Join the Insider! Subscribe today to receive our weekly insights
Willem (00:00)
Welcome to the ITOT Insider Podcast. So David, who did we bring today?
David (00:08)
Yeah, so really interesting here because we’re gonna, I would say, go back to the roots. We’re gonna talk about what it is to build an OT organization. We brought Anton Melander here. Anton, thank you so much for joining. Anton used to work for Northvolt in several roles and now he is the founder or co-founder, I have to say, of a startup we’re also gonna talk about.
Yeah, Anton, thank you so much. And why don’t you start this episode with a bit of your story? Where did you start? What have you seen, et cetera?
Anton Melander (00:47)
Yeah, thank you very much for having me. So this story probably starts with my short about myself. I’m Anton, as you mentioned, I’m Swedish, born and raised, studied something called industrial engineering management with a master in computer science at this school called KTH here in Stockholm. And after that, I basically had two options, either go down and offer from an investment bank ⁓ or join a small company.
at the time called NorthVolt that was supposed to be making battery cells. And since I’m here, it’s pretty obvious which route I took. when I joined the company, was about 100 employees, had no factories, no nothing, pretty much. And I joined in the digitalization department, which at the time was, I think, me and three others. Starting out, we were like, you know, this is going to be a Greenfield factory, trying to figure out what should we build, what should we buy off the market.
And the only thing we knew was that making batteries is fast. You tend to make a lot of batteries very quickly and it’s very tight tolerances. even small like microns of misalignment can lead to short circuits and so on. And we really wanted to, since this was in 2018, we really wanted to make sure that we could be as data driven as possible when it came to scale up production and learn.
how we can take new chemistries into large scale production. And we also knew that we were not going to get it right the first time. we wanted to make sure. Yeah, exactly. We wanted to make sure that we had a lot of flexibility and could learn as we were going along. So we ended up building quite a lot of software in-house. And I was fortunate enough to have a leadership position in the de-defilization.
David (02:21)
Important.
Anton Melander (02:40)
And we grew that from about 4 to 140 people as the company grew from 100 to about 5,000. And then I left Northvolt in…
Willem (02:45)
How?
David (02:50)
So wait,
wait, wait, four, two, 140. Jesus, for digitalization, right?
Anton Melander (02:53)
So, it’s something like that. yeah,
and the company grew from about 100 to 5,000 during my time.
Willem (03:04)
Yeah, but even then if you look at
the amount of people on digitalization compared to the total company, that’s still just quite a significant part. So, and also there’s this, you guys were building it yourself or were you like combining off the shelf software? What do I need to think about when I think about a battery factory?
Anton Melander (03:12)
Yeah, it is.
So, was 2018, OPC UA started becoming more mainstream. We had to still like make sure all of our PLC providers came with an OPC UA server. There weren’t a lot like IoT platforms capable of kind of ingesting the amount of data that we wanted. So we built the whole IoT platform.
And actually I used provisioning all the edge gateways was a challenge itself. I think we provisioned over a thousand gateways over that period. So just the whole provisioning process was something we kind of had to build up ourselves and also the IoT platform. ⁓ And in the beginning, we were very focused on like making sure we could capture as much data as we needed from the different process steps in production. And then as the company kind of actually started running production.
It was more about how can we ⁓ make operators and organizations use this data as efficiently as possible. So then we created something we called North Cloud, which was the internal MES. Yeah, so we did build a lot in-house and it was, I would say, very competent. Although, of course, we leveraged like AWS for cloud services. had…
We bought our PLM or ERP and these kind of things.
Willem (04:54)
Okay, so could you maybe walk us through what OT means in the context of Northvolt, of battery factory?
Anton Melander (05:03)
So ⁓ I mean, the first layer was kind of the data capturing and edge devices. So for each PLC, we had an edge device that would pull data off and stream it up to the cloud. And then we organized the organization around the of core functions in the organization. So we had one team focusing on enabling the operations team to know what was supposed to be produced, how much did they produce, how much downtime did they have, what were their performance, et cetera.
And then we had another part of the North Cloud product that was focusing on the quality department. So that came in more to like, what is the product that we’re making? And do they meet the quality specifications and so on. And then we were focused around the material flow. So what parts should be where and when? And for that, used, I think we had seven different AGV suppliers, each having their own AGV.
management system. So we had to talk to that and kind of coordinate that as well based on how the production were going and also how the yield and the quality was.
David (06:01)
Wow, yeah.
Willem (06:03)
Okay.
David (06:12)
There are already so many questions which, yeah, but it’s interesting, Anton, because this is a question, okay, it hasn’t, for those who don’t know, an unfortunate ending, and Northvolt isn’t a company anymore. I’m sure that has nothing to do with quality of your tools. But it is still, you start from scratch. It’s a pure green field.
Anton Melander (06:14)
Hahaha.
Yes.
Willem (06:39)
with three people,
three, four people.
David (06:41)
Yeah, yeah.
you start from literally nothing. you have the, on the one hand, you have the possibility to make choices and that has pros and cons ⁓ because there is no, well, there are obviously stories from other companies and experience from people at other companies to fall back on. But for you, for your team, for your company, there is no previous experience so that…
Anton Melander (06:52)
Mm-hmm.
David (07:10)
you have to pick a certain starting point and a certain direction. ⁓
Anton Melander (07:15)
I mean, there, you’re completely right. ⁓ The Chief Digital Officer at the time spent, it was early at Tesla and with quite a lot of like ex Tesla people. So I think they brought a wealth of experience and, know, do’s and don’ts with them. But I think, I think the situation we were like, the main decision we needed to make was like, should we buy, go with an off the shelf, which typically means you need to be very kind of explicit of what you need.
David (07:39)
Yeah.
Yep.
Anton Melander (07:43)
you know,
there’s this, you know, long requirements list and you typically rely on consultants. And that would be, you know, that would be, we would definitely do a lot of change orders and not really be in control of that destiny. ⁓ And also like their software didn’t really meet the needs in data ingest that we foresaw or should we build it in house. So that was, that was the tough decision to be made.
David (08:05)
Yep.
One of the, ⁓ let’s go into organizational stuff a bit later in the podcast, because I also like that. ⁓ But let’s stay a bit technical for now. ⁓ One of the things we often talk about is when you approach data ⁓ as a platform is the scaling issue. ⁓ So you come across so many issues when scaling. ⁓
There is the technology itself. ⁓ Can it ingest ⁓ so much data? There is the egress, course, ⁓ which is not non-trivial. There is the, ⁓ for example, the data contextualization, the data modeling, where do you start? What standards do you pick, et cetera? Now, when you start from scratch, that obviously gives you some, I would say, some extra degrees of freedom compared to starting from a brownfield. So can you also…
Anton Melander (09:01)
Mm. Mm.
David (09:04)
Yeah. What, what, ⁓ where did you start? How did you make sure that the platform was scalable? How did you, for example, deal with, with, with data modeling?
Anton Melander (09:15)
So where I think I spent the first year during my time there was to, because this was when we were sending out RFQs for our equipment makers. And this is typically not like off the shelf machines, at least it wasn’t. So we had quite a lot of influence on how the OPC UA server should look like and kind of the state model of the machines. What is like
We talked a lot about most of the machines were kind of discrete, like they ran in cycles, although we also had a continuous production of the active cathode material. But for most of the time, we defined our own standard interface that they had to comply with. So what should OPC way look like? Like the service structure, like the tag structure? And I think had we redone this today, we’d probably have used the UNS or something like that to help.
David (10:06)
Yep, yep.
Anton Melander (10:12)
this, but we did build like an effect. We had it, call it the factory breakdown system. It took like that, know, factory lines, machines. And then we standardized on like how we differentiate like process parameters and product measurements. So a lot of it started like on the PLC. And then we had our mappers that sat on the gateway that were talking to the PLC and sending packages of data mapped into our standard format for that machine. ⁓
David (10:26)
Yeah.
Anton Melander (10:42)
Yeah, so that’s where we started. then we saw, and I think this is the interesting thing with manufacturing. So take an example as measuring your performance, which is, need to know your ideal cycle time. And the PLC is not going to know the ideal cycle time. And that might even differ depending on the kind of product that you’re making currently. So that you need to sort somewhere else. ⁓ And where do you combine these and actually look at the actual
cycle time versus the target cycle time to calculate your performance. So, yeah.
David (11:16)
But what made you, like from an organizational perspective, how were you organized and how, especially more specifically, how do you see
the organization actually changing when going from 4 to 140.
Anton Melander (11:32)
Yeah, okay. So in the beginning, we were very much of like, you know, project based, we worked with, you know, we were, needed we knew we needed standardized kind of interfaces across machines to not end up in like, to heterogeneous data from the source. So that was like a project to standardize. And then obviously, working with the process engineers as they were ordering and specifying the machines. But and that kind of happened to we were like, 12, 15 people, it worked pretty well to use
Like the whole organization was extremely flat. Everyone was working with everyone wherever needed. But then as the company got to maybe three, 400, digitalization came 15, 20 people. needed some structure. So, so at some point I was the PMO. So then we became more serious about like running projects and prioritizing projects based on kind of, I would say to some degree feature needs. Like this is the features that is needed.
But at some point, when production actually kicked off and we had our first small sized factory outside of Stockholm, we also realized that technology is one thing, but it’s like, unless it actually does what it needs for the people it needs to do it for, it doesn’t really matter. So I think this was a pretty rough period where we went from a very like project oriented way of working into a product oriented way of working.
And with product, mean, like we were actually setting out like here’s a bunch of core capabilities of the platform that needs to work. And they need to help this department achieve these outcomes. like, you you can’t really work with quality unless you know this product had this quality and it failed for this reason and you kind of need to start being able to, know.
David (12:58)
Mm-hmm.
Yeah.
Anton Melander (13:25)
categorize and group nonconformity reasons to you. So you go after the ones that have the biggest impact and severity and so on. So it became, yeah, I think that’s a major thing. We moved from projects into a product organization.
Willem (13:41)
could write a book about that.
Anton Melander (13:44)
Yeah, probably.
David (13:46)
But
yeah, there is even a book called From Project to Product by, who was the author? Mick Kirsten. Yeah, it’s actually a brilliant book, but it’s easy to say this, but I would say the reality is, I would say much more ⁓ difficult than just saying like, we’re going to this product ⁓ mindset. That’s again, something non-trivial. ⁓
Anton Melander (13:51)
Yeah.
Willem (13:52)
Mick, Mick Gerson.
Anton Melander (13:54)
It’s inspired from BMW, right?
David (14:16)
What are your, I would say if you look back at it, what are some maybe some good, bad and ugly or only good and bad, whatever you come up with things you can ⁓ warn or mention to our listeners. Like if you’re thinking about going into this product mindset, these and these things, they are super important, but hey, these and these things, these are common pitfalls.
Anton Melander (14:41)
that’s a good one. am one of the, one of the biggest challenges because we were, you know, we were a, our, our customers, so to speak, or stakeholders were our colleagues. ⁓ so, and, and obviously they were always more, the backlog kept growing forever. So like prioritizing is sometimes like really difficult part where, you know, there are the things you need to do because like, you know, we can’t
David (15:07)
Yeah.
Anton Melander (15:11)
we can run this process or like this is going to be way too slow unless we get the support from you. And that might sound really important and there might be internal stakeholders that are very loud, really push it for their thing. But sometimes you need to do the things, like that doesn’t mean you actually need moving the needle of are we meeting the right business results because of this. So trying to tie down all of the initiatives to actually like
quantifiable outputs and quantifiable results. I think that was one of the most difficult part to do. ⁓ And sometimes you would have some really senior stakeholders telling you this is the most important thing. And you’d be like, but it’s actually not going to increase our quality. It’s not going to increase our throughput. and that’s, these were the most important things always for Northvolt. ⁓ So like, how do you deal with that? Then the kind of internal politics and so on.
David (15:55)
Yeah.
Yeah, that’s…
Willem (16:12)
With
the ever increasing backlogs, ideas are cheap. They’re the easiest part of any project.
Anton Melander (16:18)
Yeah.
Yeah.
And the other thing
I would probably say, the thing I would look out for is how you manage your master data. So as I mentioned, ⁓ you can’t really do OEE unless you have somewhere to, you what is the target cycle time for this machine? Or if you don’t know the planned uptime, you can’t really estimate your downtime or you have to assume it. ⁓ And I think…
making that easy. It’s very easy to kind draw this perfect architecture of how different system owns different data and how they interact. ⁓ But that also makes the whole kind of the speed of the organization dependent on certain technical achievements, which I think it should try to kind of minimize.
David (17:07)
The story of Northvolt came to a way too early ending. What was that like in the last time? What made you decide, this is the thing I’m going to do next?
Anton Melander (17:17)
Yeah.
⁓ So I left in mid 2023, pretty much. So I was not there in the very end. ⁓ But when I left, think there was a fairly… Like the morale was really high and I think that the progress was high, but apparently not high enough. ⁓ But the reason why I left is… So I had been working for Northfall for almost six years.
I had this experience of building up both the technology and the teams around it. And I started thinking more about the options we had where either we would go with off-the-shelf solutions customized for our needs. I know that would have been expensive. Or we built this internally, which we’re also a fairly large team to do so. And we started looking, me and my co-founder were like, but how do all…
Like the majority, think more than 90 % of all manufacturing companies in the world are employees less than 200 people. So like how do they deal with kind of digitalization in general? ⁓
David (18:39)
Yeah.
Willem (18:42)
team of
140 people should solve it.
Anton Melander (18:45)
Yeah, probably not, right?
David (18:46)
You
Anton Melander (18:47)
and it was this… Yeah, sorry, go ahead.
David (18:50)
No,
no, no, I didn’t want to interrupt you, but you are totally right. That’s also a bit of a problem. The success stories which are shared in the world are almost exclusively from a rather small set of global players.
Anton Melander (19:05)
Yes, yes and and ⁓
I mean, the whole kind of the modern data stack is a thing, right? Like if you go to modern tech companies outside of the manufacturing sphere, you have like, know, 5Tran and Snowflake and DBT and Tableau and all of this. it’s, you know, essentially a very powerful stack of tools that demands a certain set of competences to set up and maintain. But if you look at like non-tech companies and especially manufacturing companies, they…
Like that stack is not really accessible for them. And Norris is paying like millions of dollars to consultants to digitize their operations. So they’re kind of left with no really good options on the market. And I think you mentioned it a few times in this podcast, but like a lot of manufacturers, they live and breathe Excel. That’s where like they consolidate data, aggregate data, do their analysis and reports.
Willem (20:08)
That’s how they vibe codes, it’s just using Excel.
Anton Melander (20:10)
Yeah, yeah,
and, and, ⁓
that doesn’t really scale. ⁓ And actually when we started running, we were trying to fix the problem at source. We were trying to build an MES that would be easy, accessible and easy to deploy and highly customizable by non-technical people. But we also realized that actually to implement a ⁓ transactional system, you need to ⁓ force change in organizations. Something needs to change.
And change management is difficult and it’s risky. that tends to be a risk that you have. The barrier to get started is tend to be pretty high. And what we saw, I think we spoke to, since we started, we’ve probably spoken to more than 300, like small mid-sized manufacturers. And although most of them would like higher quality data, what seems to be the common, the more acute problem is that they have a lot of data, but they have a sense of
not fully utilizing that data. like, so there’s probably more value in the data that they are able to access. So that’s kind of the problem we’re going after with Ronja now is really, you know, trying to plug into the data they already have and make it easy to access by non-technical people, both like from a visualization and an analytical point of view.
David (21:43)
So what does that mean? Making it easier? ⁓ Is that like, like really black and white here? Is it like creating a data lake? Because that’s a typical answer. Like, or at least on the IT side, like, let’s, let’s.
Anton Melander (21:59)
Yeah.
Yeah. No, I mean, so some degree it’s a data warehouse, but a data warehouse is not going to help too many people unless you can actually do something with it. ⁓ So we do ingest data of various formats and shapes. ⁓ And then we have a fairly opinionated kind of ontology.
⁓ and a data model for how we should represent, like what is a cycle, what is an asset, what is a resource and so on. And then we make it easy for people to visualize this data. So a bit like Power BI and Tableau, but also kind of drilling down in that data inspired by like Jupyter Notebooks, like the way that data scientists would do it.
Willem (22:48)
⁓ That’s quite the pivot also, yeah? If I could from let’s build a package from something that I’ve been working on in past all the way to let’s work on visualization.
David (22:48)
The, sorry.
Anton Melander (22:56)
Yeah.
Yeah, yeah, it is, it is. ⁓
Willem (23:03)
Is that like the
biggest problem that you were seeing compared to the MES part?
Anton Melander (23:07)
Yeah.
So I think the problem is like we’re to a large degree still after the same problem, namely that, you know, empowering small and mid-sized manufacturers to be more efficient. But instead of trying to kind of control and help them run production, we help them make sense of the data that they have. ⁓ And the most like in order to kind of work with these customers, need to find like a very, the tool must be very easy to get started with.
And then it’s easy to kind of sit on top of the existing data and let them kind of analyze the data more efficiently.
Willem (23:44)
I think it makes sense because what you also said is like they already have lots of data. think compared to let’s say 20 years ago, getting the data was probably a bigger challenge than it is nowadays. It’s more about how can I combine it easily and spend less time in Excel, even if you’re a small manufacturer.
Anton Melander (23:56)
Mm. Mm.
Yeah.
Yeah. So I mean, in typical cases, they have some supply chain data in emails or whatever that might be that says we have ordered this material, I should come this day. They have their production plan in Excel. They might have some PLC data in a historian or they have operators kind of recording this manually. And then they try to understand, we, have we, know, do we have the right capacity and the right people to deliver these order and do we have enough material ordered for it?
And the data is there. It’s just like a lot of data punching and tends to be a lot of ad hoc kind of combining this data and drawing conclusions from it.
David (24:44)
So you talk to 300-ish small midsize manufacturers, which is, if I just think about it from an agenda perspective, that’s already like a huge, huge effort. There has to be some other takeaways from these 300 conversations. I’m fully with you.
surely have a data or a data accessibility type of issue. What are some other takeaways from these conversations?
Anton Melander (25:20)
⁓ I mean, I guess it’s the common things like, know, aging, aging population and attracting talent and the whole workforce and like onboarding, educating people. ⁓ I think that’s, that’s a big one. And ⁓ another takeaway is like a lot of the kind of C, C-suite people we’ve talked to, they’ve been, they tend to been in the company for quite a long time and they are.
You know, they really want to find ways to improve day-to-day operations. like, if you’re too small company investing in MES, that’s going to cost you a lot of money for two years. Like if that fails, it’s not good, obviously.
David (26:04)
No, and it blows
your entire budgets for a couple of years as well.
Anton Melander (26:07)
Yeah.
Yeah. And so like selling into these companies, efficient and really like providing value needs to be, you need to find like what is the least path of resistance to get in, and prove some value. And most typical like industrial software projects or product are sold as a project that goes for a certain period of time. And in many cases, that’s absolutely needed. If you are
putting in a new MES, new ERP, whatever, that’s not a small feat. ⁓ But these are also transactional system. need to, like, you need to change the behavior of the whole organization typically. And for a startup like ours is, you know, no one would probably trust us to deliver both the product and the governance to kind of get this in. So, you know, I think where we ended up with with Ronin now is, ⁓ you know,
David (26:48)
Yeah.
Anton Melander (27:06)
The barrier to get started is almost none. You can take the data as it is and start using it. And the interesting part of what customers that are using this today is saying is that they can focus on actually analyzing the data and trying to understand how to improve or find root causes as opposed to spending a lot of time grabbing the data from different sources, thinking about how to structure the data in order to visualize it. So you can go from like messy data to…
visualization and starting analysis much faster.
Willem (27:40)
think we’re almost running at the end, David, but I do have one question. ⁓ You’ve talked and worked with a lot of the smaller manufacturing companies, which I think is a often forgotten domain. ⁓ They have different challenges. How would you see digitalization work for them? Like if you would go a bit longer term? Is it realistic for them to say we’re gonna have… ⁓
Anton Melander (27:55)
Mm.
Yeah.
Willem (28:07)
big teams and we’re gonna go fully on digitalization. How would it look like?
Anton Melander (28:13)
I don’t know. think it depends also the type of product that you’re making. If you’re highly specialized and, you know, have a unique product, that’s very different from if you’re a commodity. But what we’ve seen with a lot of these smaller factories, like improving only like production and shop floor data might not actually have, it’s hard to translate that to bottom line results. And I think that’s the challenge. they tend, like, you typically need to look at more source of data from like,
suppliers, customers, ⁓ production, how you go from a demand to a production plan. So don’t know if that’s the thing. The thing that’s kind of struck me, which is also something which made us kind of move away from the MES, is that the closer to the shop floor, the more unique each factory becomes. And they have their ins and outs that is highly unique. And even if you could, I guess,
David (29:08)
Yeah.
Anton Melander (29:10)
vibe code an application that would fit for their needs. It’s like you need to maintain that and kind of keep that going. ⁓
Yeah. So I think like the closer you are, the close, like the more unique it is for the business. And if you can’t deal with that yourself, you’re to end up paying consultant a lot of money for it. It’s expensive. ⁓ So I think, you know, I hope there is like some, like with the whole AI thing that we can, we can kind of, there’s going to be a layer where you can put the custom things and I can feed that into something more generalizable.
David (29:48)
I that’s a perfect way to end this episode or a perfect statement to end the episode. It’s an interesting and not so easy to make trade-off. It’s one of the reasons, Willem and I also say that so many times, one of the, would say, manufacturing, yes, you’re talking about long life cycles. Yes, you’re talking about customization everywhere.
Anton Melander (30:00)
Yeah.
David (30:17)
And that has an effect on everything which flows, I would say, ⁓ downstream the data pipeline, so to say. ⁓ But thank you so much, for sharing these insights from a very small team ⁓ to become a very, very big company and then into the startup scene, wishing you all the best. ⁓
Anton Melander (30:27)
Yeah. Yeah.
Hmm.
David (30:47)
with this endeavor. ⁓ To our listeners, thank you again for tuning in. As you know, you can still subscribe to this podcast via itotinsider.com. We also have our ITOT Academy with new cohorts ⁓ starting in January. So if you wanna be trained in these topics, ⁓ make sure to go to itot.academy. ⁓ So yeah, see you next time.
when we bring you more insights in bridging the IT and OT world. And until then, take care, bye bye.