Fireside Chat with Rick Oppedisano

Good morning, good afternoon, everyone. My name is Mike Unger, and this is— welcome you to this week's fireside chat sponsored by the Industry 4.0 Club. Our Industry 4.0 Club, we exist to help promote and advance the deployment of Industry 4.0. We host meaningful conversations throughout the week, some on Clubhouse, some here on Industry on LinkedIn Live. Mondays we host Women in Manufacturing. Tuesdays we host— Tuesdays we host the Future of Manufacturing. Wednesday is our fireside chat. On Thursday, we focus on people, process, and culture. We are really excited today to speak with Rick Oppedisano, and I hope I did a good job with that, Rick. Hopefully.

Better than most teachers.

Excellent. So, so Rick is the president and CEO of Delta Bravo AI. Delta Bravo started in 2016 and today works with manufacturers like Rolls-Royce, Nucor Steel, and Continental to turn data collected on the plant floor into predictive models that increase throughput and quality while reducing waste and downtime. So Rick, we're really glad that, that, uh, we're really glad you could join us today. So, Rick, I was looking at your background and noticed that you have a poli sci degree. And I think that's really interesting because I was really curious, how'd you go from political science degree to today working in AI with manufacturers?

Yeah, it's a funny story, right? So number one, I appreciate you having me on, Mike. It's always a pleasure talking with you. Yeah, funny, funny story there. So, you know, it started out, I thought I was going to go down that path, the political science degree to law school path. And I had a job offer right after college and drove up to D.C. to go take it. And unbeknownst to me, the job offer had fell through. So I showed up, parked out in front with all of my earthly belongings in my 1981 Honda Accord hatchback and went in. And they said, Guy— I said, Rick, you know, I guess you didn't get the message. So, you know, it's no longer available. So I ended up going back to my parents' house in upstate New York and, you know, started, you know, got a couple of odd jobs and found myself living at home with no end to that in sight. And, you know, I was just basically one Camaro away from a terrible Italian stereotype, you know, living at home and listening to your mom all day. So, you know, I ended up connecting with a buddy of mine that was one of my college roommates who had moved down to Raleigh, North Carolina. His father was a former IBMer, and so he had moved down from upstate New York as well. And so through him, I had the opportunity to interview at IBM, and it's kind of a funny story there. At that point, I'd never written a line of code in my life, and I was interviewing for software, entry-level software engineering job. And so the guys at IBM said, hey, you know, we like you, you know, you fit in here, you know, personally or culturally. But we have to have you pass this technology test. You know, you got to show your stuff. And so I said, okay, well, when do you want to do that? And they said, next week. So I had about 7 days almost exactly to kind of figure out and prepare this. And so I went and I was quitting, you know, I was bartending, waiting tables, doing all the things you do right after college, right? To pay the rent.

Yeah.

And so I quit those jobs immediately and went to Borders at the time and bought the Sams books for the programming languages that I knew I was going to be tested on. Borrowed my roommate's IBM, I think it was like a PC/2 or something, and did 20 hours a day on programming books. Basically subsisted on pasta and coffee for about 6 days and went in there. I passed the exam. I think I passed it by 4 points.

Wow.

Yeah, the guys were like, hey, you know, it's great that you passed, you know, this is awesome, but we thought you would have done better. And I said, well, guys, I'm just a nervous test taker. You know, but, but, you know, yeah, just to show you. Yeah, just a short year or two after that, I was helping them, you know, I was working on design teams, you know, doing, doing things and really getting into it. You know, I really— it turned out that, you know, getting into software engineering was for me. So that was kind of how it took—

Yeah.

Yeah, that's kind of, you know, it's kind of a funny thing, right? You don't hear too many stories like that anymore. But the— yeah, so that's how I kind of got into technology. From there, you know, worked for IBM for a little bit. I had a great experience there. you know, moved into eventually starting my first business in 2001 and, you know, went through a process there. It's kind of interesting. So at the time, the idea was to, you know, use, you know, build software to do kind of internet marketing solutions, everything ranging from, you know, building websites to email marketing platforms, marketing automation, kind of the early days of that. And, you know, kind of started to— that was really first first really realized the value of being able to tell stories with data. You know, that was really the first kind of dose of that. And then, you know, was able to, you know, build that company, you know, and exit successfully. And then from there, you know, got into some more kind of, you know, enterprise software type positions. I worked with a company where we had software that remotely monitored and managed, you know, Microsoft collaboration platforms.

Yeah.

And then from there, you know, really this is where, I'd say around 2012, 2013, this is where things really started to kind coming together for me in the data side of the house. I worked with a— I was on the— I was an officer in a publicly traded company, helped run that company, and we focused on modernizing legacy data stacks. So this is moving huge banks, insurance companies, government entities from mainframe data constructs, you know, procedural applications, non-relational databases, COBOL, things like this, moving them from those legacy monolithic historic data platforms into something more modern. So, you know, moving those, you know, data stacks from, you know, your natural ADBAS, you know, VSAM-type platforms over to your SQL Server, you know, Oracle-type stacks. And what we found was that, you know, we'd be doing this with whole governments, right? I mean, you know, these massive structures. And you'd be able to start giving people access to tons and tons of data. And right around like 2013, 2014, it became pretty obvious to me that people were just getting overwhelmed right there. And just because you had access to this data doesn't mean you really know what to do with it or how to create value from it. And if you try to do that the wrong way, you end up getting into essentially anecdotal arguments and conversations the same way as you might have done if you didn't have that data at all. Sure.

Yeah. So that was really what— was that the trigger point for you to think about Delta Bravo?

Yeah, yeah, for sure. That was the trigger point where I started to say, okay, you know, what can be done? So I ran R&D for that business. And so I was in charge of kind of figuring out, you know, how to use technology to solve the customer problems. And so I just saw that as a value add, right? You know, what can we do to help these customers process that information and turn it into something valuable faster? So, you know, looked at the usual stuff like, you know, rules-based platforms and things like that. And the level and the volume of information that we were looking at was so great that, you know, I was like, you know, really the only thing that looks like it can work here would be machine learning or AI. And again, you know, the concepts of AI, machine learning, they've been around for decades, but the computing power and the actual availability of the technologies, you know, is relatively recent. Right? So, you know, I started looking into that at that point and said, guys, you know, I think this is something that, you know, that we should either build into our products or start a separate subsidiary to focus on. And it wasn't met terribly warmly.

Okay.

And so at that point, you know, I started getting a lot more passionate about that and believing that that was the course to take. And so I had an opportunity to exit the business You know, did well with, you know, with my stock and my involvement in that business. And so I was able to leverage that experience to bootstrap Delta Bravo. Started it in 2016. Really, you know, first got the business, you know, I'd say, you know, first customers and really getting moving with employees is kind of like early 2017.

Okay.

That's kind of like— that was how we got off.

Yeah. Well, that's quite a story. You know, I guess we all face twists and turns in our life and our careers that sometimes we're, we're surprised. Did you ever really wonder, am I— were you doing the right thing? I mean, you know, from, from where you started, did you ever like say, oh my gosh, what have I gotten myself into?

Oh, yeah. You know, I'd say every time around, right? I mean, but, you know, are you talking about, you know, every decision, right? No, but we— this has been a really incredible experience. And to your point, it started off with a healthy degree of confusion. So, you know, number one is, hey, we've got this incredible idea. I think there's a ton of value here. Well, the first question is, a ton of value to who, right?

Mm-hmm.

So, you know, when you think about something like, you know, the concept of AI or machine learning, it can be applied horizontally, right? You could do this for predicting, you know, financial fraud. You can do this for predicting the weather. You can do this for, predicting emergency room visit volume to help doctors staff the hospital, right?

Right.

I mean, so there's all these different ways that you can apply the value of prediction. And so, you know, we kind of fumbled around with that for a little while at the beginning. You know, we just started saying, okay, we need to kind of get into this and start figuring out where that value of prediction is highest, right? And so we did those things. You know, our first big customer was AccuWeather. We, you know, we're processing at 1.9, 20 to 40 40 billion data requests a day, you know, to help AccuWeather improve their weather forecasting algorithms.

Wow.

Then we, yeah, then we, you know, help them integrate those weather forecasts with their customers' data to help their customers look at weather forecasting and project things like foot traffic or anticipated staffing hours, right? So we really got into a number of these, like, you know, we worked with a financial services company to help them predict credit risk. And so we started to, you know, we did a lot of stuff kind of horizontally at first. So the first thing we had to do is we realized that number one, we can't keep doing that because if you want to have some level of domain expertise that provides value above and beyond your technical expertise, you've really gotta focus. So the first thing we did is we said, okay, where does the value of prediction become high? Where's the highest value of prediction? So what industries value it at a premium? So, you know, financial services obviously values that at a premium, but, you know, the barrier to entry into financial services is relatively low. There's a ton of people that are getting into—

Yes.

You know, predictive stuff for financial. And the reason why is 'cause the data sources are relatively consistent and minimally complicated, to be honest with you. We looked at that and said, you know, definitely a premium value, but, you know, low barrier to entry, lots of competition. Yeah, let's pass, let's move on, let's try someplace else. Now we ended up on manufacturing, you know, really for 2 reasons. Number one is that the value of prediction manufacturing is super high, right? Being able to predict and avoid machine downtime directly traces back to value, right? You can put a number on that.

Yes.

Being able to predict and reduce scrap, you can put a number on that, right? And then when you think about what's going on, you know, with the supply chain, you know, we've gotta make more with what we've got. You can't keep on wasting things. We don't have enough people to deal with the rework, right? So, you know, our people need better tools, right? You know, the people that are getting hired into manufacturing now, the auto, you know, these are the kind of people that want that advanced thinking, that advanced analytics, that, you know, all that stuff that you have to think about.

Mm-hmm.

The stuff that 20 years of experience gives you, right? Now, they want to take all that information and have that automated and presented to them, right? So we looked at that and we said, all right, there's a lot of opportunity to assign a premium value for prediction in manufacturing. So that was kind of how we started with manufacturing. And then I'll tell you what was interesting from there is deciding what kind of technology we had to build.

Yeah, I can imagine there's some choices to make there in terms of platforms and And, you know, I don't know enough about the landscape for AI, but I can imagine that it isn't the same for each provider, huh?

Yeah, so it's pretty different. So, you know, really when you look at the whole kind of process of doing what we do, you know, we focus on, you know, pretty much everything we do is basically 5 flavors. Plus it's just different flavors of the same 5 use cases. probably the best way to say it. So, you know, look at, you know, predictive maintenance, predictive quality, right? You know, scrap reduction, throughput analysis, right? So getting, you know, identifying and smoothing out those throughput bottlenecks, you know, demand forecasting, inventory movement forecasting. These are things that we do kind of, you know, for all the clients, right? So everything we do is kind of just a varying flavor of those things. And so when you think about that process, right, the process starts with bringing data in. So you've got to connect to these data sources, right? That, you know, there's a level of custom development involved in that, right? Because every manufacturer's going to have different data sources, you know, collecting data in different frequencies and different formats, right?

Yeah.

And there's going to be varying levels of access that you get from a compliance standpoint, a security standpoint, right? So there's some, you know, that in itself is kind of a high barrier to entry. Then you have to bring that data in and you say, okay, well, now that we've got all this data together, how do we put it together? How do we clean it? How do we join it? How do we get rid of data that doesn't matter for the use case that we've got? How do we take data that's collected once a minute or once every hour and sample that down and connect it to other datasets that are collected 3 times a second, right?

Mm-hmm.

So how do we deal with manually collected data? How do we identify error ratios or outliers in these types of data sources? So it's— so once you go through that crazy process, right, and then you've got a data source that's ready to be modeled. I saw you yawning. Look at that, man. I know you're getting to get the level of complexity that's—

No, it's crazy.

No, I'm just thinking the guy that's interested in this board. All right. So—

No, I was, I was gonna— I was thinking about when you described that. Yeah. The level of statistics involved in what you do is quite, is quite significant, huh?

Yeah, yeah, it's a pretty big deal. And then, you know, and then again, even when you get past that, you know, here's the real— where the rubber really hits the road is when you get to the point where you've got a reliable, you know, predictive model. You know, congratulations on your fancy machine learning model. How do you get that in the hands of operators or business users?

Yes.

To make— and then how can you track and measure that it makes a difference every single day, right? So there are all these, you know, all these questions around, you know, How do we take this extensive and complex process that involves a ton of integration points, that involves a ton of subject matter expertise? How do we attack that? And so the number one thing that we said is, number one, you know, we've got to focus in on manufacturing as our, you know, as the area that we are going to develop subject matter expertise in.

Yes.

We are able to leverage partnerships with folks like the South Carolina Manufacturing Extension Extension Partnership, which is a great group down here. We're headquartered in Rock Hill, South Carolina. So we were able to leverage the experts there. And these folks are experts in every bit of manufacturing, whether it's lean, you know, financial operations. I mean, these guys are, you know, they're unbelievable.

Right.

So, yes, we kind of got a crash course from them. You know, within that partnership, they brought us into customers like a Rolls-Royce or, you know, several others at the beginning. You know, Continental was another one that we did with them. And, you know, Josh, Fiber Industries, we had a number of them that we, you know, that we got brought into through that partnership where, you know, the environment was that we had an expert from the SCMEP, we had an expert from the client side, and then, you know, gradually we started to really learn the dynamics and the language of the customer, all the different pain points that go along with it.

Yeah.

And that's really what informed the development of the technology.

So that's a Yeah, that's really interesting too. And I think I'm glad you mentioned SCMEP. They really are a fantastic partner. I'm in South Carolina also, as you know, and, and, you know, they're certainly one of the leading MEPs in the country by far. They do tremendous work. So, so the— we talk about the value for manufacturers. What kind of, what kind of What value does it bring? Does this bring to manufacturing? What kind of things have you seen in terms of benefits for them?

Yeah, so it's a good question, right? So a lot of people in my space are gonna focus on kind of some sort of short-term measurable ROI when they give the answer to this type of question, right? And for me, I think that's how you can kind of identify an amateur, if we're being honest. So once you do enough of these, so Mike and I, we've done upwards of 80, 90 of these projects, right, over the years. And so, you know, we've gotten to this point where we start, we really kind of see, I guess you could say, we have a degree of pattern recognition around customers that achieve the highest level of benefits.

Mm-hmm.

The ones that achieve the highest level of benefits look at the return kind of in, I guess you could say, almost like a trinity style, right? So there's obviously going to be a measurable ROI to solving a problem, right? So if you look at a predictive maintenance scenario and you reduce unplanned downtime by, you know, 50%, there's going to be a measurable ROI that's associated with that. Reduce some, you know, like with Nucor, we're, you know, reducing scrap by upwards of 10%, 15%, something to that effect.

Right.

There's a high measurable number that you can track. Now, that's one aspect of it. And if you focus on that, I think you're selling yourself short. All right. There's 2 other aspects that are real benefits that not enough people really focus on. So number one is measurable. The second one is a capability ROI. So when you think about this, it's how is this AI or machine learning capability driving a better experience for the employee? So we all know that everybody's struggling for hiring right now, right?

Mm-hmm.

How do we use this technology to turn one person into 3? How do we use this technology to reduce the onboarding or training cycle for new employees by 50%, 75%, right? How do we give them these tools? Yeah, these tools that perform complex multivariate analysis and give them recommendations on how to make a better decision faster, right? So when you think about that capability, you know, it's the whole idea about workforce readiness. You know, people think about AI as a, you know, quick operational improvement component, and it is, but you also have to look at it from a workforce readiness standpoint.

Yeah, a lot of people, You know, a lot of people are fine. You know, you hear the kind of a— for those that aren't so informed, sometimes they hear them saying, you know, technology is going to take away our jobs.

How do you—

how do you, um, if you go kind of another level down, how do you engage employees? Uh, how do you continue to engage associates on the shop floor when you have an AI implementation?

Yeah, so a huge part of it, right? Um, because that is where the rubber meets the road. So you have to really kind of learn how, you know, how people work, right? You know, uh, how and where they can make changes in the process, right? If you say, you know, this model says that we have to make a change at this point in the process, but there's no way to do that without ruining the batch, kind of a useless model, right?

Right.

So what we, you know, our process typically involves, you know, it starts off with a process diagram. What are we doing? What data is being collected at each stage of the process? How and where can an operator influence change, right? That's kind of where it starts for us. And what will, you know, we kind of operate at the 80/20 rule these days, right? Where, you know, our folks are experienced enough where if you give us a dataset and a process diagram, we'll probably understand about 80% of what's going on.

Right.

The, you know, the 20% is what makes that manufacturer unique, right? In every manufacturing process, right, there's always going to be trade-offs, right? You know, hey, you know what, the data says this, but you told me this. Well, that's because, you know, this supplier, we kind of know that their stuff's a little bit saltier than this supplier, right?

Right.

So yeah, so there's always those little trade-offs in there. And that's that 20% where you need some level of involvement from a subject matter expert or an operator, right? So, and the other side of this too is this stuff's all iterative, right? You take your best guess at it, your models become smarter and more accurate with more information, more tuning, right? More context. So everything completely, it's iterative, even down to the user interface, right? I mean, with Rolls-Royce, You know, we started off with a completely different model set than where we're at now, a completely different interface than where we're at now, and that's been going on since 2019. So, right, so you've got kind of like 3 years of iteration. And so I'd say that that's, you know, that's definitely from a time commitment, we've got it down now. So again, big part of the value that we bring is the fact that we've got domain expertise, but we also have process expertise.

Right.

So we, you know, so I would say we typically like to have about maybe 5 hours a month involvement from the subject matter expert just to kind of, you know, continue to, as the solution's built, deployed, and managed, usually around 5 hours a month tops. And that kind of tapers off after we're into production. That goes down to like, you know, 2 or 3 hours a month. So we don't necessarily require a ton of time from the customer resources. We just know where to put them and how to use them to kind of maximize that.

Oh, that's good. So I noticed, you know, you shared with me, um, a little bit about Delta Bravo, and I noticed that in your— you talk about the challenges you faced the first 3 years of the company's existence, and then something happened there where all of a sudden, in around 2019, your success really changed in terms of your ability to help your, your clients. What did you learn in the first 3 years of the company that you applied since then to drive success?

Yeah, that's good. It's a good question. So I would say that, you know, the first— when we started out fairly early, right, people were familiar with the Industry 4.0 concepts, familiar, you know, with the concepts of AI and machine learning, but no one had really kind of really done anything yet. And the big challenge there at that time was that our customers had no preexisting frame of reference around whether or not this could work, how it would work, the level of commitment that would be required from their resources, and then the level of cost commitment. So there was absolutely no preexisting frame of reference. It's not like buying a server, right?

Right.

So, you know, that was a challenge at the very beginning because, you know, it's pretty difficult, right? You end up, you know, becoming essentially an R&D kind of exploratory investment. And so since these things take time to kind of show ROI, Right. You know, that was another kind of a challenge there, right? So when you're an R&D investment and it's taken time to kind of show the ROI and it's iterative, it's difficult moving from R&D into kind of the enterprise usage zone. Then on top, so we had to kind of figure that out first. And so in doing that, you know, we just, you know, I'll tell you, Mike, we did engagement after engagement where we'd get it to pilot. And it kind of got frozen, that pilot. And so we just started learning how to ask the right questions. We started working with, you know, again, getting back deeper in with our partners like SCMEP, the other guys, you know, that we work with, and started really breaking down the roles that we needed to talk with, whether that was an operational excellence person, VP of quality, continuous improvement people, and really starting to understand how and where we could fit into their charters, right? Started to understand scalability. And then, you know, at the same time, this is all happening while we're building our product. So at some level, you know, we were kind of changing the tires on a moving car, right? You know, building a product while onboarding people into that product, continuously updating the product, right? And continuously updating everything that was being delivered to those customers. And so it was just, I would say, just, you know, what we learned there. And again, there's some real numbers, you know, if you take a look at our our business from 2017, 2018, you know, I would say that we— I know these for a fact, these are actual numbers. So, you know, we had about probably 3% of our engagements had negative ROI at around the 24-month mark. About 5% of the engagements had neutral ROI. 6 or 7% of the engagements were kind of slow ROI, right? So there were some outliers there that really weren't providing the returns that the customers were looking for. Yeah. And so, you know, there was plenty that did, but there was definitely a subsection, you know, that weren't really stacking up, you know. So we had to go back there and kind of evaluate those. And if you look at where we're at now as a business, if you look at our, you know, 87, 88 projects that we've kind of calculated on, I would say that, you know, we're— I can tell you again, we're around 85.5% positive ROI within that 24-month period. You know, if you look at 2019 through, you know, you know, 2021, you know, we're, you know, 40, 45% of those projects are driving positive ROI within a 12-month period. So, so what did we learn? Number one, process maturity. All right.

Mm-hmm.

So when you get into this, you have to have frameworks that allow people to be brought together to establish consensus, to understand, prioritize, and score their use cases. So one of the struggles that we had at the beginning is that we would talk to these R&D guys or process guys, but we didn't necessarily bring IT in. We didn't necessarily bring in finance, right? We didn't necessarily bring in, you know, supply chain folks. You know, we learned, you know, we learned that, you know, when you bring these people in, they usually don't talk to each other in real life. They don't really know each other that well. And, you know, what they— what, you know, there's no real benefit from one to the other. So we sort of started to really kind of study that and put together these frameworks that would allow us to position the capability in a way that would help them understand the benefit and the influence and the impact of each other, right? We started putting together frameworks that would help them establish their own level of governance internally around, you know, who would participate in this, how much, what level of value they should expect from it, and how it would scale. So I would say number one is just process maturity right there. That was a huge deal for us. Another, then at that point, you know, really after that, it's product maturity. Our product is, is top-notch. You know, we're able to bring that data in, process it, model it. You know, I'll say one of the big things that we're able to do now is, you know, we're able to take these models and either express them to an interface within our application, or we can integrate them into an interface that the customer already has. And if you do that, you reduce the, you know, the learning time, right? You reduce that shock value that the operator has in learning. I'll say, Mike, so When we first started working with manufacturers, it was like, hey guys, that sounds great. It sounds great that you can solve this problem, but listen, our operators don't wanna learn anything new. You know, they don't have the time. Yeah, absolutely. You know, who's got time? Yep, and so we figured out really from the beginning that if we're gonna create these predictive models, either, you know, if the customer has an interface or has a process that we can just fit 'em into, do that. Or if they don't, be able to create an easy-to-use interface that, you know, that can be customizable to their process. And that's kind of how the application's built. So, okay, it just allows you to get to the value quicker, right? Because the value ultimately is how do I use this predictive model to influence the decisions that I make every day?

Yeah, and that's a great comment. So, do you— how do you create credibility in what your predictive model says on the shop floor? Because that's a, you know, that's, you know, I'm, you know, I can imagine a situation where someone's— they used to always do X when Y occurred, right? And now you tell them to do like Z when Y occurred, and they're like, no, we always do X, you know, what's the deal with that?

Yeah, so it starts off with just kind of an evaluation of the data. So a lot of times, you know, what you're talking about is what we would describe as kind of tribal knowledge, right? So tribal knowledge is usually based on experience, you know, just kind of human pattern recognition, right? And a lot of it's hunches. And to be honest with you, Some of these, some folks out there are really good at this, you know, really good at this. And the trouble with them is that they're gonna be really hard to replace, right?

Yeah.

So what we do is we'll go in there and we'll take all the data from all the different pieces that influence that outcome, okay? And we'll put that data together, clean it, join it, and we'll do what we call a correlation analysis, right?

Right.

Yeah, so the correlation analysis takes the historic out— the history of the outcomes. Let's just say we're talking about machine downtime.

Right.

We're going to take all the different machines that are in that process, in that line. So let's just say there's 5 or 6 machines generating information, right? We're going to put all that data together and say, hey, a correlation analysis tells us that as this metric over here goes up, that affects this metric, you know, 2 stations down, driving it down. When that metric gets driven down, the probability of a negative quality outcome goes up by 15%. Is that important? Is that something we can control, right? So very, very quickly, usually we're able to do that within a couple of weeks, right?

Interesting.

Where we'll have that level of analysis. Yeah. And so what we're able to do is, number one, kind of really help the customer understand historically how the data works together to drive particular outcomes. Sometimes those outcomes are good, sometimes those outcomes are bad. We're able to help them understand that as certain metrics move up and move down, there are different probabilities for those outcomes. So if I raise, you know, ingredient A and reduce ingredient B, the probability of a good yield goes up by 18%. So we're able to give them that kind of understanding into how the data works.

Mm-hmm.

And so usually when you start with that, and again, you bring it against historic data, there's not really a lot of credibility to prove, right? You've got the credibility there because you're not saying this is gonna happen, you're saying this happened already. And here are the factors at the atomic level that influence that outcome.

Yeah.

Right? So we start there and then at that point, you know, when you start doing the modeling, you know, so, and I'll tell you what's another thing that's interesting at that point with correlations is that the customer will come back and either say, yeah, man, that's totally what we thought, that validates a lot of our experience, or they might come back and say, shoot, man, you know, I didn't see that thing. Yeah, that one factor over there, you know, that's something somebody might have brought up. We thought that was a one-timer, but, you know, Or maybe they might associate it with a supply chain, you know, some sort of a difference in the supply chain, right?

Right, right.

So we see a lot of that now, especially 'cause you got, you know, again, with all the difficulty in supply, you know, people are getting out there trying new suppliers, starting to make stuff on their own, things like that. And so, you know, we start to see a lot of those differences come in as the supply mechanisms or raw material suppliers change. So, but to answer your question, When you build the model, this is kind of fun, right? So when you build the model, you do this and you test it on historic data. So what we'll do is we'll train the model on, let's just say, 85% of the historic dataset. So the model's in there, it's seeing information, it's making pretty accurate predictions. We use a couple of different statistic models to measure accuracy. MAPE is a good one, AUC is another one. So we go in there and we use these different kind of statistic measuring mechanisms to rate the accuracy of the model depending on the context. Context, right? So we'll train it, you know, train it on about 85% of the historic data, then we'll start throwing data at it that it hadn't seen. And usually when you go through that, that process at the very beginning, and if you can get that 70% accuracy neighborhood, that's a really good sign. That means that, hey, um, with some continuous tweaks, with some continuous tuning, with additional context, we might be able to get this thing up upwards of 80-90% accuracy.

Wow.

And if you're doing, you know, you know how the deal, you know how it is in manufacturing, man, if you're making the right call 8, 9 out of 10 times, right? Yeah, that, that's, that feels good, you know?

Yeah, no, it's based on facts too. Yeah, based on the facts also.

That's, you know, especially new operators. Yeah, especially new operators. You give these guys these tools, um, you know, who, who wants to be in a job where they're making mistakes for the first 6 months, right? Um, it's, it just kills motivation, right? Kills your credibility as a teammate. So if you give them the ability to, you know, understand these probabilities and make these higher I guess, higher success probability decisions sooner, that makes them happier in their job, takes some of the stress out, right?

Yeah. So thinking about, you know, you mentioned some of the companies you've worked with, Nucor and Rolls-Royce, to name a couple. And I'm guessing you also deal with some smaller, I'll call mid-sized manufacturers as well. So what's it like? What's the difference in working with someone who's a bit smaller versus large? Is there a difference in that? I mean, because usually the large guys will have Fair amount of internal support resources. The small to mid-sized guys won't have that. Is there a difference for you or not really?

Well, yes, it's funny. Each of them has their nuances, right? So let's start with what they have in common. Zero experience doing this.

Okay.

So, so it's funny, everyone reads all this stuff about Industry 4.0, and everyone talks about it all the time. You see, if you look at some of the 8-Ks or 10-Ks for the, you know, the big publicly traded manufacturers. There's always some content in there about smart factory, right, this type of stuff. But the reality is there's not enough people out there. Um, you know, when, when Ford and Tesla and, um, you know, GM and BMW jumped into this game, they were jumping into this game competing for the same resources that Facebook is after, that the banks are after, right? So there's a, there's a limited number of resources out there. So number one, Whether, you know, what we have found is that, you know, if you look at a Nucor or if you look at a Rolls-Royce in this context, or another one like Precision Global down in Greenville that makes, you know, pump components, you know, or we've got a number of other ones that kind of fit this bill that are smaller, midsize. Elchem Silicones, another good one.

Right.

These guys don't have internal resources that understand these processes or technologies. So the number one thing is is that they both lack, you know, the frameworks and understanding of how to deploy, manage, and scale these things. So that part's the same, right? Now, where things get different is usually the investment in data collection, all right? So the bigger guys, they kind of were in that Industry 3.0 movement, right? Where you had a lot of automation components coming in.

Right.

People are buying stuff, spinning off PLC data, right? Um, they may or may not be collecting it, uh, and storing it, right? It may just be sitting on the, you know, on the, the HMI for 24 hours, then it gets wiped, right? Which is, you know, not a great idea. Um, but, um, but ultimately, the, the guys that, that tend to be a little bit bigger have a little bit more of a significant investment in that space, right?

Okay.

So when you get a little smaller, yeah, you know, some of the guys that would get a little smaller, um, you know, they'll, they'll have, they'll have some of those gaps that are manual collection, right? Um, you know, that type of stuff, or Or the biggest mistake I see, you know, kind of, you know, biggest, I guess, real challenge that I see across both of them is that they'll drop these lines, you know, the machines onto the line. The OEM will say, hey, here's your PLC data collection, you know, your ratio. You should be collecting these metrics at, you know, half hertz, right? Because we gave you this really light storage component on the machine. And if they collect it at half hertz, it won't overwhelm the storage component. If it's networked, it won't have an effect on your network, right? Now, if you wanna do a predictive model, you're gonna have to collect data at a higher level of granularity. So in the case of Rolls-Royce, we had to move them from half-hertz data collection to 10 hertz, right? So you have a scenario now where you've gotta have a bit of a network upgrade, you know, you have to have a storage component. So the bigger guys tend to be a little more ready for that than the smaller guys. So with the smaller guys, you tend to look at like kind of more of a roadmap approach where what you do is you spend a little more time on kind of defining that 18 to 24 month roadmap, specific use cases, specific influence on impact on technologies, surrounding technologies, and you give them a little more breathing room. So you just move a little more iteratively with those guys because they're not usually prepared financially or staff-wise to handle some of those kind of ancillary changes that might be required.

Well, and that's, that's really it. That was really one of my key questions as well, is if I am one of those mid-sized guys or small guys, how do I know I want to maybe pursue this? What are some signs that I need— I know that this could really benefit me?

Yeah. So let me talk about some mistakes. All right. The signs are obvious. Read something, right? This is, you know, this is Industry 4.0, right? You've got a measurable ROI, you've got a capability ROI, you've got a strategic ROI. This is something that we didn't talk about before. You know, I've got customers that look at this from a strategic standpoint that say, hey, how much more valuable does my business become with this capability? Can I go and acquire a competitor that's moving slow on this, bring them into this, drop my architecture on them, and increase the value of that acquisition exponentially? Right? There's a strategic component to this, right?

For sure.

So we're seeing that. Yeah. So when you say, okay, let's think about, you know, the mistakes that they make. So If you're a small to midsize manufacturer and you're saying, hey, I'm not ready for this, I need to keep on collecting more data, I need to keep on working towards perfection before I make this jump, that's a huge mistake. Listen, if you've got anything coming in, there are ways to use this capability to make your life easier today. So a good example, again, going back to Precision Valve or Precision Global down in Greenville.

Yes.

You know, they have a number of machines of tons of different ages, right? They have 50, 60, 70 machines. that are, you know, ranging in age from, you know, 1 or 2 years old to 15, 20 years old.

All right.

So they have data that's being spun off. And so what they were doing is they were taking this data and dropping it into spreadsheets, big-time manual process for putting all these reports together and understanding OEE, extremely time-consuming, extremely manual. And so again, do they need to, you know, they didn't really need to do anything other than what they were already doing. We took that data, we threw it through our models, and we automated that entire process. So they're getting a higher level of quality data quality, right? Because the machine learning model is taking this stuff and putting it together, putting it into a spreadsheet. It's all automatic. They can see it in graphs and charts within the application, or it can get exported as an email into a spreadsheet. And I bet 5, 30 hours a week.

Yeah, I bet they— I bet from the shop floor perspective, you know, the folks that run the equipment are happy not to be doing that data collection anymore. I bet that's really a win for them.

Right. You know, it gives them time to actually go back to concentrating on what they're good at, right? It gives them time to start building the culture, train employees on, you know, better information, right? So yeah, so I think that the big trap that the smaller guys fall into is assuming that they can't start now. And given how busy the smaller guys are, it's way easier to kick the can.

You know what I'm saying?

Yeah. It's easy or easier to kick it. I mean, it happens to us right now. I mean, I talk to these guys and they say, hey, you know, we've got this problem, but I'm too busy. It's, uh, I was talking to somebody the other day, they, they, there was some— oh yeah, it was, uh, it was, hey General Custer, I'm saying it was, hey General Custer, General Custer, I've got these machine guns, I've got these machine guns, General Custer. And General Custer's like, oh, I'm too busy for you, man, I, I don't have time to talk right now, right? So that, that's kind of— that's the situation that I see a lot of the small to mid-sized guys falling into, right? Is because— is that they're, they're so busy They're not seeing the forest from the trees. They're, you know, it's an easy, convenient excuse just to kick the can down the road. But the problem's not going away. The problem's only getting worse. So at some point you gotta stick your— get your head out of the sand and address it.

So for those small and mid-sized guys, then do you end up partnering with like an integrator or do you provide that sort of support that goes beyond where they just don't have time to manage Maybe the projects themselves because they are so busy with their day-to-day.

I know we're white glove. We, you know, we'll do the whole thing for them. So, you know, there are certain things that Delta Bravo does not do. All right. So we're not going to go and install sensors on the line. Right.

Right.

You know, we're not going to go and, you know, do a number of, you know, I guess you could say kind of physical stuff. Right. That's not what we're going to do. So we have partners that we work with that can help us kind of hit those types of things. But ultimately, we do provide a white glove service to the customer. And so we'll manage the project, we'll do all the reporting. Like I said, typically between 3 to 5 hours per month is what we would require from our customer in terms of their involvement. So it's not, you know, if they're serious about it, look, if you want to solve a problem, you're going to spend 3 to 5 hours a month solving that problem. You know, that's pretty good.

Yeah, it's a pretty good return on time.

Yeah.

Yeah. Well, that's great. And I thought it was interesting when you mentioned something earlier too about the ROI. So if I'm going into an AI-type project, I've got to really be planning to, you know, you mentioned a lot are getting returns in the first year, but if I really want to understand the longer-term benefit, I've got to give it a couple years. Is that a fair comment?

Yeah, yeah, definitely. And again, don't focus on the financial ROI only. Focus on the workforce readiness component. Are you reducing your training time? Are you creating a culture that'll turn these employees into a longer-term asset, right? Are you giving them the tools and processes that they can use to have less stress at their job, be better at it, and become more valuable to you faster? Well, so when you think about, you know, all those things, you know, that's really where we like to get folks moving on, you know.

Yeah, that makes a lot of sense. You know, I think a lot of folks view, you know, we talk about manufacturing today and the challenge that manufacturers are having in hiring and retaining employees to work on the floor. But this sounds like to me, as you're describing it, that with implementations like AI, you end up making the shop floor a lot more attractive for folks and a lot more interesting as a place to work.

Yeah. You know, and then if you can do that and you can combine the results of operational improvement on top of that, You know, it's, it's, it's something— this is something that American manufacturing and global manufacturing is going to have to do to keep up. American manufacturing is going to have to do this to be able to compete. Right. Again, this is just reality. This is where we're at. If you're, if you're not looking at doing this in the next year or 2, sell the business. Sell the business.

Well, that's a— yeah, that's very interesting. The It's so important to making these advancements. In fact, when you think about manufacturing here in our culture, in our society, we are a high-cost labor market. And so if you're not making these sort of gains and if we're not really taking the lead on this sort of thing, then it becomes much less tenable to remain, you know, in manufacturing. I know we were talking earlier about that. You know, can you comment a little bit about your viewpoint on the importance of manufacturing in our society and Today?

Yes, this is definitely something I've gotten a lot more passionate about over the years. You know, I was surrounded, um, you know, I was exposed to it when I was younger. My father, uh, helped run an Army base that produced, uh, cannon barrels. They're a manufacturer of cannon barrels. And I got, you know, several cousins and friends that, that work there. Um, and so, you know, so I was kind of exposed to it at an early age, and, you know, I didn't really fully grasp the value until we really started moving with Delta Bravo. Manufacturing is a critical sector of our economy, right? And if you look historically, as manufacturing struggles, the culture struggles, right? When America is building and making things, America is kind of in a good mood. That's kind of— if you just look back, history speaks for itself in that regard. So I think it's critical, number one, just from a cultural standpoint, a critical economic sector. People need to work. People need to be making things, producing things, you know, feeling good about what they do, having some level of gratification around what they're doing at work every day. So, so for me, you know, from a personal standpoint, I love manufacturing for those reasons. From a— I guess you could say from a professional standpoint, we got to do it, guys. If we are not making our people more valuable, if we are not reducing waste, if we're not reducing our energy footprint, these Industry 4.0 things, if they're not happening, you know, we're going to get swept away. I mean, if you don't— I mean, we have partners in Asia, so we're aware of what's going on with manufacturers, you know, around the world. And these guys, you know, in Asia particularly, they're— this is part of how they're being built. You know, if we're not, you know, if we're not keeping up, we're not keeping up. You know what I mean? Like, it's pretty Pretty real. So again, going back to this, if you are not planning on deploying these capabilities in your plant, I urge you to sell your plant now because it's just gonna get harder and harder. It's just gonna get worse and worse. You know, the value of you, what you do, and the value of your customers is just gonna get harder and harder to maintain if you don't start putting this stuff in place.

Yeah, thank you for that insight. There's a question we have from the audience, and Scott asked a great question, and I wanted to ask you if you could give us some feedback on this. He mentions that the small to mid-sized manufacturers, you know, who are interested in getting started in AI, you know, they're challenged because there's a lot of folks that claim they do AI out in the marketplace today. How do they— how should they go about vetting a company, an AI company, in terms of being able to choose one? What would you suggest?

Yeah, hold on one second, Mike. I just had a— my phone started going off in the headphones, and I thought I had disconnected it from the Bluetooth, but There is no escape. No escape. So question again. So the small— this is about a small manufacturer.

Yeah. Small to mid-sized.

How do they vet an AI vendor?

Exactly.

Yeah. So number one, I would say, you know, body of work, right? Body of work. Have they done it before? If you're small, you have way less margin for error, right? You cannot afford to hire these consultants that become, you know, earworms that just stay with you and keep on telling you that your baby's ugly. and not how to fix it, right? So number one is body of work. Have they done this before in your field, in your area of manufacturing with someone your size or bigger, right? The reason why I say your size or bigger is because the guy that's gonna be your size or bigger is gonna probably have a degree of complexity or scale that you are gonna wanna have eventually.

Mm-hmm.

So you're gonna wanna, so yeah, definitely number one, body of work relevant to your area. And then, you know, the other side of that is flexibility, right? You know, so when you take a look at how these engagements are run, all right, if you come in with a fixed price and a fixed scope and fixed promises, I can tell you right now that's going to be a recipe for change control and a credibility buster. Because what will happen is you'll get in there and you'll start to realize that this dataset is not exactly what we thought it was, or this model that we thought would produce this result We've got to change that a little bit for this different use case at a different point in the process to get the same value, right? So there's a lot of flexibility. This is data science, right? There's a degree of experimentation that has to happen. So beyond the experience is going to be looking at the process for that vendor. Are they deploying this using the agile process? Is there a degree of flexibility in the statement of work that allows you to change without getting dinged on change controls? Yeah.

Oh, good. Thank you. I appreciate the insight there. You know, as we get ready to wrap up here, we're getting close to the end of the hour. I did want to ask you—

Time flies when you're having fun.

I know. I want to turn our conversation a little bit differently now. You know, the future of digitalization and things like AI is really quite tremendous, and a lot of new careers in that area. What advice would you give to young people who want to get involved in Industry 4.0 and AI as they, you know, who may be in college today or want even to change careers and get into this? What advice would you give them?

Well, there's the great, great field. All right, so number one, yeah, that's the first piece of advice I'd give them is do it. But when, you know, I guess, yeah, you know, when I say do it, it's because the challenges out there are so interesting and diverse and the need is so strong that if you're good at this, you'll always be working on something that's kind of cool, kind of interesting, and you'll always have value, right? So I really feel strongly about this as a career for our best and brightest. Now, the advice that I would give them coming out, there's 2 pieces, right? Number one is know the technologies, learn Python. There's a million resources out there for for learning Python, learn the principles of data science. Start and then, you know, so come with the ability to be able to, you know, do ETL, right, extract, transform, load, this is the data cleansing and joining stuff. There's a million resources out there that you can start to learn how to do that, takes a while, so be patient. And then the second bit is start to work with somebody that focuses on communicating this stuff with a customer. So everyone, you know how it is, everyone rolls out of college, you know, they don't— they may have some level of skill from a, you know, practical standpoint, but they don't have the experience in communicating at a professional level.

Mm-hmm.

And to do this work, communication is a premium value item. The ability to communicate these concepts to an engineer or to a management person or to an analyst you know, that, you know, the ability to, again, just communicate the value of the data to those different audiences is a premium value item. So I would encourage them to get an internship at a place like us, you know, or even at a manufacturer where they start to learn that language as soon as they possibly can and be able to understand how to communicate that value in the way that they feel comfortable. Because that, I mean, that's it. If you can, if you've got the technical side of the house, Mm-hmm. communicate it and understand the value and the risk and the process, man, you'll be, you'll be writing your own ticket.

No, that's excellent. I think I really appreciate you emphasizing both the technical and the communication skills. I think that's sometimes we miss that element of how strong you have to be in communication in order to really be effective in this area. So thank you for that. So as we get ready to wrap up, you have any last thoughts or last comments you'd like to share with our audience today?

Oh man, I'll tell you what, we, we did a pretty good job of covering everything, you know. Uh, I, um, yeah, I would say that if, um, you know, if our audience here is, is local, um, you know, South Carolina is the state to be. Um, you know, look at folks like the South Carolina MVP, look at, I mean, the Fraunhofer Institute. There's a number of different, um, uh, options out there that if you want to dip your toes in the water, um, and you don't want to, you know, take it, make a huge financial You've got partners out there. You know, we have partnerships with those agencies that we, you know, we can leverage some degree of funding to help offset those those investments. So there's no real, you know, again, if you're in other states, maybe there's those options. I can definitely speak for the state we're in. If you're in South Carolina, there's no there's really no excuse not to to start looking at this. Like you know, the OPEX Forum. I mean, is another great one. Mike Demos, what he's doing over there. You know, those guys are doing great work, tons of value. So there's a lot of different ways to get the train rolling. You know, the only thing that you can't do is not get the train rolling. No, you know, thank you.

Yeah, I appreciate that. So I just, again, I want to thank you, Rick, for spending the time with us. Just to let folks know, this has been the fireside chat that we do every week with Industry 4.0 Club. Our club hosts meaningful conversations. Monday we do Women in Manufacturing, starts at, um, 12:30 PM Eastern Time, sometimes on Clubhouse, sometimes on LinkedIn Live. Tuesday we do the Future of Manufacturing, we focus on a technical topic, again at 12:30 PM. Wednesday is our fireside chat, normally starting at 11:30 AM Eastern Time. And Thursday is People, Process, and Culture, which normally starts at 12:30 PM. Again, sometimes on Clubhouse, sometimes on LinkedIn Live. We've been really excited today. We've been talking to Rick Oppedisano, who's the president and CEO of Delta Bravo AI. Rick, can't thank you enough for taking your, your time to join us and share your insights about AI and Industry 4.0. Thank you so much.

Yeah, thank you, Mike. Sure do appreciate the opportunity. It's been a lot of fun.

Yeah, thanks a lot, you guys. Everybody have a great rest of the week. Thank you.