Gen-AI Meets Determinism: Bringing Speed and Certainty to Complex Product Configuration

Welcome to the Industry 4.0 Club's fireside chat. The club's purpose is to engage the power of diverse worldwide talent to accelerate the global evolution to Industry 4.0. We accomplish this by hosting, supporting respectful, thoughtful, and interdisciplinary open conversations, and creating quality content on wide-ranging Industry 4.0 topics to expand the reach of our community. My name is Mike Ungar, and I am the CEO of the Industry 4.0 Club. I have over 40 years of leadership and management experience, most of that in manufacturing at Michelin. I'm joined today by my co-host Mike Yost, who's one of the co-founders of the Industry 4.0 Club. Mike, would you mind introducing yourself a bit further, please?

Of course, Mike. Thank you. Welcome, everybody. As Mike said, my name is Mike Yost. I've been in industry almost 40 years myself, with the first half of my career on the industrial solution provider side and the last half on the industrial nonprofit side. So Mike's co-host here on the fireside chats as well as the volunteer chairman of the organization. So really excited to continue the conversations that we like to bolster here by speaking with Laura Beckwith, who is the Director of Product Management for ConfigIt. We're going to speak to Laura about GenAI meeting determinism. To bring speed and certainty to complex product configuration. So Laura leads innovations that help manufacturers manage the growing complexity of configurable products. Her work focuses on combining advanced technologies like generative AI with deterministic configuration models to ensure accuracy, traceability, and repeatable outcomes in manufacturing environments. Laura is a strong advocate for responsible AI in industry helping organizations harness AI speed and flexibility while maintaining the rigor required for engineering and product configuration. So we welcome you, Laura.

Thanks.

So before we get into actually our discussion with Laura, I wanted to highlight that ConfigIt's hosting their 11th Configuration Lifecycle Management, or what we'll call CLM, Virtual Summit on May 6th and 7th. This event will focus on how manufacturers can master customization, complexity by establishing a strong data foundation for future growth. Held virtually over 2 half days, the summit brings together B2B manufacturing leaders to explore how connecting systems, data, and logic enables organizations to manage complexity more effectively. Discover how CLM approach links systems and configuration logic across the enterprise, turning product complexity into confident decisions throughout the product lifecycle. Anyway, Laura, so we're really glad you're here to talk with us today. I mean, I'm already like— I was— as we were talking before we got started, you know, this is quite an interesting and new topic for me, so I'm excited to learn more. So let's just try to get into talking about what GenAI is. And, you know, we hear things a lot about that seems to be quite a buzzword today. Maybe you just introduce GenAI and and then we'll get into a little bit about what determinism and manufacturing, how this all fits together. Would you mind?

Yeah, of course. I think actually I'm going to start a little bit with the configuration problem just to frame a little bit there before we move into the GenAI part. Because I think one of the, you know, when you were talking, Mike, and introducing about this, I thought, gosh, these configurable, complex configurable products, like what, that can mean a lot of different things, and are people getting the right picture in their head? And one of the examples I really like to think about is a car, an automobile, that all of us may have had the experience that at some point we have to buy a new car or buy a used one. Maybe we've gone online and we see that there's a button on a website that says configure, and we click on that and we start to, to make some choices. And essentially a car, you know, there's— it's complex, right? There's a lot of choices that you as a consumer can make. Underneath the scenes, there's a lot of regulatory rules. There's a lot of differences in how something can be engineered, how it's being manufactured. All of this complexity, you know, that's kind of— it's fundamental to something that consumers order, right? They want to get the car that they want. They want to get that electric SUV. They don't want it to be too expensive, but they certainly want it to be able to tow their boat up to a lake house. So they've gone through some configuration and they start, you know, choosing this, making some selections, and then they say, but I need that better towing. And all of a sudden, there comes up a conflict. Well, you can't have that. battery, you need a long-range battery because that's what's going to make it possible for you to take your boat up to the lake house. And on the front, it might seem kind of simple, right? Like, hey, these are just some combinations, but it's hiding a lot of complexity underneath. And now you might kind of wonder, okay, well, where is the GenAI coming in here? And I guess for me, this is like, even if you could put a more chatty-like interface, you know, I could instead of making selections on the computer on a UI, I might talk to it or write to it and say, hey, I want an electric SUV. I want to really, you know, I want to keep it inexpensive because I don't know, you know, there's some uncertainty in my life. But I absolutely, you know, I really like my boat and chat that in there. And if I, you know, someday I'm able to place an order that way, the manufacturer wants to know, and this is where the determinism comes in, they want to be sure that they can produce your car and it will be legal, it will be possible to manufacture it. We don't have to come back to you and be embarrassed that, by the way, we sold you a car without the right size battery. And this requires this, what we call determinism, this idea that even though GenAI can be very, you know, it deals with probabilities, what is the probable outcome? Probability that you want the big battery, not so much because you said you want to keep it cheap, but you need to have it because you want to tow something. So that is, that's the determinism underneath. And the GenAI is great at trying to map like what is it that you might need. So maybe that's a big mouthful, but that's kind of the way I'm thinking about it.

Well, it's really important too. Having grown up in the automotive tier 1 space, I understand what you're talking about and the implications that it has in the production environment when you choose one option versus another option. And if all 3 of us order the same type of vehicle going down the same assembly line, it could have completely different connections and configurations and such for what each of us wants. And I think, as we talk about Industry 4.0, one of the things that gets really exciting about it is it really lets and encourages companies to be more customer-focused. So all the things that you're talking about as far as a front-end configurator and using AI to help you find the right configuration for what you want at the right price point, those are all things to better serve the customer, right, which drives the opportunity. And then the back end underneath it is where it seems to me the CLM, the configuration lifecycle management, comes in to manage that complexity. So is that the right way to look at it?

Yeah, exactly. I think that Basically what we're saying is that, you know, manufacturing, they need to know. Probably correct is not good enough. Like, I do not want to have mistakes when my car goes down the manufacturing line. I want to make sure that the right battery is there to put into that car. And maybe yours is next and it needs to have a different battery. Like, it needs to be deterministic. I need to know. And You're absolutely right that when we put this layer on top of it, that, you know, that's what our customers are expecting is all of this to make that experience even better for our customers.

Yeah, and to go further from, you know, my personal experience, if we get that wrong in the assembly process and it shuts down the assembly lines, you're talking about tens of thousands of dollars of penalties per minute to fix that kind of a problem because it's so important to keep that from getting out into, you know, a finished product out in the marketplace. So yeah, it's— yeah, so that the concept of determinism, making sure that it's right, that it's always right, is really an important point that you're highlighting. Exactly.

Yes. Yeah.

Excellent. So yeah, go ahead.

No, I did. That's— you got me thinking on my background in the tire industry. You know, this idea of thinking about, do you want long life, comfort, good gas mileage, you know, and being able to then actually— I mean, today we— you have to look up a table and there's different— but being able to actually configure that based upon what you would like, knowing that there are some trade-offs. Going back to your point about long battery. In order to get that, you can't have— you can't always have everything, but knowing where the limits are and then being able to actually request that, and then us being able to actually customize attire to the individual's needs rather than, you know, here are 4 choices, pick one. Yeah, pretty cool stuff when you think about it.

It is, it is. And I think also, I mean, it's also interesting because of course when we're doing all this GenAI stuff, I, I'm playing on on, you know, some large language models, and I'm saying, hey, let me model something kind of relatively simple and see how far I can get. And one of the challenges is, I mean, it seems to do quite well up to a point, but I have a very hard time when I then ask it, oh, could I get this feature and this feature together? And it, you know, does some reasoning, and it's like, but is it right? Like, is it absolutely— like, can I believe it. And, you know, I'm then going back to our software to say, is that right? Oh, it did pretty well there. But the fact that I have to go back to our deterministic software in order to figure that out also shows that, you know, this is the human aspect. We need to be kind of aware of how we can get that variability, but then we need to be like, where, where, how do we know the truth, right?

Yeah.

Well, it's interesting because there's always the trade-off of what's the reward for doing what we're doing. So this idea that we're going to introduce AI, that we're going to make the customer experience more configurable, we're going to make it more automated, et cetera, et cetera. There's a commercial value for that, right? That companies, I think, see and everybody's excited about. We're having these global AI races, right? But then what's— how do you protect? And so that's the point where you're driving as far as the importance of the complexity management on the configuration side, and then, you know, having this deterministic backbone of your software. So can you talk maybe a little bit about how your approach and having that deterministic infrastructure helps manufacturers get the benefit of AI and managing that risk that you were just talking about?

Yeah, certainly. And I'm going to switch a little bit. So it's very interesting, we can think about AI in different parts of an organization or different parts of the process. So, so far we've been talking more about the selling side. What is our, you know, in a manufacturer, we want to have that determinism because we don't— probably correct isn't enough. But we also have AI on the definition side. And actually, that's the area that at the moment we've spent quite some time looking very deeply in. And in this side, you talk about, okay, so fine, when I get out there, that, you know, I can't have the towing unless I have the big battery. But who is it who wrote that rule? Where did that rule or that constraint come from? And those constraints, they come from all over the place. So, when we talked to our customers, when we're out there, of course, many of them are on this digital transformation. At the same time, many of them are using Excel, they're using some technical spec documentation. This is where they're recording what is it that our product does, where is the variability, what can go together with what. And, and when you take that information, it's— somebody has to translate that into the language that our VT technology, our virtual tabulation tongue-tied at the moment, the BT technology, what it is that that enables. So there's a step between all of this spreadsheets and PDFs into what, what we call a product model, this thing that has all the rules about what can go together. And we're using AI there to say, listen, your data's unstructured, your data's in a PDF, like it might be in a specific format, but there's natural language in there. That is what AI, GenAI, is amazing at. It can take pieces of that or take the PDF, feed it into this tool that we've been building called ACE Prompt, and get that. You let the AI along with the tool that is trained in product models, what is it that these look like, and bring that together. It will make suggestions for you like based on this information, it seems that there's a battery and the battery comes in different sizes. Oh, and there's a relationship and it turns out there's a relationship to towing and maybe there's a relationship to the heated seats and maybe there's a relationship to all sorts of other things, of course. Let's start building up this product model from that information. There we're using AI to make suggestions into how we could what you might want as rules. But here, it's also super important that this VT technology is also saying, by the way, if we write these rules, here will be the output of them. Here are the combinations that will be allowed or are disallowed. And that is sort of that determinism backbone, right? So AI suggests something, the determinism of the underlying technology maps it to you and shows you straight away, this will be the impact if you enter that rule into the system. And you as a human then get to make that choice. Does that make sense?

Well, it makes sense to me because to your point of things in spreadsheets and, you know, one of the things from my background being primarily in the manufacturing execution world growing up, we would use common language ERP vendors and MES vendors and PLM vendors, and there's a common language that people would use. And oftentimes those decisions about what size battery goes with what product and stuff, oftentimes those were dictated by the system that you bought, right? So you have an ERP system and it tells you that this is how things should work and what they should be. And we always contended as the MES folks in the plant making this stuff happen that that's a suggestion or a best guess, but really this is what's happening. So it sounds like you're applying tools, modern tools, and forward-looking advances to help with that process, improve that process. And I guess the question I have is, who do you typically work with? Like, who sort of owns your solution set, or who do you partner with?

Yeah.

Is it on the product side more or in the manufacturing world or?

It's sort of— the honest answer is it depends. So we can be— in many organizations, projects like the ones that we run with our customers, they might be the manufacturing/sales side where they say, well, you know, we're not— you know, we needed a new CPQ system. And then you work your way backwards and you say, well, where is the data coming from? Oh, well, it's sitting over there. By the way, the engineers, they have their PLM system and they give us some data and then we need to remodel that and then we put it in this system. We've worked with different parts of the organization. The challenge, I guess, that you could say with CLM is that we're trying to say your silos of engineering and of MES or of a CRM system or of an ERP system, all of the silos, you need very similar data, you overlap. Like, you know, maybe you don't overlap on all of it, but you have to meet each other. And basically what we're trying to say is, no matter who brings us in, can we figure out how you overlap? And shouldn't we have some of that engineering data so you don't have to re-implement that? And shouldn't we have a bit of that manufacturing data also? And all of this is so that your ordering system, maybe your CPQ system, maybe an online ordering system, is actually accurate and you're not having to redo a bunch of— A bunch of work there. I think that, yeah, so, but it isn't always easy, right? Because we have these silos.

Yeah, yeah, yeah.

It's, you know, I was thinking about your example earlier about the towing and now you need a larger battery. You can imagine with an AI system that comes back to you and says, Now, if you have a larger battery, you know, for no cost, you can have these things as well, you know, and you begin to, wow, I didn't know that. I mean, I get, you know, and I get this, I get, and, you know, so it can be really, I guess it's kind of an exciting thing. The other, the other thing you caused me to think about, so I spent a lot of time deploying lean in my career, and it really points to, for a manufacturer to really leverage this all together, to have a a single-piece flow or being able to actually create, you know, the more, the more flexible your manufacturing process is, the more you can leverage this sort of opportunity. Am I thinking about that correctly?

Yeah, I think so. So one of the, one of the things that we like to talk about, about, and actually to the CLM Summit and like this data foundation wording, you know, there's something about You know, are we, are we a P— are we trying to do what PLM is doing? Are we trying to do what ERP is doing? And the answer is always no, because, you know, engineers, they don't want this, the, the stuff about how you want to sell it, dear marketing team, right? We know you want to put in packs, but we don't want to put that in our PLM system. That is not really relevant to us. But on the other hand, it's, it's a bit Like that engineering data, they need to understand that if I replace this battery, it might have follow-on effects towards the manufacturing or towards the marketing team that they're completely blind to. And I think that part of saying, getting this foundation of the data and trying to build up, you know, bringing this data into this centralized vision, the CLM vision, is that now it will be easier to do all of this. Like, it will be easier to do the analysis of what's the impact, like how many departments are gonna be affected by this if we make this change, because you have it all, you can bring it in. You don't have, you can still author it over there and just make sure it's coming into the bigger picture and then do that analysis. And I think that that's— I think that's pretty awesome. And that establishing that foundation is what will enable a lot of stuff to be able to— transparency with AI. It will enable, you know, more classic AI or GenAI. It will enable new discussions with the data because it's all here. And whether or not that's for speed or for— yeah, for intelligence of What data has been configured? Like, what is it that people are wanting to do out there? All of those kind of questions that most organizations are hungry for.

Yeah, I just want to take a minute with everybody and remind you that we're speaking with Laura Beckwith, who is the Director of Product Management for ConfigIt. We're getting into conversations around complexity management, configuration lifecycle management, leveraging GenAI to meet the demands of, you know, increasing speed of of markets and opportunities and balanced against the complexities of product configuration and such. And so it's a— there's a lot of it. One of the things that always gets me, Laura, is that like the level of depth that we're talking about is obviously very high at this point, right? And when we talk about complexity management, we can say, okay, I get that. But like, you change a bolt that's going into the engine block on a car, like you talked about, the ripple effects of that are enormous. And especially if you're tracking and tracing that kind of information, and like, there's— so it's very easy for us to sort of go, yeah, I get the concept of that, but the levels of complexity here and the costs associated with it, to Mike's point about in production with lean manufacturing, are severe. So I think it's just a really deep deep topic. And I guess that leads me to the point that you have your summit coming up, and, you know, we're only going to talk for a few more minutes here today, but, you know, you have an opportunity for people to go deeper into those, you know, topics, which is obviously very, very important. So, well, and I guess so if I could just— I wanted to come back to a point that we had talked about earlier with regards to AI becoming more capable in engineering and configuration workflows. And then how do you think about accountability, or how should manufacturers be thinking about accountability when you're talking about what should be trusted to AI, what should remain governed by these deterministic models, where does human judgment remain essential? Can you talk us through your thinking on that?

Yeah, so I mean, Of course, the thing that we're all hearing a lot of is make sure we have a human in the loop of some of this. But of course, it's not quite enough. My background is in user experience. I think a lot about how people use software. One of the things I'm noticing is just how easy we can get caught up in the AI said it. We don't necessarily know how to challenge it. It's so confident. The way I like to think about this problem is you really— I think about it in roles. So the AI, it's fantastic, it's powerful, it can speed things up, it helps to generate options, it can interpret data, it can do all of that. But sometimes it's a, it's a bit too confident, it's a bit too sure of itself perhaps. But then this is where we use this deterministic model. So for us, we, I mean, we live in the world of Boolean logic. I mean, it's down to the— at the bottom, it's like true or false. Everything becomes true or false. And at that determinism, every time the AI makes a suggestion, we can run it through our little engine and say, dear user, the AI suggested making this change. Here's the impact on it. This is sort of what I referred to earlier. And find a way that we can use exactly what humans are strong in for owning that decision and giving them the information they need so that even if the AI comes back and is, you know, very confident, That you have that information right there. So from a psychological perspective, it's like, well, great, thank you very much. I can see here that's not actually quite what I meant. Let me reframe what I'm asking you to do. And that's the area that I think is super exciting, that, that we can leverage what it is humans are really strong at and—

Yeah.

And boost them up in that capability.

Right. No, that's a great, that's a great example. I think you hit it on the head, you know, using AI to take care of the things that are rather, that don't require us to think, get that out of the way and then put it in front of us where we can actually think and then reflect and make those decisions. That's where, that's the value add I think that you're talking about. That's wonderful. So you're going to be speaking at the conference, I understand.

What do you hope people are going to learn by attending the I, I, I'm excited because, you know, just like in this conversation to some degree, we've talked about making sure our foundation of the data is in place, right? And the, the summit is going to talk about, you know, why that's important, different ways of getting the data into place, and also then why it matters, what you can do with it later. So I think there's, you know, some great speakers that are going to be there on the technical day. We have a number of people from the product management team who will be talking about how to think about this. I think one of the other things is there are a lot of potential existing customers or people who are working in manufacturing with pretty complex products, but they might approach it, they might approach it in the way that they just make a ton of variants. And then, you know, like a TV manufacturer, they don't seem to have configuration. They seem to make just, we have 500 variants. Well, there's pros and cons to that. Is that making it easier for you to sell your TVs? So I hope that people will come and try to understand when is it that going toward a configuration-to-order process would make sense for my organization and what are the future benefits I can get out of that.

That's great. Well, we've been talking for almost half an hour now. We're getting ready to wrap up, so I think it's important to share maybe a final thought for our audience. So Mike, would you mind sharing maybe a final thought first, and then Laura, we'll let you have the last words?

Sure, appreciate the opportunity. You know, I think that we're barely scratching the surface of this topic, and I understand why the summit— I guess 2 things. I guess this is the 11th year you've hosted this summit. And I think the importance of education, of awareness, of having dialogue around these conversations is so critically important. So I would certainly encourage people, even if this just sounds a little bit like there's something that might be worthwhile to understand, to take a look at the summit. It is virtual because the focus on education, there's so many layers of depth here. And so I think it's really exciting. Again, growing up in the automotive Tier 1 space, I'm sort of vibing with what we're talking about here, but I'm also sort of scared of, you know, where GenAI is and all that stuff. So I would certainly encourage people to learn more, to follow Laura and her teams to dig into this topic more deeply.

Thanks, Mike. Laura, how about you?

Yeah, I think it's exciting. I reflect a little bit with you, Mike, that, you know, there's a lot of unknowns right now. And I think that one of the things that we who are in the industry and are exploring GenAI about what it can and can't do, how we might be able to use it, that there's excitement, there's a bit of apprehension. And I think that trying to find these middle grounds in between where we can still feel like we're not giving it all up, we're making sure that the humanness also comes through in some of these applications that we're building. I think that, that to me resonates in my heart.

Well, thank you for that. And, you know, I just want to take time to thank you for taking time out of your busy schedule, Laura, to join us. Without question, your enthusiasm for this topic— I mean, I'm motivated, uh, to participate now, you know. So I really appreciate your enthusiasm and really your insights today into really how GenAI can meet determinism and, and how this all impacts, uh, the management of complexity. So thank you for that. We just want to remind everybody that if you'd like to sign up for the summit, here's the website to go to. It's a virtual summit, as you've heard today already. It takes place on May 6th and 7th, and please register. I think you'll learn a lot and enjoy, you know, hearing from folks like Laura. So the Industry 4.0 Club would like to encourage you to take action. Again, attend the summit, attend a workshop, read an article. Or even start a smart manufacturing roadmap for your factory. We'd love to hear from you, hear what you're doing. So please drop us a note on our website at www.industry4oclub.com or at our LinkedIn page. We also host conversations like this weekly. So if you want to be on a future broadcast, have something you'd like to share with our audience, we'd love to have you. Again, reach out to us on our website or our LinkedIn page. And with that, Again, Laura, thank you so much for, for your insights. Mike, you as well. And we'll, we'll wish everybody a great rest of the week.

Yes.

All right. Take care.

Thank you so much.