EMS@C-LEVEL
As Forbes, Entrepreneur, Fast Company and SCOOP writer, Philip Stoten, continues to talk to EMS (Electronic Manufacturing Services) executives he learns more about their individual and collective experiences and their expectations for their own businesses and for the entire electronic manufacturing industry.
EMS@C-LEVEL
AI, Digital Transformation, and Innovations with Arch Systems Cofounder and CEO, Andrew Scheuermann
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Uncover the path to digital transformation in electronics manufacturing with Andrew Scheuermann, Cofounder and CEO from Arch Systems. Andrew shares invaluable insights into how Arch, with 85 professionals in 13 countries, is partnering with major EMS players like Flex, Jabil, and Plexus to redefine efficiency.
Learn how their innovative Action Manager tool is revolutionizing the transition from data to intelligent actions that significantly boost KPIs. Discover the strategic approach behind scaling from proof-of-concept projects across multiple facilities, and why large EMS companies are cautiously embracing digital change to stay ahead in the competitive landscape.
Be inspired by the transformative power of AI in manufacturing. Imagine harnessing generative models like GPT to automate complex tasks such as downtime analysis, retaining expert knowledge, and driving the fourth industrial revolution.
Andrew walks us through Arch's ambitious plans to expand capabilities from SMT lines to box would and to new industries like semiconductor advanced packaging and injection molding. By collaborating with MES providers such as Aegis Software, Arch aims to unlock untapped efficiencies in less optimized sectors.
Join Andrew amd I as we explore a future that promise to reshape manufacturing and the exciting innovations that lie ahead.
For more information about Arch Systems visit https://archsys.io/
Listen to Andrew's own podcast "Manufacturing Intelligence" at https://archsys.io/blog/introducing-the-manufacturing-intelligence-podcast/
EMS@C-Level is hosted by global inspection leaders Koh Young (https://www.kohyoung.com) and Global Electronics Association (https://www.electronics.org)
You can see video versions of all of the EMS@C-Level pods on our YouTube playlist.
Digital Transformation in Electronics Manufacturing
Speaker 1Hello, I'm Philip Stoughton, from my House to Yours. Welcome to EMS at Sea Level. I am joined by Andrew Scheumann of Arch Systems. Andrew, we chatted recently in Silicon Valley. Pleasure to be having you on this podcast and, as I mentioned to you then, I'm a fan of your podcast. Congratulations on getting that off the ground and getting it started.
Speaker 2Oh, thank you, Phil. No thanks for having me here. And yeah, we just got the Manufacturing Intelligence podcast off. We got the first couple episodes out highlighting different thought leaders, especially in digital innovation. First couple one in semiconductor, but then the next few exactly in kind of EMS electronics assembly.
Speaker 1Yeah. Yeah, I enjoyed the first three and it very much felt that we were covering different areas but with kind of an overarching theme, and that was nice to see. And looking forward to episode four dropping and you're on my follow list so we'll make mention of it in the show notes so people can follow. Thanks so much. So let's start with a brief introduction to Arch, because I don't think people perhaps realize how substantially you've grown and how many of the Arch team are actually out there now.
Speaker 2Yeah, I appreciate that. So Arch Systems is headquartered in Silicon Valley, but, yeah, we have about 85 professionals across 13 different countries now. So we built a large distributed team that works with some of the top manufacturers worldwide. So we're helping, I think, about 125 different factories today and that's been growing really rapidly 80 to 100 percent often increase year over year and how many different factories we're installed in and helping, and especially that we've done some work in six of the 10 largest EMS. If folks search us online, they'll see press releases with Flex and Jabil and Plexus, notably those three that are publicly announced our partnerships together.
Speaker 1Yeah, and those partnerships are really important, aren't they To have that kind of buy-in from companies that have multiple facilities, multiple locations? They've got legacy equipment they should have of. They should have all the problems that you're trying to solve and a lot of those requirements. And when you look at a company like Flex that has grown so much with acquisition it's a very complex equipment set Is connectivity. The first thing you think about when you look at a facility Is it getting all that data before you can homogenize it?
Speaker 2Yeah, I would say one of the phrases we have at Arch is data to insights, to actions, end to end and so kind of. In a practical sense, yeah, we do start with data and end with actions, although we really like to flip that on its head and think about the intelligent actions we're going to drive first. Data is a means to an end. We don't sell connectivity at Arch just for the sake of connectivity. We sell efficiency improvements and guided actions, and data is the first step to enable that.
Speaker 1Yeah, and I think when I look at guide actions.
Speaker 1It's interesting you use that term data insight action. For ages I've been using the term data insight value and to a degree that you know, that's where the actions come from. But my sense when I look at digital transformation in our industry is what has slowed it down has been one having the building blocks and we'll come to that later because part of that is connectivity but part of it is elsewhere. But it's actually being able to see the value, being able to say, hey, if I spend this much, I can save this much, or if I do this small proof of concept project, I can see a return on investment and then, if I can do that, I can scale it to the rest of the facility. My sense is that up until now that's just felt overwhelming and people haven't known where to start.
Speaker 2I completely agree with you. Yeah, a lot of people have done data integration projects. They've integrated all kinds of data. Maybe they built a data lake, maybe they built a dashboard and they've gained new visibility, or they've been able to count how many megabytes or gigabytes of data they've been building up somewhere, but they've said you know what's exactly the value. So, yeah, I like data insights value very much as well.
Speaker 2And for us action not to jump too far ahead, but core of our tool is called the action manager, which is moving away from dashboards that you know people are supposed to look at. Do you even know if they're looking at them? Are they creating value or not? You know, let's say, the OEE got better, the efficiency got better. Was that because of these cool dashboards you had, or was nobody even looking at them? And they solved it with all the great knowledge in their heads, right, and so we have a core of our platform called the Action Manager, these guided intelligent actions, and you can directly see if guided actions were followed, were they completed, and so we do measure the KPIs and the ROI for all the sites that we're in and the best ones that are achieving, you know, 30 X ROI plus on their license costs, are exactly also doing 10, 15, 20, 25 actions per line per week, right? You see exactly that they're clicking off actions and so you know there's a one-to-one with having advanced technology and really using advanced technology to improve the KPIs.
Speaker 1Yeah, and actually tracking that KPI performance back to the route which is actually using your product. When you look at those companies, and particularly those large EMS companies, are they typically conservative in their starting approach? Do they go for a small proof of concept project and then look to scale? And is your job to kind of create something that's small enough and tangible enough to get them started and then help them to spread that across the facilities and across the processes?
Speaker 2Definitely. Yeah, nobody has started with us with. You know 10 or 20 factories all at once. But on the flip side I will say that the successful proof of concepts are often in two or three factories, not just one. So it's not you know, you go to one factory, one little corner of the factory, and go here's five machines or here's one line that we don't really run very much but like maybe you can do something impressive with that, like that doesn't make for a good pilot. What makes a really good pilot is a couple lines in, for example, two different factories. So you see the visibility, you see the all the infrastructure and security points proved, but you're also on top of a higher number of lines that have real impact to the organization. So just a quick example on this. One thing I say is let's say that in a given week there's a one out of five or a one out of 10 chance that our software would catch a really important action.
Speaker 2Not just like okay, it's nice if you can save a little bit on attrition, but it's only gonna be worth a couple hundred dollars. Like there's a major downtime. It's a major quality issue, a major, you know, loss of materials like, and then if you see this and guide it and have a faster mean time to repair, this is gonna make a huge delta one out of ten times a week.
Speaker 2Okay, if there's one line in the pilot every five weeks, every 10 weeks, you'll see one thing and you'll still ask you know, was it? Was that a trick or was it repeatable? If you have 10 lines in the pilot across two plants, every single week you're seeing something. So you run a four to eight week pilot and you've seen four to eight significant things happen and then you say, wow, so you know, do they start somewhere?
Speaker 2and then scale it out. Yes, yes, absolutely, and we always work with them to include at least a significant base of lines. And if you're not a tier one, if you're a tier one, you know, say, do five to 10 lines in a pilot? That's still a small percentage of your total. Tier two, you're a tier three. Five to 10 lines might be a good chunk of your base still. And so this is something I always emphasize, because if you're a tier two or tier three, we'd love to work with those companies as well, and it's, it's. The pilot can feel a little bigger right from the start, but that is important, not to limit it to too small a scope, and then think, yeah, see, I'm not seeing the improvements in just two weeks yet, you know. Yeah.
Speaker 1And is that? It's interesting? You mentioned that. Is it typical that you have had more success with the larger EMS? I mean, we obviously talk about those flagship companies. They're great logos and brands to have on the website, but within the US that tier two, tier three is actually doing pretty well at the moment and is, you know, has constraints with respect to talent, is really looking to improve efficiency. So when I think of like $100 million EMS, I see a lot of success. There Are those companies understanding the value of what you're providing.
Speaker 2I think so. Yeah, we have more of those starting to sign up. One of them that shared our name publicly was SMTC, for example, mickey who was there, and there's a new leadership team there shared about that. They were doing some work with us and other ones like that that are in the, you know, 100 to 300 to 500 million range. Yeah, I mean, we absolutely can start a pilot on just one. We've done it. You know, start a pilot on just a single line, shown incredible value and moved on from that and it's helpful now.
Speaker 2We didn't start there because of exactly what I described in that pilot. If you have more volume, you can really see the value of this big data and AI faster. You know there's more likely in a Flex and in a JBL. There's so many lines. There's more likely in a flex and in a J-Bowl there's so many lines, there's so much volume that you know, in just a very small amount of time, this analytics can see major problems. And so from a C-level EMS at C-level your podcast here, right At C-level, you can really quickly see the impact, whereas in a tier two, tier three, you've got to be a little more conservative with your investments and make sure that it really, you know you don't have such a base that you don't belong somewhere.
Speaker 2But yeah, we absolutely been seeing, over the last I think year and a half, two years now, more adoption in mid market, and that's something we're really excited to do more of. We have so many of the connectors already built, so much of you know the stuff, that when we were in kind of we ourselves as a company were in r&d phase, it wasn't the right time to engage. You know, oh, we've never seen your mes, we've never seen your machines just most machines we've seen before. And uh, not to jump ahead, but one thing that we announced about a year ago is a partnership with aegis factory logics, and we have a really exciting capability out together called Aria, and I think Aegis is well known to a lot of mid-market manufacturers as a great MES that they depend on, and so we've been making real efforts to work with more machine types and work with leading MES like Aegis, and I think we have a really powerful solution now, not just for big but also for medium-sized manufacturers.
Advancements in AI for Manufacturing
Speaker 1Yeah, I think FactoryLogix is one of the ones I see regularly when I'm in EMS companies and it's appreciated and well thought of, so that's important. What you say about connectivity and having all that ability to connect and adapt to all the different softwares and all the different pieces of equipment is really important. It felt in the early stages of industry 4.0 adoption, if we have had that, the connectivity was just taking up too much airspace, absolutely, and like many, I felt that connectivity should be like electricity. You should have it. It's a given. Everything connects to the network in the same way it connects to the mains or compressed air or everything else. That's going to give particularly those small companies, but also the large ones, a much faster deployment. With you have we kind of moved on a generation? Is most of the connectivity question in the rearview mirror?
Speaker 2So I'll answer that in two ways. I think we've done more than half of all SMT machine varieties out there, which is a huge percentage. But the other thing is our speed at doing them too. You know, I don't know four or five years ago, when company was still earlier. You know, some of the first ones we're doing are like Fuji machines and some machines and Koyang, by the way. Even like Fuji ASM Koyang, they have different generations of technology.
Speaker 2So we've written molt it's not like one connector for all of Fuji, like are you talking to Nexum? Are you talking to, you know, the host interface of interface of Fuji tracks or Flexa and et cetera. But we were, you know, we started out writing, you know two connectors in the first year or something like that. Right, and now you know we routinely do four or five, six, seven, even eight in a month and we have the team, the capacity to just move extremely quickly at them. So you know that's across our customers, not customers, not, not, I don't think maybe one of our big customers. We did like almost 10 a month for, you know, just one customer but usually that's varieties.
Speaker 2but if you're a manufacturer and, yeah, probably we've seen most of what you have and maybe there's two, three, four new ones, and you know, we need to do those within a three to six month period as we're going to roll out and that's, yeah, it's no longer an issue at all for us.
Speaker 1Yeah, yeah, I wanted to come to the whole issue of kind of ingredient technologies, and obviously connectivity is one of them. Another really important one is AI. A lot of talk about AI. For me, it feels like AI is what could push us along towards that fourth industrial revolution that we've been talking about for 10 years, and when you go to a trade show, it was industry 4.0 on every single booth. Now it's AI on every booth. Do you feel that that's a game-changing technology and do you think that the timing of AI getting to where it's got to now has actually been really good timing for arch systems?
Speaker 2yeah, I, yeah, I've said this to a number of people we, when, uh, tim and I founded the company, um, we were not natural language processing experts nlp, that was like the buzzword. Yeah, theseis because it was a more specific field that you were into. My background is building semiconductor equipment, built x-ray machines and deposition machines, and Tim's background was in large-scale analytics. His PhD was in Markov chain statistics, this kind of thing, but not NLP, not language processing, and so we didn't have that in mind. But around the time that GPT-2 was coming out so the models behind chat, gpt we came out with our action manager concept, which was like the dashboard is dying, nobody has any time to look at these dashboards.
Speaker 2And when you get to a technology like ours at first, you create even more dashboards. And when you get to a technology like ours you create at first, you create even more dashboards. We have like 40 of these advanced analytics dashboards. We call them process insights. For the SMT line, you've only got one or two dashboards to even put something on right. So what the heck do you do with these? Nobody can look at them, and so obviously you need something that looks at it for you and just tells you what to do gives you a guided action.
Speaker 2So we came out with the action manager and it didn't have any of this generative AI or language AI in it, but it was doing exactly this copiling mechanism, where it was supposed to read all this data dashboards, rapidly put it into a ticket and bring guidance to someone. Here's the one, two, three, four. It's what you do, right? And so when then, gpt 3.5 came out and chat GPT, it was like, oh my gosh, this is extremely exciting because this is this huge accelerator where the same value proposition that we offer to our customers can now be turbocharged. Right, we can collect tacit knowledge, expert knowledge. We can manuals. We can grab whatever they happen to have written down so far. We can collect tacit knowledge expert knowledge. We can read manuals. We can grab whatever they happen to have written down so far. We can accelerate doing things like tacit knowledge capture. And, if you want, I could go into detail on the first kind of production generative AI use case we have is one that I don't think people necessarily would have expected, but it's very cool.
Speaker 1Okay, yeah, no, do tell me about that. What I want to do, though, is circle back on the whole idea of tribal knowledge and skill and the fact that we do have a lot of skill resident in gray-haired people. We need to extract that. Maybe AI is the way to do it, but also, alongside that, we need to think about what impact that has on training operators and technicians. If we're just training AI, give me your example of the use of AI first.
Speaker 2So the first production use of AI that we're doing is automating the understanding of downtime. So a lot of systems out there MES systems and OEE systems have the ability for you to click some buttons or manually record why you had a problem. 4ms plus E is a really common man method materials, machine, environment. So I'm down and it's my machine and it's it's the other nozzles broke. I need a new note. I'm down because I'm out of materials which one? The specific resistor capacitor right, et cetera, and so you can write these down.
Speaker 2And if you took all the time to write it down people don't really like doing, you know then you would have this thing to dig through and figure out what are my top downtime reasons, how do I know tomorrow? And it's, it's slow, it's error prone and it competes with the time to just solve the downtime right now. Yeah, so this has been a huge problem that our customers have been worried about for a long time, but it's one where there is no perfect rules based analysis, youbased analysis of just like hey, every time this machine throws this error code, you know that it's a nozzle.
Speaker 2Even that you would think, oh, it's perfect. There's so many great error codes in Fuji and XM, but it's complicated and different error codes fire, there's no error code, et cetera. But this is amazing for generative AI, because what we do is we take all the rich data we have from machines and also we now have a mobile app where people can talk in their native language broken English, spanish, chinese, whatever it is take a picture of just what they're seeing, and they might not be the expert that knows exactly the root cause of what's happening, but they just say whatever they're seeing. And we have all the information in the machine and we give it to the AI, which is getting smarter and smarter, and we say, hey, this organization has 50 downtime codes. Which one is it you pick, and it's extremely good at picking.
Speaker 2And so then, not only have we now gotten automated understanding of which downtime happened, without the person having to click through the menu, but also we've started capturing this really interesting information from whoever is on the floor of like what's everything that you saw, right? If they see and understand very little, maybe it's just enough to label the downtime code right. Understand a great deal. We now have richer information, not just from the machines, but also from the people. So this is the first thing we're doing. It's both automating downtime labeling, which is an immediate efficiency time saving, and then it's starting to capture the tacit knowledge on the problem and, like you know what are the deeper problems happening in your factory from your frontline.
Speaker 1Yeah, and that's really interesting, isn't it? When you think about those other issues that people are seeing in their periphery, if you can record those, that's when you get to that level factory, the first thing you want to do is walk the floor. And you want to walk the floor with the guys that are operating the machines or the guys that are responsible for that particular line, and those guys are all unique. They all have some really interesting knowledge that has come from years and years of operating a particular machine set. They know how the machines interact. How do you first get that data in? And I can see that you're doing that by saying, hey, just speak to me and show me what you're looking at. It's almost like you give them the meta Ray-Bans or whatever and you record at the same time as doing that. And is it possible to actually replace some of that tribal knowledge? Because part of the problem with that tribal knowledge is it's doing one shift a day, not three. It's potentially retiring and leaving the industry. How do you make sure you've gained that knowledge?
Speaker 2Yeah, I think there's a future, maybe five or 10 years away, I guess, where there's a lot of headsets in the factory. Maybe it could be sooner that people build amazing headsets. My wife builds headsets at Meta. They're doing a great work. But I think right now the key to capture is about mobile phones and mobile devices. People are familiar with them. More and more factories are comfortable with these coming into the factory, some where they're still not allowed right, but easy interfaces that people are familiar with in your own language, plus knowing when to say what.
Speaker 2So the other thing that I think don't work doesn't work. That some companies have proposed is just like, just like find the expert, just dump everything you like. Just like find the expert, just dump everything you know. Just like pour your head out. Like, what does that even mean? What do they do? I didn't even do that. Yeah, how do you even do that?
Speaker 2But on the flip side, if you have something like this action manager which tells you you know, right now there's a downtime or it could be at the end of your day, you know, at the end of the day there were six major downtimes that we want to know more about, or there's six major scrap events.
Speaker 2So don't tell me about everything you know. Tell me about these six events that happened today Maybe two of them you did already and four are still unlabeled or undocumented and tell me about these four downtimes in some detail easy talk, whatever language you're comfortable with and put it in, so making it really easy, user-friendly for the right thing right now. And then I think what people can also see is experts get to see their knowledge being put to work and immediately creating value, and we're not treating them like this resource. That's like oh, you're the problem. It's like you are the like, please stay. Like we need you in the back, like you are the last bastion holding these things together and please put your knowledge into the system and start collaborating with it so that your life becomes easier. And like you know, um, you know that that's, that's how we see it working.
Speaker 1That's how love our customers to use it yeah, that makes sense, andrew, and you know, obviously the concern if you're an expert is they just want to suck all my knowledge and leave me the other issue that has been discussed with AI. You know we could talk philosophically about AI all day and you know that's a completely different topic, but it's the idea that if we are taking this expert knowledge and we're putting it in the AI as we train the next generation in the AI as we train the next generation although is that next generation missing out on training? Because we're busy training, you know, doing reinforced training on the AI system. How do you think you balance that?
Speaker 2I personally have a strong opinion on that, that the next generation desperately needs training and they need it way faster and more often because the data I'm sure you've looked at this, like turnover rates, just on average have gone up from a couple percent to 10 to 15, in some cases 20% annual turnover rates. Annual turnover rates and I've talked to factories where it's far worse than that and if you look at the implication of that, I was in a factory in Southern California a couple weeks ago where they have one big part of the workforce that has an average tenure of 35 years in their factory and then the next part of their workforce it's like a bimodal has an average of three years of working in the factory average, and so people are leaving after just one or two, and a lot of them are leaving after four or five years.
Speaker 2By the way, in our industry technology, people leave in three to five years. All the time that is the expectation. So, like if I built my business thinking people needed to stay 10 years or we were screwed. We're done for. We're absolutely done.
Speaker 1Yeah.
Speaker 2I think that's happening to everybody. So the these young people tend to be younger, but this new generation workforce tends to be younger. Who's only staying, say, two to four years, desperately needs training and they're never going to be the people that stay 30 years. So if you have a couple of those gem, those foundational employees that are going to actually stay 30 years, don't train them the AI way. Train them the good, old-fashioned way. Make them the good, old-fashioned way. Make them the good, old-fashioned experts. But for this huge part of your workforce that's going to be coming going really fast, enabling them with a technology co-pilot that your experts train so they can get really fast with lex expert time and when they inevitably leave and the next person comes in, that expert doesn't have to like we're on, like, like I have to start again, are you?
Future Milestones in Electronics Manufacturing
Speaker 1kidding, I expected this already. Yeah, yeah, fascinating. I really like this idea of the, of this kind of power triangle of the expert with the, with the tribal knowledge, the, the, the younger generation that want to learn and need to learn quick, and the um and the AI co-pilot that is learning from the pilot but also training the next pilot and training the next co-pilot that you know it's a nice synergistic triangle that works in every direction and I think that's yeah, that's hugely valuable. What about the future at Arch? What do you see as the next big milestone? What are you looking to achieve in 2025?
Speaker 2Yeah, so maybe I'll answer in terms of EMS and outside of EMS. So EMS, which is top focus for us. Major milestone is I'll start with the product angle really significant focus from front to back of lines, so being able to do the same thing that we do for all manufacturers.
Speaker 2You know we connect all kinds of sources of equipment data and logs and MES information and drive these intelligent actions. You know, help quality, downtime scrap, but not just do that. Some of our customers already work with us front to back, but the majority work with us on the SMT side today, and so a really exciting angle we're going and partnering with MES like we're doing with Aegis and Aria.
Speaker 2For sure. A lot of companies have home-built MES. We work with those very closely as well. You know that capability of going front to back, kind of the whole context of the production line and doing these guided intelligent actions, that's a really big one. And so our product enabling that for our customers, driving end-to-end ROI with that, a higher level of KPIs, you know. So, yes, continuing to focus on, say, smt machine utilization and scrap rates, but also focusing on like end-to-end rolling throughput yield and even on time delivery of the product and how AI is guiding an action to help you with those things, not just with utilization, that's really exciting and we hope with that's going to come far more customers and collaborators and partners of ours.
Speaker 2Another one is so kind of like adjacent to EMS, but a little bit outside EMS is we're expanding into um other processes and industries that are right next to it, and two focuses that I'll mention are semi advanced packaging and injection molding, which are wholly outside of EMS.
Speaker 2That's a. There's a big part of what does as well. Wholly outside of ems. That's a big part of what he does as well. But there's also a lot of non-ems collaborators, partners and customers where these processes are really important and they absolutely need the same guided, intelligent actions and being able to help our customers more holistically. That's the unifying answer across these two things front to back a plant.
Speaker 1So yeah, and I think it's really valuable. And when we look at solving that connectivity issue, the first thing we did was solve it on the SMT line and the SMT line feels like there are still some important efficiencies to be gained. Change over time, as you say downtime, all those issues but when you get off the SMT line there's a whole world of exciting opportunities in terms of increasing the performance of the product and moving into other sectors. Really exciting, I think. Aegis are in a couple of thousand factories around the world and a large percentage of those are not EMS factories. So actually having a partner that's already kind of trodden that path is pretty interesting in terms of seeing where the opportunities lie there.
Speaker 1And there are obviously some industries that are actually way less efficient than the EMS industry. You know the EMS industry has survived on this business model of you know, making a relatively small margin by making products that OEMs used to make themselves more economically. So it has to, by nature, it has to be efficient throughout the supply and value chain. So exciting times ahead. Keep doing what you're doing, keep doing the podcast. I'll continue to download and enjoy and comment on those and thanks so much for your time today, thank you. Thank you for having me, phil, it's been great.