Beyond the AI Hype: What Actually Works in Manufacturing - with Nick Leeder
Show notes
Welcome to the Trend Detection podcast, brought to you by Senseye Predictive Maintenance – which gives you visibility and insights into all your assets, from single machines to full plants to help you reduce downtime, increase knowledge sharing and accelerate digital transformation across your organization.
What you will learn in this episode:
Why many industrial AI initiatives fail to move beyond pilots, despite heavy investment and executive attention.
How AI hype and “fear of missing out” often lead companies to start with the technology rather than the business or process problem.
Why process maturity, data relevance, and context are essential prerequisites before applying AI at scale.
How to identify repeatable, scalable AI use cases—with predictive maintenance highlighted as a strong example.
What to measure to prove success, including operational impact, financial value, and real improvements to frontline workers’ day‑to‑day work.
You can find out more about how Senseye Predictive Maintenance can reduce unplanned downtime and contribute towards improved sustainability within your manufacturing plants, by visiting: www.siemens.com/senseye-predictive-maintenance
Connect with Nick on LinkedIn:
https://www.linkedin.com/in/redeel/
Transcript
I work at that difficult intersection between technology and the processes that are used, and I sit on the process side more than the technology, so I look at how technologies can be employed in manufacturing processes to create more value, increase productivity, increase profitability of margins, and I work all the way from companies in Formula One all the way down through to smaller SMEs in the hydrogen sector. Um, and large multinationals, so my interest is really at that intersection of how technology can add value to a business, particularly in manufacturing. Welcome to the trend detection podcast powered by Siemens, and now your host, Nel Sullivan. Hello and welcome to the latest episode of the Trend Section podcast. Uh, really pleased to be joined once again by, uh, Nick Leader, a, a regular on the podcast. I say not just a regular, but, um, I don't know how, how many podcasts you've done now, Nick. I'm always trying to count with different guests how many podcasts you've done. I must be either 3 or 4 now together. Yeah, you're getting onto the leaderboard that or the podium of, um, if there's a prize on there, yeah. Exactly. Maybe, maybe I should do that as, uh, encouragement, but, uh, but no, really fantastic to have you, um, back on again. Um, and we'll talk about industrial AI, of course, topic close to your heart, close to our heart at Siemens as well, and it's really to dive into your experiences about why sort of industrial AI is a sort of a broader term, why those projects sort of fail. What, what we could do about it, we're gonna talk about, you know, you know, some of the solutions to that, of course, for people, anyone listening today can sort of take away into their own day to day work, but, um, but maybe Nick, cos it, it's probably been a little while since you've been on and for new listeners, uh, just a little reminder of who you are, what you do, etc. Sure, um, well what I do is a bit of a nightmare to really explain, but ultimately, um, I, I, I work with a number of different companies always around engineering and manufacturing of products and in the processes themselves. And really what I do is I work at that difficult intersection between technology and the processes that are used. And I sit on the process side more than the technology, so I look at how technologies can be employed in manufacturing processes to create more value, increase productivity, increase profitability of margins, and I work all the way from companies in Formula One all the way down to, to smaller SMEs in the hydrogen sector. Um, and to large multinationals, so my interest is really at that intersection of how technology can add value to a business, particularly in manufacturing. Yeah, it's interesting that you said in the sort of diversity of the businesses you, you work with. I mean, are there overlaps in the conversation we're gonna have today when we talk about industrial AI and the application of AI and some of those other industries, you know, what, what sort of similarities and differences there, massively, I think, um. I mean, it, it, it, it frustrates manufacturing companies that I work with going the business of manufacturing is actually the same irrespective of whether you're making a one-off vehicle for a race or millions of products or bearings every year, you know, look at my background in SKF Group, yeah, and it's like the business of manufacturing is the same, the technology has the same opportunity, but you need to employ the technologies and the use cases that add the value, and that's where it differs. So, you know, I think the business is the same, the focus is the same, they ask the same questions, but the solution is not necessarily the same for all industries. No, exactly, exactly, and I mean, so I guess let's, let's start us off, I guess with looking at why, you know, open it up, why do industrial AI projects fail, you know, and, and, and, and I guess we, a lot of the time we talk about pilot phases because a lot of these projects start off very small to get started and do testing, but, uh, maybe we could sort of mix those two together. Well, I think it's worth taking a little step back and saying we're in this, let's call it a halo moment for AI. Where if you're not mentioning AI in a conversation with a client, you're all or in a manufacturing in a meeting now with leaders, you're almost irrelevant, which is real absolute real shame because there are many things we can do that don't require AI that increase productivity and I'm sure we'll come onto some of those conversations today. Um, and also at the same time, there's a lot of headline grabbing pieces of work that go out. That create this uh excitement around AI and at the same time disillusionment. So you know, I, I can't, you know, I think not a week goes by where Nvidia are not having another announcement of a relationship that they're building around AI, um. You know, I think there's every company in the world is currently now a partner of Nvidia, you know, so everybody's going, we've gotta be part of this because everybody else is. So you've got that fear of missing out moment. And I, I, I won't use acronyms, but we all know what the acronym is there. But at the same time, you know, within like days of that, uh, MIT put out a report that says 90% plus of AI AI projects are failing. So you've kind of got this strange world we're living in, it's, it's the perfect hype curve we're we're going through, where you've got the excitement of the halo uh opportunity that comes from AI but at the same time the reality is very, very different on the ground. And um I think that is probably one of the root causes of why AI projects are not moving beyond pilot pilot phases, because everybody's jumping in with 2 ft. They're jumping in 2 ft and going, well, we've got to do an AI project, there's a relevance question there. Um, and then you go, OK, we're gonna do this on this one thing. And that one thing, it may or may not solve the problem, but the problem is that's not repeatable, it's not scalable. They've not stood back and said in this process in this factory, we do it 50 times, we do it 20 times a day, or we may do, uh, we may have 5 factories with the same process in there with pretty similar setups. Um, what are the opportunities to use AI? Where will it create value, and then let's focus in on that. They're jumping in almost kind of straight into proving AI works. And I'm sure we're gonna go into, there are I think some preconditions that we need to establish before AI becomes relevant. So, you know, um. A great example is Rockwell came out this year, you know, 60%, nearly what, 56% of companies are piloting smart manufacturing. But only 20% are using it at scale. So to show something's working and then to prove it a scale are very, very different gigs, um. And the global Lighthouse Network, uh, approach from the World Economic Forum, you know, they start, the people need to start thinking and they're recommending, thinking of assets use case, use cases, so things can be reused and not rebuilt every time. And again that's another flaw I think we've got with AI is we're kind of going in, proving it, but then we go, OK, but it won't work here, it won't do this because of X, Y and Z. And I think there are fundamental preconditions that perhaps are being overlooked that are are probably why AI is not at scale yet. So is it a case of sort of going in and thinking about this, this small problem here, or is it being distracted by lots of all the different use cases for AI I'm thinking where to start first, or spreading yourself too thin, I've got to do everything at once, but end up spreading yourself too thin. Well, we do see this problem of, uh, motion versus progress, uh, illusion that happens a lot. So I, I've recently talked about this in a couple of, uh, presentations I've given where, you know. There are the, the ability to spin up 50 projects is quite easy. To actually see progress against 50 projects is very hard. So what you've got right now is that there's a lot of motion to do stuff, but progress is pretty hard. And then progress, if you're not making progress, things get killed quite quickly. Or sometimes don't, which is actually another problem, but, um, so there's illusion, disillusion there, um, that or illusion that we're creating or if we're doing more things we're we're moving forward and we're not actually doing that. I, I think also from a frontline workers' point of view, you know, they're faced, you know, in the current environment versus say 10 years ago, they're faced with 5 times the amount of change that they've had. So there's change fatigue in the people as well. So you've kind of got this and at the same time because of the AI hype, we're going, we've gotta do AI and that that varies between very big programs with grand aspirations and grand statements up here all the way down to well we're gonna use a little bit of AI agentic AI and some SOPs down here. And so, and it's all disconnected. There's no, a lot of companies don't actually have a fundamental strategy with the ripple down approach. And if we get into things like the ethics of AIs, the security questions, you know, then it becomes more complicated, but right now I think we're really at that experimental stage still. So. You know, that, that formal approach and the way that we can scale is missing, that question's missing because we're experimenting and that's good, we learn from experimentation, we're not necessarily getting value and I think that's why you get 97% of Egenic AI projects are not, according to MIT, uh, uh, are not actually proving any value. And I get, I mean, some of these, yeah, some of those stats you're sort of reading off there, Nick, are really quite stark, aren't they, you know, especially when you consider the, the, the hype around AI, I know. Because the way you'd think of it, you know, it's just perfection and everything's going perfectly, and it's, I guess it's a part of the sort of social media thing as well, isn't it, that, you know, everything's talked about in such a positive light. But I guess if we dig into sort of the reasons behind it, to me, it seems like it's more than just a. Simple technology issue, you know, behind this, and we're talking like people, processes, culture, and what you said before about, you know, that gap between what's being set up here and what's being said down here. So, there's this big, let's say, vacuum of information in the middle from what you've described. I thought that's quite interesting. Uh, I mean, look, let's start with a bold statement, yeah? Automating an efficient, oh sorry, automating a broken process. Just means you break things faster. So Let's take a step back. AI requires actually processes that work. If the processes don't work. Then you're automating something that's fundamentally flawed. You're using capability. So I think the process debt within companies. If you're trying to put AII over it, ultimately becomes a tax later on because it fails because the process is broken and it's not gonna fix, let's be honest, AI will not fix a broken process. That's people with process knowledge, the data that you need. And actually that's another key component is that. You need to think about is the data, what data we need to make AI actually work or your use case of AI work. Um, And obviously the process is the precursor to getting the right data points that you need. And again, I see that manufacturing does not suffer from a lack of data. What it suffers from is relevance of that data and context of that data. So actually pulling that together in a way that says right, we have a good process, we have relevant data that we trust that's providing context. Um, I think those two things, if you don't have those, then your AI project is ultimately going to be at risk. Um, and actually just, uh, I'm actually gonna make a quick call out on, um, SESM in the North America. They're doing the Smart manufacturing interoperability, uh, program process or SMIPs. Um, and I, I, I, uh, this is really interesting for me, um, interoperability platform, sorry. So SMIP is the Smart manufacturing interoperability platform by Sesme. And It, it looks at a number of core capabilities and this is actually the things, this is where you've got a world of competition and different vendors and legacy applications. By the way, this stuff is really cool on a greenfield site site where you build from scratch, but most of us work on brownfield sites with a legacy equipment, legacy processes, so. The the SMMIP approach looks at connectivity, yeah, how do you get the data from the different sources, legacy machines, systems, old historians for example, um, how that data is ingested, so how, you know, connectivity then acquiring that together, as I mentioned a minute ago, providing the right context in a consistent way. That context gives you ultimately a springboard for scale. Because a motor of a certain type in one factory, if you can then connect it and say that's the same motor in a different factory, you've got ultimately Uber context that provides around that. You know, security of that data, the workflows that data goes through maybe even gets enriched and how it moves between the, you know, some of the, the systems and the OT components you've got. And doing that in a neutral way so we establish some standards that enable AI, you know, I had a wonderful conversation with a company called Think IQ and Doug Lawson and Paul Turner, um, the other week and it was actually last uh last Tuesday, uh, a presentation in London they were giving, uh, at the Quorum principle event that was really going over a little light bulb moment going, why is that not there in manufacturing? And the problem with standards like this is standards are like toothbrushes, yeah. Everybody agrees we need one, but nobody wants to use anybody else's. So, um, I'll let that sink in for a moment. Um, you know, so I think something like that Smith initiative by SsMe is really, really important for the data and context. But actually let's get the process right now if you've got those preconditions and you've got a workforce that doesn't feel threatened or, you know, cos there's an, you know, the other problem with this bubble of AI is it's a link to workforce reduction as well. Um, that doesn't feel threatened, uh, threatened by AI, but actually sees this as an opportunity to enhance their work, and that's your job, by the way, if you're trying to do AI projects to position that correctly with the frontline workers. I remember we've talked about this a couple of times before about actually buying and making the value relevant to them and not just the boardroom. That's, that is then the right environment that you're creating for AI projects to be successful. Yeah, and I think, I think as well, it falls into this, again, everything you, you've said is, is correct. It, it just makes you think a lot of the, the projects can sometimes just be sort of like, you know, theater. You know, it's, it's just, uh, something to showcase that we're, we're in innovative, we're doing AI, but like you said, without some of the fundamentals, is it, is it a case that some of the fundamentals, you know, you mentioned connectors, you mentioned data, like some of that's. Maybe not forgotten, but sort of being put in the shadow a little bit, not as cool at the moment because AI is here, but those are the things that you need to have in place to actually, uh, apply AI on top of it, that's what you're saying, right? Yeah, it's the, it's the boring things you need to get right to make the exciting things work and get value from it. Yeah. Your, your, your car, you know, you get attracted by all the kind of sat nav and all the, the funky stuff you get inside of it, the self-driving features, the fact I was in one the other week, it's not mine by the way, had a cup holder that was heated and cooled. I mean, my mind's exploding, what a great cool feature that is. I can't find, I can't think of a use case and by the way that's a great example of like AI kicking off. Why do you need a cup holder that chills or heats something up? But if the engine doesn't work and you don't put oil in the engine, I'm, I'm working on combustion, not electric, uh by the way, in my example, but yeah, if you don't put the boring thing and maintain that engine, the car doesn't go. So yes, you get a hot cup of coffee, but you're still stuck with where you are, and primarily the main reason why you drive a car is to go from A to B. So if you're stuck at A but with warm coffee, it doesn't help where you need to get to, which is to get to B. So it's always been the case with technology, and unfortunately I've been around too long now that we forget about the boring foundation bits and we get attracted by the shiny hype. And where I am seeing AI have value is where it's being deployed in scalable use cases where they spend time and effort getting the process right. Getting the data right and then the use cases are straightforward to actually implement and people have trust. People have trust in then what the computer says. The computer says this and we know the process is working, the data's right, and they're able to validate what the computer is saying. Uh, that's where I'm seeing it, so things like predictive maintenance is actually those sort of use cases are scalable. You can provide very simple context and it makes the workforce's lives better. They make the operators' lives better because the machine is running more smoothly. Maintenance moves out of kind of predict uh preventative firefighting mode and into predictive modes, yeah, for the brake fix world they live in. So their lives improve and things happen then from there. Importantly, there's a tangible value case to that as well, you know, unplanned downtime is very expensive. So it's that kind of compound effect and the environment that that creates. So we move away from the theater and we get to the practicals, um. Yeah. So, I mean, uh, I mean, maybe let's, let's go into that a little bit then, because a key component of it, you know, everything we, we've talked about, you know, what type of projects to look, look at, look at it at scale as well, which again in a predictive maintenance world's important because we know the value of predictive maintenance in a Siemens or a Sensai world is, you know, at scale, but. What are sort of the measurements companies can put in place, I guess, to, you know, to see this, and then we can start with predictive maintenance, but of course, maybe there's other use cases we could talk about as well, but what, what type of measurement could they put in place for these types of projects? Uh, first of all, let's call out, there are 3 measurements you need to put in place for these things. Um, And the first one is a fundamental disconnect in most manufacturing companies. Which is the operational metrics, so things like defects, cycle times, unplanned downtime, um, those things. Safety, let's let's let's mention safety. I can't not mention safety, but. Those things are measured. They're often though measured in percentages or other values. Now OEE is a classic there, yeah, it's measured as a percentage which ultimately is quite meaningless as a statement. And people will shoot me down for that maybe, but I think a percentage of nothing is still nothing, yeah, um. And actually there's a disconnect then, first of all between the operational KPIs and the board level KPIs. You know, the, the harder metrics, you know, the return on, uh, assets, the net assets, yeah, those sort of things, the EBITD statements, the financials ultimately, which we look at what was and step one is to get metrics agreed and understood operational, but actually then establish a very strong metric connection between that. …