CONTRIBUTOR
General Manager and Editorial Director,
Techstrong Group

Synopsis

In this Digital CxO Leadership Insights series video, Mike Vizard speaks with Ralf Haller, executive vice president of sales and marketing at NNAISENSE, who explains how AI is transforming manufacturing.

 

Transcript

Mike Vizard: Hey, folks, welcome to the latest edition of the Digital CxO videocast. I’m your host, Mike Vizard. Today, we’re with Ralf Haller, who is executive vice president for sales and marketing for a company called NNAISENSE. They are in the AI and manufacturing space, and they have launched this academy where you can go and learn how to apply AI in more sophisticated ways into manufacturing. Ralf, welcome to the show.

Ralf Haller: Hi, Mike, thanks for having me.

Mike Vizard: Why did you guys feel the need to create this academy? What’s driving that, and who do you think is gonna attend?

Ralf Haller: Yeah, good question. Of course, there is a lot of things out there, in particular for people who want to deep dive into programming and into learning to know how machine learning works, to learn the algorithms. Then there’s a lot of talk on the business side what AI can all bring, you know, all the management consultancy firms and these pundits. But interestingly, in the middle, where you actually have to, then, do the work, there is not that much. You don’t find that much about machine learning, engineering, and project management, how such projects are actually done and how you can make sure that they’re successful, since many of them actually are not successful, which we may be talking about in a minute.

So, we clearly saw the need to provide you something and to educate the people about what is possible, how you can do it, and what to avoid, and so, to make use of the many opportunities that industrial AI offers.

Mike Vizard: People have been moving towards robotics in manufacturing for years, now. How does AI change that equation? What, ultimately, is gonna be different?

Ralf Haller: Yeah, sure. I mean, there are different approaches to using AI. There are people out there who implement AI technology, of course, in technologies like robotics. And we help such companies as well, if our algorithms make them smarter, apply more efficient algorithms. But then you have industrial processes and smart or manufacturing landscapes, all of them are entirely different than the other, even within the company. One manufacturing plant is not the same as the other. One industrial process plant is not the same as the other; they evolved over, sometimes, 100 years or so. I’m not kidding, we have customers, like, in the oil and gas or in the chemical industry, or in the glass production or steel that are literally 100 years old. Of course, they evolved over time, but still, all these things are entirely proprietary and, in order to do something there and apply AI, you really need customized solutions, and this is what we help them to provide so that they can also make use of the opportunities.

Mike Vizard: What impact has the pandemic had on all of this? Is there more interest in AI and manufacturing because people are looking for ways to keep companies running without necessarily putting workers in harm’s way?

Ralf Haller: Yeah. Well, at the beginning, of course, when the pandemic started and no one really knew what to expect and, you know, they basically started going into wait and see what’s happening mode that, of course, impacted us as well and projects that, maybe a few months before, were earmarked suddenly got delayed or even canceled for the year.

Then, once this was more under control, we saw, actually, a lot of activities happening and just last quarter, we had maybe our most successful quarter ever. We could close four new, big customers—two of them actually in North America, one in Silicon Valley. They’re working on at least a half a dozen more for this first quarter, and so, people have used this time to look at their efficiencies, which, of course, is something that AI can provide, and are now getting out of their wait and look mode and implement things.

So, at the beginning, yeah, it stalled it, but then, actually, it helped quite a bit since it became more and more necessary to do something on this front.

Mike Vizard: What is the impact on the labor requirements or manufacturers, then? Do I still need people or are we just more or less focused on augmenting them? Where does, where’s that line between the man/machine interface gonna be?

Ralf Haller: Yeah. That’s difficult to answer in general, because every situation is, of course, entirely different. In general, we don’t try to replace people, we try to augment them, make them smarter, make them more efficient, but of course, there will be, also, situations where you automate certain processes and you can do this, then, with fewer people. So, as I said, it’s—in general, you cannot answer this in general.

Overall, we don’t see, now, a huge impact on the labor force, since these are still highly specialized people. But for them to be more productive, to be more efficient, to get that same industrial process faster into a stable situation by AI helping them, right, and complementing their, over many years, learned operative excellence is, of course, something they appreciate as well and secures their job, in fact. But then, maybe other, in other areas where this would be a completely automated process, then you might not—might need maybe less people are doing maintenance since the system takes care of that and automates that process, yeah.

Mike Vizard: Do you think that this will also, maybe, push manufacturing to the point where the goods are being consumed? Because the automation allows you to kinda be less dependent upon labor, so therefore, I don’t need to have everything on the other side of the ocean somewhere, I can actually start manufacturing some stuff locally?

Ralf Haller: Yeah. That maybe was exactly the reason why we saw more demand and now also customers signing on in the second half of this pandemic, the second year, I would say. Because yeah, they, of course, also for sheer political reasons, maybe pull out of China and bring this back to other regions and then they don’t wanna, of course, do the same, but automate these more, use the latest technologies.

And therefore, yeah, this can have an impact or might be even needed to move manufacturing back to, for example, North America or also Europe. We have a little bit of a similar situation here, although there’s still more manufacturing done here compared to probably North America, as I understand. But this—yeah, seems to be inevitable to be able to make use of the latest technologies and not fall back against your competitors, yeah.

Mike Vizard: You not only are the VP of marketing and sales, but you actually teach some of these courses. What is it that you wish organizations appreciated or understood about AI and manufacturing versus when they show up in the course in the beginning?

Ralf Haller: Yeah, yeah. Well, we have two tracks. We will offer the first track for technical management. Again, it’s for industrial use cases, mainly, or in terms of titles that we would expect, it might be CTO, might be VP of software engineering in an industrial process firm, and these people, they don’t have, typically, maybe nowadays, they start having, but traditionally, the ones that I have met so far, they don’t have a Computer Science degree. They for sure don’t have a machine learning Ph.D. or something like that.

So, for them, but they’re still very technical, so, they might have, at least, a Master’s degree, Mechanical Engineering, Chemical Process Engineering, things like that, right, or a Ph.D., also. So, they’re relatively technical, they understand the physics, chemics, and the math, but they never—and they understand some software projects may be for their space as well, but they have never, ever experienced the difference to, in AI engineering projects. So, that’s something they have to learn. And they feel uncomfortable, often, if they talk to their experts, the data analysts, right, who maybe speak a language that they absolutely don’t understand. And then they have to make a decision about something that is partly also here and there a bit hyped up in terms of expectations in particular by cloud companies that we often see.

And so, to educate them on the basic terminology to make sure that they, for them, it’s not some magic tricks that they think they cannot, that they don’t understand. We’d like to enable them to judge this better and be able to then make a more informed decision, because ultimately, they have to—they’re still responsible, right, if it doesn’t work out and if you don’t understand what’s coming, then you are, of course, maybe not doing this project. So, for us, it makes total sense to help them understand that, and since they are technical, we can go relatively technical in these courses. Of course, not on a level where we teach them to program or teach them algorithms, that would be going too far. But they understand, based on their background, many things much better and then feel more comfortable to actually start such projects, yeah.

Mike Vizard: Do you think that these efforts require or are pushing organizations to break down the silos in their organizations? Are you seeing more collaboration across IT people, manufacturing people, data scientists? I mean, it seems like it takes quite a team of folks to put together something, so maybe manufacturing is becoming more of a team sport.

Ralf Haller: Oh, yeah, absolutely. So, I mean, to break down silos is, of course, something you cannot do overnight. You need some time for that, but then you don’t have years, either, to wait until that happens, until you have maybe a data platform and data access, you know, for everyone missing out on lots of opportunities.

So, you have to do this in parallel, meaning that you have to start to do, combine, break data sets and open up silos. But then, at the same time, also start out with use cases, using these data sets that you have combined. So, it’s—yeah, it’s a multi-step process, and that’s actually also something we mentioned in this course. For example, it is not smart to, let’s say, have a four or five year lasting data platform project where you’re collecting every corner of your company, everything you can find, and then after five years, you figure out more than half you can actually not use or that you don’t need for what you wanna do.

So, it’s—yeah, it’s a combined effort of looking at use cases, of course, business-driven use cases, and we help them with that. And at the same time, of course, also build and break open these silos and combine these data sets, make them accessible, at least, through a data catalogue, you know, so that they can make use of that. Because without access to data, AI is pretty useless, as well. We need data, and the better and the more comprehensive, the easier, and the faster we can do such projects, yeah.

Mike Vizard: It seems to me, at least, that end customers are kind of forcing the AI issue in manufacturing, because they’re demanding a level of visibility into their supply chains that you really can’t do with humans, you really need some form of AI to kinda really understand how complex the relationships are between, you know, every supply chain is dependent on another supply chain. So, you don’t really know what the goods and services are without some help from some AI.

Ralf Haller: Yeah, absolutely. So, I mean, when we implement these systems, we group them in three main categories. One is inspection monitoring, where basically, we just look at, okay, things like defects, anomalies, and predictive maintenance. These are common buzzwords.

Then the second phase is the modeling of smart manufacturing or manufacturing plant or an industrial process plant or machine. So, that’s then a digital twin, but it’s purely data driven, of course, AI powered digital twins are completely different than what people—although I understand it isn’t digital twin, but we still use this term, since, in fact, it is a digital twin, we think it’s the third generation of digital twin. The second generation are complex, expensive simulators, but ours are, we collect hundreds of thousands of sensor data, control data, and also static data. And then with that, we basically learn through the data these environments and have a digital twin.

And then, on top of this digital twin comes the third step. That is where we then do, for example, user assistance systems, which we have implemented for several of our customers, but one of our customers is also an investor like Schott Glass, a leader in specialty glass production, a German company. And there, we have built a user assisted system or an operator assisted system that basically tells the operator what settings they should put so that 12 or 20 hours later, the glass quality is optimized. And for them to look at, you know, hundreds of sensor data, even with all their expertise? They get it sort of done, and have it done always, but it’s absolutely impossible for a human being to comprehend this complexity, regardless of how experienced you are.

And this is the strength of AI. Humans are better with other things, but here, you must have. So, there are lots of such situations and use cases where AI can suddenly, even a 100 -year-old process, optimize further. And they often tell us from the beginning, “Yeah, if you get 1 or 2 percent efficiency improvement, it would be already huge and save us millions.” We often get 10 or 20 percent, in some cases, even more. And this is earth-shaking for them and hard to believe. And like, in Schott Glass’ case, they then decided even to invest into us, because it became very strategic for them since this is, of course, the biggest cost factor they have, the glass production itself, yeah.

Mike Vizard: We also hear a lot about Internet of Things these days as it applies to manufacturing. It seems like it’s getting easier to collect the data that we need to inform the AI model. So, is that part of the equation?

Ralf Haller: Oh, absolutely. As I said earlier, without data, we can do nothing. Of course, we have very sophisticated algorithms that we have developed ourself. We have expertise, more than 20 years, in fact. Many of the algorithms that are used today in open source have been done by parts of our team, like in evolutionary reinforcement learning, reinforcement learning and so forth.

But, of course, the last seven years since the company exists, we have developed new things that are now used. And yeah, absolutely, that’s correct, yeah.

Mike Vizard: So, what’s your best advice to folks? What do you wish they would do before they showed up for class and what do you see as the common missteps?

Ralf Haller: The technology is here, so you can do it, but the humans are often not ready. We see here in Europe—I’m not sure how it’s in the U.S., but here in Europe, we often have management that is quit risk averse. If they’re new things, then they don’t jump at it, you know, they first wait and see what’s happening. That, of course, is the wrong approach, here. You need to start with some use case, get your feet wet, and learn, and also train your own people on things. And, most importantly, then judge by yourself how much you can do by yourself, because not everyone, in fact, hardly anyone can be a Google. And the state of the art, here, are these big Internet companies, you know, and you can, of course, not invest what these guys are investing.

Therefore, you might need outside help. We have the caliber of these mentioned companies, you know? In fact, they’re even using some of the algorithms we have developed in the past, but of course, to much more extreme and further develop that and have an army of people now. And an industrial company can just not do that. So, to just admit that, still learn how much you can learn. But then, you know, then be open to outside help when it’s needed and the time to market is then, of course, is very crucial and important.

Since IT is moving into all these industries, that is something they have to still learn. Because, for them, it’s like—yeah, it takes seven years to change the plan. It takes seven years to, for example, in the past, at least, to change a car manufacturing platform. Today, you don’t have the time anymore, right? I mean, the equation has completely changed to must set this up much, much faster and use as much technology as well as outside expert help that you can get. And maybe acquire companies, if needed; something, also, is more done in North America, still, than in Europe, as I can see. But I think that they have to learn that, and they will, if they wanna survive.

Mike Vizard: Alright. Hey, Ralf, thanks for being on the show.

Ralf Haller: Thank you very much for having me. [Cross talk]

Mike Vizard: Alright. Thank you, all, for listening to our show. You can catch this podcast and all our other podcasts on DigitalCXO.com. We’ll see you next time.

Show Notes