AI Without the Hype: A Practical Guide for Mid-Size Manufacturers - with Nick Leeder

Trend Detection Podcast ·

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.

Why it’s time to move beyond the AI buzzwords—and how to focus on solving real industrial problems with technology that delivers value.

How starting small with predictive maintenance or AI projects can build trust, prove ROI, and create momentum for wider adoption.

Why simplicity and usability matter more than complex data science—and how making insights accessible drives real operational impact.

The essential role of high-quality, contextualized data in the success (or failure) of any AI-driven initiative.

How to avoid common pitfalls by linking AI projects to business KPIs, building scalable plans, and being honest about what’s working and what’s not.

Real-world advice for industrial leaders on scaling AI and predictive maintenance across global operations—without getting lost in the innovation theatre.

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

Transcript

So, you know, when you kind of start in that small wing approach, you know, again back to the lighthouse, um. These transformations tend to happen 25 to 50% faster when you kind of start small and scaling. Um, and that, that, that comes from remaining focused on what you're trying to prove. And accepting that you're gonna do this in an iterative way. Because I guess that we've heard this phrase yet, um. So, start small, scale fast or fail fast. And I think that's really important because resources are always going to be limited. So if it's not gonna work, you don't want to spend 6 months trying to prove something and find it doesn't work then. You want to prove it within 2 or 2 or 3 months that it's gonna work or not. Have a plan to scale it or a plan to say look, let's be honest with ourselves, this isn't working. 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 detection podcast. Um, I'm really pleased to be joined once again by Nick Leader, a familiar face to the podcast now. How are you, Nick? All good, trying to, trying to, trying to remain cool on this really hot summery day in the UK. It is, I don't know, I mean, most people will be listening to this actually, but yeah, if you happen to see us, we, we might slowly melt through this podcast as we, as we get through, but we'll, we'll try, try and survive the best we can. Um, but today, yeah, today we're talking about, um, a, a not unfamiliar topic, AI, but hopefully we're gonna have a new sort of angle on it, cos we're talking about AI beyond the hype, which I think is a very challenging topic actually, Nick, cos it's impossible not to hype AI, I think at the moment. Um, but, um, but yeah, so I guess let's dive straight in really around this topic. I mean, why did you, you, I mean, you came to me with this topic as a suggestion. What sort of got that sort of thinking in your head, Nick? Yeah, well, to be, to be honest, when you say hype, I mean there's hype and then there's hyper hype. And I think where we are right now is that everybody, vendors, my, my clients in the manufacturing industry, everybody needs to, or they feel at least they need to be seen to be doing something with um AI. And you know, even some of the execs I've been speaking with, I mean I was speaking with the CTO of a UK PLC company and he's going, he's sitting in board meetings and they're talking about what are the AI projects that they need to do. Not, what are the problems we need to solve and how a, how AI can help us solve those problems or change the customer experience or provide new functionality that hasn't existed or insights. So uh the, the thing that's frustrating I guess for me is that AI however powerful it is, I don't think we've yet really understood the true value and potential, and we're probably. Spinning it in the wrong way. And that's. I mean, you, you've sort of hit the nail on the head there in terms of, you know, understanding the value and the value. Do you think, and that, that must be an issue where people hear about AI and needing to do AI projects. Is it sort of going too quickly ahead with things without actually thinking about, you know, how, how can we apply this to, you know, business challenges? Yeah, there's a phrase actually I'm, I'm hearing more and more these days, um, or at least I'm saying it more and more, which is. The capability of these tools is outstripping our ability to take the value from it. So, um. What I, what I mean by that is, uh, you know, I, I think that. Just because a tool can do something doesn't mean we necessarily have to use it. Um, and there are so many clever things in simple things we can do in our operations in our facilities that take us a step forward, that AI as a second or maybe third step could be incredibly powerful or incredibly valuable, but once we've established perhaps core data that is trusted, that is contextualized in the right way that you can train off effectively. I, I think those things that just having that foundation actually takes you a massive step forward. So we're jumping, we're jumping the technology, so technology essentially is outstripping our ability to consume it. I'm sorry there's a little bit of background noise perhaps going on, but you know it it's outstripping our ability to consume it and therefore we are now, we're almost kind of being driven by the technology pace rather than our pace to. Both, uh, capitalize on it, but also, you know, let's face it, we've, we've been living in this world of digital transformation for, um, I, I, I'm, if you can't, you can't see this, but I'm doing air quotes, yeah, digital transformation, um. We've been living in it for, well, at least I've been doing it since the year 2000, you know, 25, nearly 25 years into this realm. We've all got a little bit of change fatigue and then, you know, getting pushed now to do the next thing, the next thing, because the technology can do it. Actually, we've got to get back to, right, how is this gonna improve the quality of my product, the reliability of my machines? How is this increasing the throughput without increasing my capacity? How can I avoid the net Capital expenditure? You know, those things, how can you know, how can I, it's very topical from a UK point of view, how can I optimize machines' performance to use less energy to produce the same output. Let's focus on those core problems and then see how the technology supports us rather than the other way around. And you mentioned some of those, um, some of those use cases, so around energy optimization, predictive maintenance, and, and things like that. I mean, do these use cases often spring to mind for manufacturers or is it something usually that needs to be, you know, it's not there until it's sort of in front of their face kind of thing. I'm just wondering whether it's like an obvious thing. It, it, it kind of, it kind of is and it kind of isn't. I was about to say. If you're a manufacturer that goes to trade shows, I mean, I, I guess most companies if they have in the UK, if they've been in the UK, the big show is at the beginning of June at the NEC with the digital manufacturing, um, week, um. And you could not move for AI stands, yeah. Oh, we do AI this, we kind of have AI enabled tissue dispensers for your toilets, you know, all this sort of stuff. You can track the paper usage from it, um, it, it's gone mad. It has gone mad and it's, it's like there are some very, very important things I think that. Are best suited, I guess, to AI applications. So you know we talk about predictive maintenance, you know, predictive maintenance, as long as you can get constant signals from your sensors, so vibrational sensors, uh, thermal sensors, oil quality, whatever you need. Link that back to what you were producing at that time because there are variables there and that's the important thing is understanding and building the correct variables into these models. Then what we've proven and then what companies have proven over time is that you can build gradually more models that have higher degrees of confidence to predict potential failures. Now the word predict is really, really key here. Because it's not prescriptive, it's a prediction. So that you still need people to first of all, go in, look at the data, interpret that data to understand yes. OK, I'm looking at these various signals. Perhaps my peak acceleration's gone up. I've done some analysis and looked at the spectral, uh, you know, at the FFTs and going, OK, yeah, I can see that there's a probably a wear on the inner raceway or shaft misalignment. And, and then. Use that data to go and verify and then take corrective action. Which is better than waiting for something to break than basically companies do today. But we're not yet in that full prescriptive view, and I think there's actually a a confusion here, um. Where people think also for AI I think it's a vendor's role, so apologies to everybody from Siemens listening here, but I think there's also an important role for the vendor. So let's get back to the predictive maintenance piece. You know, I, I've mentioned FFTs, I've mentioned, you know, looking and analyzing the data here. That requires a certain level of competence and skills and training. Whereas if we use AI correctly, we can use that knowledge, we can embed that knowledge into the tool and actually then produce better insights for people to take action without needing that detailed information. Isn't it better to say, hey guys, this machine's gonna potentially fail cos the shafts are misaligned or you need to go and check the lubrication on the machine. And just have it as a statement like that rather than, hey, there's a warning on you know your RMS or something else like that, you've gotta go off and do something. So we need to take uh the world of complexity and I've had, by the way, I've had a recent report or recent blog written and uh published by the World Economic Forum on this, which is, do we have a skills crisis or do we have a usability crisis? And I think this is where AI could put massive benefit. Is to simplify the complexity of all this data to simple actions and information that people can understand irrespective of whether you're a manager, whether you're an a line technician or whether you're a detail VA2, um, in the maintenance team. I guess there's another, uh, I, I completely get, um, the, the points you're making, but I guess in order to take action from AI insights, there has to be a high level degree of trust in AI as well, because there's one thing saying AI is everywhere and we must use AI, but that human instinct to actually trust what the AI is saying is crucial as well, right, to the adoption of these tools. Yeah, and it, it occurs over time. So, um. You know, I, I think it's good by the way, to be skeptical, to ask for the tools to prove themselves to be right. Um, but actually, particularly with that usability question answered. I, I've seen real success stories happen where people, you know, operators get and use um use the tool, the barrier to entry is lower. They actually go and find out yes the tool was right or sometimes the tool was almost right. I mean I've got examples of where we think there's a a bearing getting loose on a fan. And when we actually look at the fan, what's happened is that the fan's imbalance because of some um debris on the fan blades, but it gave the same signal as a kind of vibrational signal as you would do with a a a warm bearing. So You know, I think there's trust and I think that you need to understand the limitations of these things as well. And as you know, you can only do that through experience. That's not saying I'm not gonna use it because I don't trust it. It means that I have to take everything with a degree of salt, a pinch of salt until it's proven and repeatedly proven to give me the right information. And I think that's my gripe a little bit from these trade shows cos you go around, you speak to the vendors touting their latest AI solution. And they don't necessarily have the, the process knowledge, the detailed process knowledge. To actually explain really how it works, um, as opposed to, well that we have models to do this, you know, there's a, there's a big discrepancy between again between technology capability and practical application. And I, but I think it comes back to another word which I know you use quite regularly, Nick, is about buzzwords as well. So if you talk about algorithms and things like that, I guess people think. Or vendors, some vendors are guilty of thinking that that impresses people, right? If I say, oh, we've got these algorithms, people are like, oh, brilliant, that's what ding, that's what we need, that's what we need. Is it that, it's that kind of thing as well, right, that's playing into it? Yeah, I mean, obviously if you've got algorithms, there's creams for that, we can, uh, solve, solve problems, but in reality, I, I mean, something like AI, if we kind of boil it down, it, it isn't a thing, it isn't a product, yeah. It it's a capability that you're gonna use and develop and it, it's about turning data. Into a way that you can make better decisions. So again, let's let's strip these buzzwords back to actually what it's actually doing for you. You know, it's algorithm, I mean who cares about an algorithm? It's about helping you use the data you've got to make better decisions, yeah. It's about augmenting the team um and providing, you know, what AI is brilliant at doing is spotting patterns of things and often patterns with disparate pieces of information. Is it a combination of the machine running speed, the, the humidity or the temperature in the in in the factory at the time and the mach what you're actually producing. And those are quite disparate bits of information, but it can connect those things together and predict outcomes because it starts seeing patterns of failure or issues that occur in specific situations, um. And you know, I think when you think about an approach which is can you sense it, can you decide what you need to do and then can you take action from it. You know, tools like AI kind of bring the horsepower to do this faster, where traditionally you'd have you imagine the amount of Excel spreadsheets you would have had to gone through that previous scenario to try and do that. You know, you can speed it up, you can do it more accurately, and you can bring many more variables in that a human traditionally couldn't analyze effectively. So I think when we talk about words like algorithm, let's just distill it down into those simple things that are really relatable to the operating environment of um your your customers and the people listening to this podcast. Yeah, no, absolutely, absolutely. Um. And OK, another thing sort of stands out to me as well because you, you mentioned, you know, the, um, Smart factory event where, you know, there's AI everywhere, and I had a similar experience at the same event, not, not jumping on this event in particular, but, um, around predictive maintenance as well, where predictive maintenance was on everything, like everywhere as well. So it sort of leads me to think when people are trying to select tools to use, you know, where do they even begin to start when everything's AI, everything's got algorithms, everything, you know, what, what's the key criteria that people should focus on, um, yeah, yeah, where do they start? It was like, it's like that kid in the sweet shop moment, isn't it? You walk in and go, wow, which one, which one do I want? So, like a kid would do in a sweet shop, the first thing is decide. What sort of suite you're really after, and then lets me translate that into into more meaningful terms, yeah? What's the problem you're trying to solve? Is it about energy use? Is it about quality? Is it about maintenance and unplanned downtime or availability, yeah? What's the biggest problem you're trying to solve? Understand. What wouldn't, what good would look like? And then if you've got those two things, then I think you can start narrowing down because you're, you've got a real focus on an outcome, not a piece of tech. Um, and therefore not all immediately, not all the tech is gonna help, you know, do you want that sweet to be sweet, uh, do you want it to be a gel kind of based thing, do you want it to be a sour? You know, those are your problem kind of comparables, and therefore you're narrowing down the sweets in the sweet shop. Um, And then I think there's other questions you need to think about um like how interoperable are these tools, can it talk to your systems you have, you might have an MES system, a SCA um an ERP system, can it connect to those things really easily? Because you're gonna need that data to feed these sort of AI based tools with more contextual data. Um, do you need a PhD in data science to use it? Or is this kind of a low code, no code type deployment, yeah? That's another kind of criteria, you know, I, I remember going to an event and then going we need more. The problem with manufacturing is we can't attract enough data scientists because the financial services sector is taking them all. And I'm going, hold on, this is wrong. We need AI to be used and data sciences to be used by everybody. You don't need that, we shouldn't need those sort of skills. And the, the final question is, what's the ROI? Is the problem big enough and worth it to solve, you know, are, if you deploy this, will the ROI be 6 months, 12 months or 18 months? You know, if you get the, if, if you get a 6 month ROI on this, you've got a big problem that can be solved with a very rapid return. That, that helps your case to get the funding to actually launch these initiatives as well. And I wonder with, with AI as well, whether it, it also comes with expectations of success because everyone knows how amazingly productive AI makes everyone. But does that also create Yeah, an issue sometimes where you expect instant results or otherwise, I'll no, that's, that, that, that obviously doesn't work. There's a, there's not, there's a bit of patience required, right, for depends on the technology we're talking about, of course, but there, there, there is, but I, I, I guess, you know, I don't think there's ever been instant payback on anything. Um, I've seen some phenomenal paybacks or at least phenomenal results, uh, with, with some of these tools, you know, and going just simply measuring stuff, you start to see stuff you didn't know existed. So for example, I worked with a customer that the HVAC system was working at, um, 33 a.m. in the morning when the factory was shut. It's costing them money, you know, that's not AI by the way, that's just simple. Well, what's, what's that system turning on at that time in the morning? But um. You know, I, I think the problem here, the other problem is when we start initiating these projects, we, we make the projects also too big. So are we trying to eat the elephant in one go, you know, or do you just take bites to prove. And starting with very, you know, starting with something smaller. With very clear metrics that you can demonstrate whether that's like reduced and unplanned down um, unplanned downtime, uh, yield or energy, yeah, um, run experiments as well so that in, you know, every 90 days you're kind of moving it forward, you're capturing the benefit. You're able to share the story of what you've done and the results that you've seen, that builds confidence and buy-in within the organization. Um, and by solving one problem at a time and gradually building, this is something also really important for AI, yeah, is gradually building up the training data and showing that it's actually working and correlating correctly. You know, building that momentum, one problem at a time, you know, perhaps one per quarter. That I think is the big difference here. So it's not the, the big transformation that is gonna take, you know, the classic world of ERP, yeah, 18 to 24 months at least, it never works first time round. Start smaller, prove it, and gradually build it. But I think it's important, you know, we, we talk about. A little bit about, I, I mentioned eating the elephant, you need to at least know what your roadmap is and where you're gonna go with it. Cos if you only have a small project to prove, that project could well prove but may well never scale because you don't know where they're. You know, how can you apply this to more facilities or different lines within your factory or different machines, is it scalable that way or where do you build from the next capability on it, so it's important to have a view of what the elephant looks like and the steps to get you there. But you don't start trying to deliver it all in one go, you take the first steps, you prove it and you grow it. …

Full transcript (WebVTT)