Podcast EP 143 - How can OEMs transition to a service-based business model - Gijs Meuleman, CEO, Sensorfy

EP 143 - How can OEMs transition to a service-based business model - Gijs Meuleman, CEO, Sensorfy

Aug 26, 2022

In this episode, we interviewed Gijs Meuleman, CEO of Sensorfy, specialist in predictive solutions for industrial companies. Sensorfy designs and develops predictive maintenance solutions to processes sensor data near the source, with machine learning on the edge. 

We discuss how to transform a transactional business model into service-based, data-enabled business that is more closely aligned with benefits realized by the end customer. We also discuss how predictive maintenance impacts operations cost structures to allow leaner maintenance organizations while sustaining service levels.  

Key Questions: 

  • Why is predictive maintenance one of the most widely adopted IoT use cases? 
  • How do you evaluate the different parameters to decide if predictive maintenance makes sense in a specific situation?  
  • How do predictive analytics enable business model innovation? 
  • What is the current level of maturing in industrial equipment markets in terms of adopting service based business models? 
Subscribe

Erik: Gijs, thank you for joining me today. 

 

Gijs: Welcome. Yeah, great to be on the podcast. 

 

Erik: I think this is going to be an interesting one because the topic here is predictive maintenance, which has to be use case number one for industrial IoT. So, it's something that most of our listeners will be familiar with. But I think we're going to find some interesting ways to address this from a business perspective and look at it from different angles. So, I'm looking forward to this because it's a topic that — at least, for us at IoT one — comes up all the time. But before we get into that, I'd love to understand how you ended up setting up Sensorfy. I think it's always interesting when talking to a founder to understand a bit about the backstory. Where did the concept come from? 

 

Gijs: Great question. Thank you, Erik. Well, my name is Gijs Meuleman. I do have a background in Electronic Engineering from Eindhoven University, specifically into RF Microelectronics. After finishing my studies in 2012, IoT was still relatively new. Actually, I thought I can do a better job here. I saw a lot of potential. I just started as an entrepreneur. To be honest, I didn't think about it that much. I started as an engineering consultant developing IoT solutions in general. At the time, I was also not so busy with thinking about scalability. Over time, we developed various IoT solutions in, actually, a wide range of industries. From those solutions, we've built our technology stack — a set of modular hardware and software components — that helps us now to make predictive maintenance solutions for our clients. 

 

Erik: So, you were a researcher for a couple of years before you set this up in 2015 at Eindhoven University of Technology. Were the things you were working with there directly transferable to predictive maintenance, or was it more fundamentally similar technologies but different solutions? 

 

Gijs: Well, it was similar in a sense. I actually did a PhD in RF Microelectronics. I know quite a lot about RF transceivers, which is a part or an important part of IoT solutions. I didn't like being a researcher, so I actually quit after two and a half years. Then I decided to start my own company. 

 

Erik: Yeah, got it. Aside from the fact that you thought you could do predictive maintenance better than what was on the market in 2015, why start your own company? Why not join — I don't know — Siemens or some large player here? What gave you the confidence to say, "I, as a relatively fresh grad, can come out here and win in this market"? 

 

Gijs: Well, a really good question, Erik. Well, as you can see on my profile, I was an entrepreneur already since 2009. I did some of the sides job during my studies, so it was more like a natural choice for me. 

 

Erik: Okay. Got it. I see, also, you've set up an investment company focused on real estate, which is very interesting and ambitious. Actually, it looks like a fun company, what you're doing there. But I can see that you really do have the entrepreneurial life. 

 

Gijs: Yeah, true. It's a hobby that got a bit out of hand, but it's a different topic to discuss. 

 

Erik: Yeah, that's a good way to describe it. Great. Well, let's talk about this topic of predictive maintenance. Again, it's a solution that most of our listeners will be fairly familiar with. Why is it the specific solution that you've decided to focus on? Actually, I think that's quite interesting because a lot of tech companies will say, "We focus on a particular vertical. We focus on a particular layer of the tech stack," but you've chosen a specific use case to focus your business on. What about predictive maintenance that struck you as a suitable scope for business? 

 

Gijs: Really good question. As I mentioned before, we started off as a generic IoT solution provider, and we did a lot of various IoT solutions. Three years ago, we decided to bring more focus and concentrate our efforts on a certain niche. We look hard at the products we developed over the years, and quite a lot of them were about predictive maintenance. I think the business case in this field is pretty clear. That helped us to make the decision to focus on this niche, mostly because we have acquired some experience in that area. 

 

Erik: I think it's a smart decision. I think entrepreneurs often have hungry eyes, right? So, they wanted to find a problem as big as possible and envision their unicorn in the future. But it's much easy to be successful if you're focused. So, I'm interested in looking at this from a couple of perspectives, just based on my personal experience with predictive maintenance. Maybe the first perspective is defining the business case, just from looking at it from a cost and efficiency perspective. Then we can look at it also from the business opportunity or the opportunity to refine your business model around this. But just from a cost perspective, there's a couple of cases that I've seen recently or been involved with. 

The first was a bit of a negative. So, that was looking at automotive production lines. You look at it and you say, okay, predictive maintenance makes a lot of sense here. Then you get deeper into it, and you say, well, the reality is that, at least, here in China, there's excess capacity so they're running at 85% anyways. They have a lot of scheduled downtime. Then you've got all these engineers that you're not going to fire because they're state-owned enterprises, or because they need them anyways for other purposes so you're not going to save headcount. Then the production lines all have slightly different equipment, so you're going to end up doing a bunch of different mini projects if you want to deploy this. The business case just gets killed. 

 

Then you look at another case, and this is offshore win. You think, okay, flying somebody out there on a helicopter costs $100,000 round trip, and it's a killer business case. So, you have two different situations. One was difficult to justify, although I think that company is actually taking a second look at it. The second is very easy to justify. As really an expert in this, how do you evaluate the different parameters to decide does this specific business case make a lot of sense, potential sense, or no sense? 

 

Gijs: That's a really good question, Erik. First, what you have in the predictive maintenance space, there are different parties. You have the asset owner, that usually has a direct benefit of using predictive maintenance based on the industry, of course. But we mainly serve the industrial OEM market. We do that by integrating our technology into their equipment or assets. That's our focus. Of course, we evaluate every opportunity to see if there's a clear business case. What we see is, most of the driver, especially here in Europe, is the shortage of technicians. IoT and predictive maintenance, in particular, will help to serve more customers with the same amount of technicians. That's something that's very valuable over here. 

 

Erik: Yeah, it makes sense. I think, actually, everywhere in the world right now, that seems to be an issue. Okay. So, labor shortage. Nonetheless, I guess, in the cases that I just gave, there's one where you would assume that there might be a shortage but, in reality, there's not. The other, there's an extreme challenge of getting technicians out there. Are there other variables you look at? I guess, you have to evaluate your customers to an extent, and determine is this a customer that we're going to be able to serve long-term, or is it going to be they're going to figure out that this is actually is not providing value? Anything else on your checklist that you would be assessing to determine who's going to make a great long-term customer as an OEM? 

 

Gijs: One of the other parameters we look at is how far — what happens if the asset fails? Sometimes the impact is very high, costly. That's, of course, an important parameter. The second one we look at is, okay, is this OEM also able to scale it? Usually, a startup is much more difficult. If it has the OEM as an established business over the world, that helps to scale the solution, and also make sure that they can gather enough data sets to really enable predictive maintenance. 

 

Erik: So, there's another case that we're working on, which is a company that does very large pieces of equipment. It could cost $10 million. They are often going to be customized according to the requirements of the customer. So, maybe some percentage will be standardized. But usually, they're going to have different customized parameters. What that means is that it's, I think, often going to be more difficult to train a model, a predictive maintenance model, and then apply that to all of the machines in their fleet. Because they're all going to have slightly — they're going to be collecting different data. They're going to be operating, behaving in different ways. Whereas if I'm just selling a pump, that pump model where I might have 10,000 units, fundamentally, the hardware is all the same so it's going to be much easier to scale that. So, how do you look at these situations where you'll have customization in the OEM equipment? 

 

Gijs: So, we mainly work for — what I said, we mainly work for OEMs. They have very similar equipment. So, one of our customers is a large supplier of automatic doors. You see that although the door might be in its form might be different, the movements you detect and the algorithms you can develop based on that, they are very similar. So, I think if you serve the asset owner, then there will be much more variability in the kind of assets. It's much more difficult to provide a machine learning model for all the different types of equipment with, well, quite some variations in them. 

 

Erik: Okay. That's a good point. Okay. So, you're looking at it when you're working with OEMs from more of the component level, of saying, okay, they're different. But they all use the same hinge. They all use the same engine. You can standardize around that. Is there a shortlist of components that you find to be the most valuable or maybe the easiest to do predictive maintenance on? 

 

Gijs: We've focused on, basically, three markets. That's building assets. That's real components, and that's industrial equipment. Within those markets, we see that we add value quite easily if it's linear motion. So, any vibration data or linear motion, we have quite some experience with. Then it's easy to detect or easier to detect the failure modes and mechanisms and to help a customer quickly. So, that's actually one of the parameters we're looking at when we select new clients — to make sure that we can be successful. 

 

Erik: So, linear motion, what is that like? A piston fire? Give me a couple examples there. 

 

Gijs: Yeah, good question. Within the building assets, automatic doors, with elevators, escalators, you can think of within the real. We've done quite some preparation analysis of real or the real itself, real component turnouts, turnout systems. With industrial equipment, it might be indeed hydraulic equipment or other moving, linear moving parts. 

 

Erik: Okay. Interesting. Why do you find that linear is easier to assess? 

 

Gijs: It's not easier to assess, but I would say we have most of the experience with that. So, it's easier to be successful. Also, there, you have to find a niche. There are other players in the market that can derive predictive maintenance from voltages or currents, for example. 

 

Erik: Oh, interesting. Okay. So, it's more that you've built up this experience over X number of past projects. How transferable is this? What is the workload of supporting customer number 100 versus customer number 1 or maybe customer number 10? 

 

Gijs: Over the years, we've built IoT technology stack. It's a set of modular hardware and software components. That enables us to quickly build tailor-made predictive maintenance solution for our clients. So, that helps them to scale more quickly, although there will be some integration. Always will be some integration. So, we look at the scale of each OEM, in particular. 

 

Erik: Okay. But can you give me a number of hours? I'm just curious. If we look at customer 10, or if you were to try to do a new — if you were to say, "Okay, we're going to extend beyond linear and we're going to start looking at voltage," what would be the effort required for you to figure out how to do voltage versus doing the next linear, the next project for a linear client? I'm just curious, because these economies of scale are so important in IoT. It's such a fragmented industry, that finding economies of scale is actually challenging. When you do find them, they're very valuable if you can really figure out how to scale them. 

 

Gijs: Yeah, indeed. Number of hours is difficult. But what I can say is, it's like the throughput time to really industrialize a project. It can be anywhere from 3 months up to 12 months, I would say. That depends on how difficult it will be to detect the failure mode. Mostly, most of the time required is to really analyze the data and develop the model. 

 

Erik: That seems to be a big issue. I mean, a lot of these industrial systems are designed — of course, they're all designed not to fail. A lot of them fail very seldom. So, you have maybe a lot of data but very few failure data. What is that data collection and training process? Do you basically work with the customer to put their equipment under a bunch of stress tests to force it to fail? How do you make sure that you have the data that you need to work with? 

 

Gijs: That's something we do together with our clients, indeed. Sometimes they can simulate failures to make sure that we capture those data. 

 

Erik: Okay. So, it's mostly a simulation. But I guess simulating in a real-world simulation. 

 

Gijs: In the real world, yeah, indeed. 

 

Erik: If it's 3 to 12 months, what's the bulk of the time here? Is it manufacturing the training data, or is it more of collaborating with them to figure out the business model, or how they're going to integrate this into their service offering? 

 

Gijs: It's, basically, a cycle of product development. So, you might have to make sure you have the requirements right. Derive first a prototype. You develop the software, integrate it into the system, start collecting data. Then from that data, derive a model. That takes quite some time. One of our clients' external systems, they are developing such a product that detects failures within turnout systems. It's quite a challenge. 

 

Erik: What is the turnout system? 

 

Gijs: It's used in a railway switch. This is a similar name. In railway, when you have two tracks in the intersection, that's called a turnout. 

 

Erik: Okay. I've seen the term railway switch come up in newspaper articles about railway crashes, so I can appreciate the value there. Okay. Great. So, that's on the deployment side. Then we discussed some of the benefits from more of a cost savings perspective. I think one of the things that you focus on as a business is helping your OEM customers to also develop new business models around this, look at how they can use predictive maintenance to also modify their revenue model. What have you seen as interesting cases or success cases in also developing new, maybe higher profit revenue streams around this? 

 

Gijs: That's a really good question. If you look at OEMs, traditionally, they have more of a transactional business model. So, they sell their parts. They sell the asset to the asset owner. Then when something breaks down, they provide spare parts. One of the interesting models that can be enabled by predictive maintenance is monitor your asset from a distance remotely, and give timely warnings to the asset owner when the product will fail. So, one of the opportunities is to use predictive maintenance to enable them to sell a long-term service agreement. Then as time matures, they learn more about the failures of their assets. That can help them to transition to a more outcome-based business model. where they are actually guaranteed the performance of the asset and the asset owner. That's in direct benefit with them, because they want to use the asset as long as possible with the lowest amount of total cost of ownership. That's an opportunity for OEMs to transition to a service-based business model, which we see happening more and more now. Predictive maintenance can really enable that. 

 

Erik: I guess you could do this in different ways. So, the simplest way would be just to add a SaaS solution on top. So, I sell you the hardware, and then I sell you for whatever, $100 a month per asset. I sell you this predictive maintenance solution. The more transformative way to do that would be to say, I'm going to lease you the asset. I own the asset, but you use the asset. I then use this software to make sure that I can optimize the uptime and so forth. Where do you see the market today in terms of adopting different models here? 

 

Gijs: Yeah, so, that's correct. You start off usually with a monthly fee. It's where you give them some insights on what the health of the asset is. Then as your confidence as an OEM matures, you can shift your focus more on an outcome-based and indeed, what you said, leasing the equipment and making sure that the equipment is always running, has no downtime. That's a fully outcome-based business model. What I currently see is that there is still a mix. So, companies that are already quite far in this cycle still use a basic fee, a more basic monthly fee. There may be a bonus when the uptime is higher than expected or it minus when the uptime is lower than expected. 

 

Erik: Oh, it's interesting. I haven't heard about this bonus structure, but that makes sense. Is that common? 

 

Gijs: It's not common yet, but I see several companies, industrial companies, moving towards this. If you can figure it out and it really, really works, then you can, I think, be one of the winners in your particular industry. 

 

Erik: It's an interesting concept. I mean, this is one of the most interesting areas for me around the IoT space. It's that there's so much room for invention of the relationship, right? When you have data, you can really reinvent the relationship with your customers. Maybe another thing that sometimes sits as a hidden benefit here is the use of the data that you're processing and the insights for other functions, right? If you think about the value creation as a stack, you have one element of the stack. It's reducing labor expense for your maintenance team. Another element is reducing downtime. You have other things like maybe providing data back to your R&D team, so that your R&D team can understand where failures have been more accurately, and they can design the next generation of the product to be superior. Do you see much effort from these companies to really dissect the data and look at how it can be used by other functions outside of the core job of predictive maintenance? 

 

Gijs: I haven't noticed it that much yet, but I do see the value there. You can actually see how the equipment is used in the field. Then based on that, you can make your next generation of your equipments. You can derive specifications from that. I absolutely agree. But those are long term cycles, which I haven't seen an example yet. 

 

Erik: Yeah, exactly. That's the problem. It's always easier to envision something in this market than it is to actually realize it in the field, right? 

 

Gijs: Yeah. 

 

Erik: Okay. Cool. Well, tell me a bit more about your company. We already know now what you do. You mentioned the industries. Who are you typically working with? Is it more medium-sized businesses? Is it large businesses? Is that, I guess, primarily European? Who are the typical customers for you? 

 

Gijs: That's a good question. What I mentioned is, our customer is our OEMs. Usually, a well-established OEMs with existing market, and they really see the opportunity in servitization of their equipment. Mostly, we serve European customers but they ship their equipment worldwide. 

 

Erik: In terms of — I don't know if you have that much visibility into this. But in terms of the end of markets, where especially the servitization models are adopted, I feel like Europe might be a bit at the cutting edge. I think China, where I'm sitting, it's a bit challenging to get people to adopt. Even cloud, often it just won't work in China right now for industrial. So, there's some issues that need to be resolved more at the human level before some of these systems are adopted. But do you have much visibility in terms of the end markets, and which markets are most ready or willing to start looking at cloud-based software or new business models? 

 

Gijs: I do have some visibility there. Indeed, what you mentioned, while sending data to the cloud is still a challenge — especially in the the oil and gas industry — they don't want to share their data, usually outside the plants. But over the years, I've seen quite an adoption of the cloud. More and more companies realize the benefits and are willing to send the data to a cloud system and see the benefits of that. 

 

Erik: There's a clear trend line, right? It's just a question of industry, by industry, country by country, how this moves. 

 

Gijs: Yeah, true. 

 

Erik: Well, how about in terms of who you're working with? If a company is saying we're going to start looking into offering this new digital service. Whereas in the past, we were selling just hardware, is this treated more as just a pure technical integration where it might be the product manager who's leading the engagement and is your touch point? Or is it more of a strategic topic where the CTO is getting involved, or the business unit head? Who are the typical stakeholders who are making the decisions around adopting these solutions? 

 

Gijs: We see them from two layers, actually. Of course, management has to be involved. There has to be a vision on digitization and how the company wants to make use of that. Then our typical touch point is with R&D. So, they already defined a strategy and see that predictive maintenance is one of the topics they want to address in their next product. That's where we come in. We help them shape the requirements of the strategy and help them integrate the technology, really make it, remove the complexity for them. Afterwards, we help them with the analysis of the data and also designing the model, which can be done only after the collection of various data points, of course. 

 

Erik: Each of these is a discrete project itself. So, what does your offering look like? It sounds like you have some standardized software you've built up, but you're also probably using more an array of off the shelf hardware, and you have your services. What does that look like as a package? 

 

Gijs: Our technology — we built our technology stack that are different components. We have standardized hardware for our sensors. We can customize that. We have an integrated software layer in the cloud. We have an inside hub where we can analyze the data and really integrate the model, predictive maintenance models, and show them the insights. But usually, it's an integration with the edge system as well. 

 

Erik: Got you. When it comes to the cloud, are you built on any particular cloud? Are you agnostic? 

 

Gijs: We've worked with many different cloud vendors, but we have most of the experience with AWS and the technology stack they're using there. 

 

Erik: Okay. Got it. Then when you look at topics like cybersecurity, is this something that you're leaning on AWS, for example, to secure the data? Are there any areas there where you have to get involved in these topics? 

 

Gijs: Yeah, so, most cloud providers have solutions for that, using certificates and making sure that the data you send from your device or your gateway to the cloud is actually secured. We're building on top of those solutions and that could be in AWS but you have them in Azure or Google Cloud Shell. 

 

Erik: Okay. So, that's not such a big issue. Is it always the case that predictive maintenance is a self-contained system, in that you deploy the sensors and so forth? Are there cases where you might need external data from upstream, downstream in a process? Is that ever the case, or can you always basically accomplish the job with the sensors that you're going to be deploying directly on the OEM equipment? 

 

Gijs: Well, it's a combination. Since we integrate the technology, we also integrate with the system of the electronic systems that might be already available inside the equipment. So, we extract data from there. On top of that, we usually use several different sensors based on the specific equipment we're looking at. 

 

Erik: Okay. So, you might be looking at, I don't know, voltage or just energy flows, things like this that could also be relevant. 

 

Gijs: Yeah, indeed. 

 

Erik: Okay. Got it. Great. Is there one or two cases that you could walk through, just to give us a perspective in terms of what it looks end to end? I'd also be really interested in hearing some results that some of your customers have had? 

 

Gijs: That's a good question. One thing I can dive into is the product we built for a large OEM of automatic doors. We've built a sensor that's being put on a door that measures different parameters based on vibration, temperature. Based on the vibration events and some edge analytics, we can detect several failure modes. All of these failure modes are sent to the cloud. Technicians can then directly interact with those, with these data. Then based on that, they can also do an automatic scheduling of a maintenance technician that will go there before the actual breakdown. That helps to serve more customers, of course, with the same amount of people but also to reduce any unplanned downtime. 

 

Erik: In a case like this, what is considered a good success rate? I guess you could look at it in two perspectives. What is the time period before a breakdown that you can predict? Are you predicting six months in advance or two weeks in advance? Then what's the accuracy level of the predictions? What would typically be good ranges here? 

 

Gijs: Yeah, that's a good question. But I don't know, actually. 

 

Erik: Okay. Why is this? I mean, for this specific case? 

 

Gijs: For this specific case, yeah. 

 

Erik: If we just look more generally, not at this case. But in general, what would you consider to be — I guess it depends on the situation. But what would be the reasonable ranges? Because I think the challenge here is, you can always accomplish something. Right? But the question is, is it accurate enough that it's going to make a significant impact? 

 

Gijs: Yeah, that's true. You don't want too many false positives as well. So, it's being accurate. Well, it's possible to have a three to six months window for some type of equipment. But it really depends on which equipment you're looking at. 

 

Erik: Okay. So, somewhere like the — But we're looking at months, months in advance notice. 

 

Gijs: Yeah, that's possible. 

 

Erik: Then accuracy rates, is there any standard range that you would say is good, or is it just all over the place depending on the situation? 

 

Gijs: Yeah, that's something I cannot generalize. 

 

Erik: I guess, in some cases, you're looking at like 99.XXX accuracy. In other cases, somewhat less. 

 

Gijs: Yeah. 

 

Erik: Great. Then, I guess last question for me. I think we've covered this from a few different perspectives. Is there anything that we've missed here around predictive maintenance that would be important for folks to understand? 

 

Gijs: Well, predictive maintenance is, what I would say, it doesn't start with the technology. We should really start with the business model in mind. It's a great technology. There are very potential, a lot of benefits can be reached from them. But you should start at looking at the business model first. Then the technology and all the technology stack will come later. 

 

Erik: Yeah, it makes sense. I think that's probably the opposite from how a lot of companies look at it. They want to do a pilot project, and then they want to look at very direct cost benefits, which is a very simple way to look at it but often not the comprehensive way. Great. Well, what is the best way for folks who are interested in learning more about predictive maintenance to reach out to you, or to get in touch with the team? 

 

Gijs: The best way is to go to our website. It's www.sensorfy.ai/contact. You can reach out. If you are an OEM and you want to to increase your business through servitization and predictive maintenance, or if you're involved in predictive maintenance inside an OEM company, we can help you to eliminate complexity and quickly set you up for success. 

 

Erik: Awesome, Gijs. Thanks for coming on the show today. 

 

Gijs: Yeah, thank you for having me. 

test test