort.
With a worker that now maintains their quality of life and their wage at 25 plus dollars per hour, and they're doing it in a more efficient way. Whereas before it would've been 10 times the effort, way tighter margins, all of those things come into play. So it's a really fascinating, I've liked textiles specifically for that reason.
It was gutted over decades, and now it has this potential to come back in such a big way. And it sounded like you have some customers in the space.
Bryan DeBois: Yeah, and a lot of the commodity industries, I'd love to see comeback steel and paper and all of these types of things. Yes, we wanna be doing, making chips and, and advanced technology.
We wanna do all that too, but let's also, let's not lose sight of things like textiles and some of the commodities and things like that, because I think they can be made here. I think they, we can challenge all the assumptions about how they were made in the past and go in and build a greenfield plant, like you said, that's smaller, tighter, faster.
Leverages AI, leverages automation, leverages new technologies, and is able to stay competitive with I, I think it's absolutely possible to do that.
Matt Horine: Yeah, we certainly hope so. And I, here's the on the [00:33:00] spot question and maybe a chance for a bold prediction. Looking five to 10 years out, what's your vision for the American Smart Factory?
Bryan DeBois: If we're going to stay competitive? We need to get to a level of automation that we have not achieved yet. And again, I, if you have not had a chance to go inside and I think there's videos online of Amazon Fulfillment centers, but you really need to see. Now they're not m making products, so it is a kind of a different thing.
They're moving packages and things like that. Uh, so the, the state space is a little more constrained, but it is amazing what the level of automation that they've achieved. And one of the things that, that, when you talk to companies that have highly automated processes, one of the things they'll say is, we don't want the operator to necessarily think or make decisions.
That's not their job. We want them doing what they do best, solving problems as they come up in that. But they're not really making decisions. The machine is making all the decisions and they're there to do the things that humans do well and leveraging their ability and multiplying their effort, right?
That's what good automation gets you. And so we need to get to, frankly, to McDonald's [00:34:00] level of automation. So if we think about McDonald's. McDonald's doesn't go out and hire cooks like a diner. Hires cooks, right? McDonald's hires people who have all the advantages that people have. But you've seen the videos online, right?
They're loading frozen pucks into a machine and they press a button and the machine does the work of cooking the burgers perfectly, and they press a button, they load the frozen fries in, and the machine does all the work. That's a level of automation we don't have yet in these factories. So five to 10 years from now, that's what.
It will look like where you can take a person with maybe not a lot of training, maybe who's not gonna be there for very long, but can get good product out of that machine. I think that's what it's gonna look like. And it's gonna be the adoption of automation. Traditional automation. Yes. Ai, yes. Robotics and innovative thinking.
Breaking outta some of these mindsets. Some of these are very old processes, old industries, thinking about them in a very different way and leveraging all the technologies that we have in our toolbox today. And how would we go about building a plant to make that product if we were starting out today?
Really, that's [00:35:00] what I think it's gonna look like over the next five to 10 years.
Matt Horine: It's an exciting future, exciting prospect for leaders listening who are considering some type of digital transformation or exploring AI tools, what's your biggest piece of advice to get started?
Bryan DeBois: Number one piece of advice I always say is get started.
You don't have to go and do a full AI project right away, but you need to take the first. That could be an AI readiness project. Maybe you look at some of the issues that you have with industrial networking, but you've got an old industrial network. Maybe you've got a bunch of disconnected skids. Start to get those things networked.
If you just did that, maybe you aren't one of those people who in the early two thousands got religion and started to collect data. Maybe you were one of those companies that said, why are we gonna collect this data? Okay. Now's as good a time as any start collecting the data, right? You can take those first steps today and then I'll give you one more bonus, and I know that this sounds self-serving.
Find an OT system integrator, even if it's not rosis. Find an OT system integrator that you can trust because this is a world, uh, this digital transformation, this AI world. It is [00:36:00] full of vaporware, it's full of AI vendors who don't really understand our world. It's full of these IT consultants. The big five consultants are coming down into our world and boy, they've got a great pitch.
I've heard it. But are they going to be able to actually build something that can be operationalized on the plant floor and run 24 7, right? That's the world that that you're venturing into. And so make sure you've got an OTSI that you can trust next to you who can help you break through the hype and really build you solutions that are gonna solve actual pragmatic problems for you.
Matt Horine: Great advice, and I think we'll avoid the big five and go with companies like yours on the, the operational expertise. If that is the case, where can our listeners follow your work and learn more about what robes is and doing the industrial AI next step? Where can they go to find out more about that?
Bryan DeBois: yeah, so the easiest way is on LinkedIn so they can find you there. Uh, Bryan, BRYAN, DeBois on LinkedIn, if you just search me up or RoviSys, R-O-V-I-S-Y-S. Or go to our website, rovisys.com/ai, we'll take you to our AI portion of our website, but reach out. I love to talk to people [00:37:00] about this, even if it's early on in the process and you just wanna pick my brain.
If you're a company that's looking at doing this stuff, please reach out. I'm always happy to have those conversations.
Matt Horine: Excellent, Bryan. A ton of great insights, and we really enjoyed having you on the show today and look forward to learning more about the next steps in this big revolution that we're in.
To stay ahead of the curve and to help plan your strategy, please check out our [00:26:00] website at www.veryableops.com and under the resources section titled Trump 2.0, where you can see the framework around upcoming policies and how it will impact you and your business. If you're on socials, give us a follow on LinkedIn, X, formerly Twitter, and Instagram. And if you're enjoying the podcast, please feel free to follow the show on Apple Podcasts, Spotify, or YouTube, and leave us a rating and don't forget to subscribe. Thank you again for joining us and learning more about how you can make your way.