AI isn't a future promise. It's here, and it's already changing how operations run. Real impact happens when AI becomes part of the process, not an add-on. This isn't about predicting what's next. It's about getting started with what AI can do right now. Combine artificial with human intelligence. The organizations pulling ahead use AI every day. All over operations, they learn fast, they move fast, and continually reinvent how work is done. Across industries, this shift is very real. AI is scaling across functions, and it's breaking down silos, solving operational challenges, and providing measurable results. AI won't replace people who understand how operations truly run, but those who embrace it, practically and confidently, will shape the next era of operational excellence. AI won't replace people who are in the same place. Hello, everyone, and welcome to OPEX 26. My name is Chris, and together with my colleague and co-host Pedro, we are so pleased to have you with us today. Over the next 90 minutes, we will focus on one core question that is top of mind for many leaders in operations right now. How do we move AI from promise to impact? You will hear from Ethan k Eben Jackson-Moloch, who will give us how AI and its goals, which is from nightmare about you, and what theçители, any other way to set aside from learning edge A SAANG let me hand over to Pedro to frame why this moment matters. Fantastic. Thank you, Chris. And welcome, everybody, and thank you for joining us today. So what is different in AI and operations today than it was a few months ago? Why does it matter and why are we talking about this? Well, behind me, you're going to see a graph. And in there, you're going to see every single model that was published within the last couple of months or maybe a couple of years and their performance improvement for the next model right after. What you're going to see and probably notice is that each new model has a big jump in performance. And why should you in operations care about that? Well, there are several reasons, but one of the main ones is that every single leap that you have, there are more problems that you can solve in operations, whether that is in supply chain, in procurement, in manufacturing, in planning. Now, this opens a whole lack of possibilities for us to work with a completely different ways. Now, there's other reasons why this matters as well, Chris. And one of the main reasons for that is because every time that we've been through a technology cycle, there was a lot of companies that were basically allowed to sit and wait. The response was to do nothing and wait for the front runners to basically test the waters, create the playbooks, outsource that thinking to IT, and get the solution when everything was ready. That reality is fundamentally different today. It's fundamentally different because if you wait, there are going to be the front runners that are actually trying things out and creating the right capabilities to take the next steps, learning from it, and take those things from pilots to real scale productions. So now what we see is not just one single cycle. We see a compound curve from the laggards that are being really, really left behind and the real front runners. And this distance is just completely increasing. Yeah. It's really, really interesting to see that this is where the implication becomes extremely concrete for leaders within operations. It is not a technology discussion. It's an operating model and leadership discussion. It affects how decisions are made, how the work gets organized, and how quickly we're able to turn insight into action. What we see in the leading organizations today is a very clear shift in the focus. AI suddenly is treated as part of the core operations. That means that ownership gets made very explicit. Value has an owner. Decisions have an owner. And accountability for turning insight into operational results is very clear. And importantly, organizations invest in their people. The planners, the engineers, and the leaders are all trained to work with AI insights and output, to question it, and to act upon it. That means that human knowledge still remains central, but it is augmented. And I think what a very interesting part of that, right, Chris, is that when all of this comes together, learning is becoming very institutionalized, meaning that you know that a new break will come in a month or two or three, and then you basically need to catch up. That becomes a capability. That becomes something that the organizations, and particularly in operations here, you need to cope over time. Yeah. And today what you see, if you fail to move ahead on this journey, you get stuck in what we often call pilot purgatory. That's the place where promising proof of concepts don't scale. Where the tools you have are disconnected from a platform and from real processes, and the energy of the technology moves a lot faster than the organization itself. The reason for this is really very simple. The technology moves so fast, but our way of viewing processes, organizations, and structure means that they move a lot slower. And this gap is where a lot of AI transformations stall today. And unfortunately, impact fails to meet the expectations of the organization. scaling AI is suddenly not about deploying the latest and greatest model because they come along all the time. It's about building the system around it, allowing for the organization to experiment. And when that system is in place, then a new model is no longer a transformation program. It's a drop-in upgrade. And that's when impact to operational result becomes very tangible. And that's what we're going to be looking into today, right, Chris? But before we move on, we also have two simple tasks to you guys. The first one is that we really would like to interact with you. So when we have a poll, please answer it and let us get to see some of the stuff that you are thinking about or some of the options there. The second is be active on the chat. We're going to conduct a lot of interviews. And throughout those interviews, you have the opportunity to participate and ask the questions that you're sitting with. Absolutely. But enough from us. Let's get to the first part of the session today. We want to start off with a rather thought-provoking piece from Ethan Mollick. He will hone in on how AI actually changes work, organizations, but in practice. He is one of the voices that is shaping how leaders think about AI today and focusing on the very concrete efforts that sort of exist within the collaboration between humans and AI when you want to scale. And these perspectives are exactly where we want to begin the journey of this session. Without further ado, Ethan Mollick. See you soon. Bye. So one of the most confusing things about AI is that it's actually kind of a reused term that suddenly changes the meaning. So we talked about AI prior to 2023. We would be talking about big data, set up your data lake, organizing your data really carefully. It would be a very technical thing with data scientists. Post-2022, right, with the advent of chat GPT and generative AI, it's a very different beast. It's something that actually you don't want just tech and data scientists to lead. It has to be led by operations, by HR, by people who are hands-on with these kind of systems. And so one problem that companies have had is viewing this as a technology problem rather than a people and process and creativity problem or opportunity has led people astray a lot. So I have this model I tend to follow when I talk to companies about AI and then I talk about leadership, lab, and crowd, the three things you need to do to succeed. And so leadership are C-level people inside the organization, the CEO, the COO, the people who are in charge. And they have to have a vision for what the future looks like. This can't be bloodless. This can't be like an email you send out saying, AI is changing everything. Use AI. It has to have a vision for what the company is. I thought about how we align rewards and incentives in the right direction. And basically, how do we think about what we want our company to be in a world of AI? The crowd is everybody using the system. I talk to the AI companies all the time. I talk to all sorts of people about AI, consultants, organizations, CEOs. And the thing that comes up over and over again is the fact that nobody knows anything, right? Like OpenAI tells me they use my Twitter feed to figure out ideas of how to use AI. So there's no magic secret out there that is like, here's how AI should be used. So you have to invent uses. And you can do that with the help of consultants who are very close to the ground of what you do, but also more broadly inside your own organization thinking about how to solve these kinds of problems. And so that requires expertise. Like having an outside person tell you whether or not AI is good or bad for your particular job in your particular country or context is expensive. But having your employees figure out how to use it is pretty cheap because they'll know instantly, is this a good result, is this a bad result, is this useful or not, what kind of feedback to give. So the crowd is everybody discovering AI uses. The last piece you need is a lab. You need actually to be doing R&D, but in a very practical way. The lab is what we call ambidextrous. It does two things. It both releases immediate products for people. So a new invention is discovered by people who are like, this prompt works really well for producing our annual report. Great. Then the lab helps come up with that report and send it to everyone, that prompt to prove it and send it to everyone. They also are benchmarking to figure out what AI is good or bad at. And finally, they are doing the kind of work where they are thinking about the future. What does it look like in a world of agents? How do we build things that don't work quite yet to figure out what they will do? So you really need the whole organization working. And to kind of return to that leadership role, the vision needs to come from a senior person who outlines why and how to use AI. Because otherwise, the rest of the engine doesn't work, right? Think about the crowd. So let's say you give AI access to everyone in your organization. Great. They have access to AI. They already did, right? In fact, in surveys in the U.S. and in Denmark where we have the best data, we're approaching sort of 50% AI use. People use AI at work. Most of them are using their own secret system. They're not telling anyone about it. But they're reporting personally huge productivity gains that don't show up in the stats. Why are they going to show up in the stats? Because I'd be a fool in most companies to show you I'm using AI. First of all, maybe it's banned. So I'm just using it secretly. But even if it's not banned, I look like a genius right now because AI is answering all my questions. Why would I ever want to show you that I'm not a genius, that AI is the genius? Or there are companies where they know that the goal is cost-cutting. And as soon as I show you that I'm productive at all, what's going to happen, I'll get fired or my friends will get fired. Why would I want to show you I'm using AI? Or they're working 90% less. Why would they ever want to show you they're working 90% less? Or they don't think you'll be rewarded or they don't even know who to go to about this. So leadership has to set up a bunch of goals but also a vision for what the future looks like and incentives so that people are willing to talk about this and use it. It's not just one thing. And all of this comes back to the bigger issue, which is invention. This is an R&D process. It has to be led by the company. You're going to have to spend effort to do this. You hopefully will fail a bit while you're doing this because if you're not failing, you're not being ambitious enough. And you're going to figure out where AI works in your organization and where it doesn't. Sometimes that means using outside vendors and consultants to help you. Sometimes that means doing the work on your own. But you're going to have to think about how you reinvent things. And ultimately, the nexus where everything is going to happen is in operations, right? Operations at HR. It's going to affect how your people do work and what kind of processes and approaches they use. And you're sort of off the map. Nobody, there's no canned answer to this question. Nobody knows what the solution is going to be. We're all going to work together to try and invent them. So this is an urgent problem. And waiting for someone else to solve it for you really isn't an option. Because by the time they solve it and put it into a product, the opportunity is already gone. So there is this kind of effect, which is like the basic request you put into AI. Everybody can do this, right? So everyone can get the same information out of these systems. So there's a few ways to think about that. First of all, for non-core functions, the fact that you can get up to 80th percentile and as good as someone else, like, that's great. That's it. For an operations person, the idea that I could just kind of get good operational excellence out of the system. You know, we did an early study with consultants. We found consultants, for 80% of consultants, using GBD-4, just turning the GBD-4 answer was better than their own answer. But what really was happening when you looked at it was that whatever they were weakest at, the AI did a better answer than their weakest answer. But whenever they were strongest at, they delivered their own. So the idea that it's samey is not that much problem. On the other hand, if you want very different kinds of answers from the AI, you can do that. That just requires you to do better prompting and give the AI better context. So if you give the AI your specific context, you can push in directions no one's ever seen before. The meta problem that we have is this is a new form of innovation. It's not actually new. Like, if you look at the industrial companies, and especially the great industrial companies of Europe and the United States, right, they have a long history, in Japan as well, they have a long history of organizational innovation, right? The reason why American companies often succeeded and a large part of their gain, according to some of my friends at HBS, is management. Good management matters, right? And so there's lots of innovation and management techniques, whether that was inventing the modern organizational form of General Motors with Sloan, whether that was thinking about GE and its heyday and how it would think about, you know, moving people through management roles, or Toyota's management system. All of these are breakthrough ideas. If you look at the last 20 years or so, though, there's been a slowdown in management innovation. The trend has become, I don't need organizational innovation. I don't need management innovation. I should be outsourcing those things to SaaS vendors and focusing on just product and service innovation in my narrow area. And I think that's blinded us a little bit to, yes, we have to reconsider how organizations work. And that's a challenging problem. If you start breaking up your IT group and putting senior managers directly in the field, what does that mean for IT and engineering? How do we start thinking about projects? They're hard problems. They're not necessarily easy ones to solve. I would also say that we are seeing, I mean, we're, you know, only a couple years into this, right? GBD4 came out in early, you know, 2023. Like, companies have just started adopting this stuff. I'm kind of shocked that we are actually as far along as we are. But process takes time, right? The technology is moving ahead much faster than the actual process inside organizations. And we'll see an adoption curve happen. So people are seeing kinds of gains. The question is, how do you harness those at the organizational level? All right. So a little bit of complexity and some promise in these words from Ethan. If we just try and grab a couple of talking points out of this, at the nexus of this, in operations, we could have a tendency to think that AI is not really aimed at us. And that someone will either solve the problem for us or it's kind of IT's issue. But for AI to be successful, this is not a sustainable way of looking at these implementations. So let's talk a little bit more about what do we need to succeed? We need to nourish innovation. We need to take ownership of that within operations and encourage this experimentation also at the details within the process levels. But then what are some of the logical steps that we could take? Yeah. And I'm very looking forward to say something around that, Chris, particularly when we had that nobody knows anything kind of remark from Ethan. But, you know, if we zoom in and to say, what are the others doing? What have we seen the patterns out there? There is some kind of evolution into this. There is a lot of logical steps that people are taking and that we're seeing the tangible results out of it. And we basically try to illustrate this for you into these five steps, right? The first one starts at the individual use. And that's where most companies actually start, of course. It's this stage, AI helps you work faster, but nothing changes in how operations really run. So we're talking about engineers using to search manuals, procurement teams to draft RFQs. The value is very, very real, but it stays very personal and local. It's personal productivity level. Then you go into experimentation. Very interesting here is that organizations start testing AI in operations, but mostly through pilots, right? So you have maintenance experiments with automated work. You have supply chain tests, demand forecasting on machine learning. Each case shows a lot of promise, but scaling is hard because data, process and ownership stays exactly the same as they are today. Now things start getting very interesting when you go into operational adoption. That's level number three. We basically see that AI is now embedded into specific process that really, really matter. Now, forecasts generated by AI are used in monthly SNOP cycles. We see production planning using AI to balance capacity and constraints within a specific business unit. Impact is measurable. You can see bottom line and top line, but you're still limited to defined areas. Now, 2026, everybody talks about agents. And here is where agents comes to play. That's level four. That's where agents also starts to matter. Because you have planning, procurement, maintenance, logistic agents rerouting shipments when delays appear. You see a lot of 3PLs already working on this and using that. And you can see the impact at scale across different geographies. Humans stay still in control, but agents handle repeatable decisions inside. And the word here is very important. Standardized processes. Now to the holy grail. To the system level advantage. Of course, there's very little companies that are actually reaching that. At the moment, a lot of them are still in the innovation like Ethan was mentioning before. At this level, networks of agents operate across end-to-end flows. So you don't have a separate silos. But planning, quality, maintenance. Everybody's basically working together to try to optimize some kind of workflow. Operations leaders then they shift from approving actions to setting directions. And more on the guardrails. To say where can you operate within. This is not automation of tasks. It's a really new way of running operations. The bottom line in all of this is that most of organizations that we look out there are basically in level 1 and level 2. Which you have a lot of productivity level at the personal side. But you don't have a systemic one or in the workflows. And real advantage starts at level 3 to 5. Getting there is much less about the tools. And much more about process clarity, ownership. And really, really a lot of operation discipline. Yes. Exactly. And I think if we want you to take one point out of this whole thing coming from Ethan and us talking about this model is scalability. Your ability to move from what seems like a really promising pilot to something that can over time become an enterprise-wide implementation. That will support multiple use cases and multiple processes. But having a focus on sharing data, increasing reusability over time. And standardizing the processes and the governance that you need around this platform thinking. Without this, you end up with standalone solutions that will fragment your effort. And that kind of exhausts the people in your organization because it becomes a new thing on a new thing on a new thing. But with it, you're able to create velocity and room for the innovation. As you listen to the conversations that we will be having today, try and consider a couple of these things. Which capabilities can be made reusable if you look at your own organization? How would you scale from a pilot to a platform and having a portfolio of use cases that you could slot in quite simply? And how would your leaders, the people you work with, sponsor the shift in operating model so the teams are empowered and the outcomes improve? And just remember, this is not a technology challenge. It's a business opportunity. So with that, let us introduce our first case company, Metis. It's a leading Nordic retailer operating in a highly campaign-driven context. Imagine rapid growth, creating forecasting complexity, changing product hierarchies, and heavy manual effort. To solve this, Metis embraced advanced machine learning to automate daily order forecast. They moved data and reporting into a cloud architecture and have cut manual work significantly. Please join us in welcoming Mere here with us, who will share how leadership alignment and great pilot ideas is turning AI into an operational engine. Thank you for joining us here today. Thank you for having me. Let's try and start with the bigger picture and you describing a little bit more about the old world or the problems that you were trying to solve when you decided to embark upon this journey with AI. All right. So I think, imagine you're hosting a party and you're planning it and then suddenly that party turns into a giant wedding. Then that's going to be difficult to keep on a sticky note. And I think that's essentially what's happened to Metis' e-com business over the past seven, eight years. We grew from being absolutely insignificant within the business to now being more than 30% of the business revenue. And growth has happened really fast and so has the amount of complexity forecasting. And complexities has been added at a level now that manual forecasting is really becoming very time consuming and less accurate because the human brain just can't comprehend all the kind of attributes you have to work with. And we're starting to see that it's affecting the warehouse staffing. We're pulling in too much over time or we're having too many staff at work not doing anything. And just decided that we have to do something because if we don't do something now, then worst case, our delivery promise to our customers is going to be hurt. And let me pause you just a bit here because when you mentioned, you know, going for a party to a wedding, there's a lot of things, there's a lot of complexity. Can you put a few more words on what this complexity actually is for you? Yeah, absolutely. So we have many different order types. So we work with the classic e-com order, but because we're an omnichannel business, we also receive almost 20% of our orders that comes directly from the shops. And the amount of orders that will come from the shop is very much affected by the stock amount in the shop, the traffic to the stores, the weather, whether there are people in the street, the campaign we're running. We also have a complex product mix. So we have both the digital products and we have physical products. Some create orders and some obviously don't. We have operations in both in two main warehouses. So one is our specialty warehouse and one is our main logistics centers. And that means an order can actually be split into two orders having to be operated in different warehouses. Then we have, because of our omnichannel setup, then we actually do fulfill 8 to 10% of our daily order count in our stores. And we also work with drop shipment from our suppliers that they do the fulfillment directly from them. And then we have different sizes of products. So some are odd size, some are very tiny, and it also depends on what you can put in one order. Okay, that's a lot of questions. And I actually just want to add also, like, because we're fast moving consumer goods, then we have a very heavy campaign load. And the kind of campaigns we're running, and there's a new campaign every day, really affects the sale we're driving in and the basket sizes and the amount of product lines that goes into each order. So I think, at least for us, but also hopefully the audience understands that this is a very complex world, of course, to navigate within as a human being. So what makes you think that AI is a really good solution for this back at the sort of origins of the projects that you've been undergoing? Well, as I said, we're at a scale now that a human can't comprehend the data we're working with anymore. And then we also had a vision or a goal set from leadership that said, well, you have to improve accuracy and you have to do it with the same amount of staff that you have now. So taking that to an operational level, well, obviously, we have to automate something because we can't work faster than we were doing. And then together with implement, we decided, well, advanced machine learning is what we need to do. And also because there are like three parts to it. So one is impact. The second, I would say, is speed. And then the third is scale. So we can load a lot of historical data into a machine learning platform. We can work with all our seasonalities. We can work with our deep analysis of our campaign data. And we can comprehend the amount of different kind of orders and where they're being sent to. And then the speed. So we could get pilots up and running really fast. And we could test as we were going. So test and learn, test and learn, and then improve as we were going. I think that has really been like the two main drivers. And then because we are the biggest retailer in the Nordics within our field, then we also have to look at scale. So right now we're just operating on the Danish market, but we're going to want to scale to all the other markets we're working on. And we can easily do that with the algorithm we're working on by just loading in new data. Yeah. No, that sounds great. But so if we talk about the impact of this. So now I know you're still ongoing. So it's a never-ending journey. But for what you're seeing now, what has changed both in terms of the measures that you would like to improve, but also on a sort of human level? What is changing when you come to work today compared to previously? So first of all, as you said, we're still in hypercare mode. And the goal is that we will improve our long -term forecast by 2% to 4% over the next year. And I'm very positive we're going to get there. On a human level, what has changed right now is that we're actually spending a lot less time searching and aligning historical data and just putting that into an Excel spreadsheet. And now we can spend a lot more time understanding the results of the forecasting and spotting the exceptions that are coming in. So we can react a lot faster than we used to. And that just creates a lot more space for value-adding work. So if we think about the change or adoption journey that's happening within your teams today, are there a couple of things that you want to highlight as having been either particularly challenging or that perhaps surprised you when you've been undergoing the journey? I would say it surprised me that tech was actually fairly smooth. I think that's not common. But so skipping that part and then looking into a challenge that I did not foresee we would have would actually trust. So when you're used to doing something a very particular way and it's a manual process and you've done it for eight years and you're damn good at it. And then you now you have you get this machine coming in and you have to trust the machine that it's better than you. That becomes competitive and you have fear of losing control. You have fear of losing importance within your company. And we're trying to challenge that by just making it very transparent what data is coming in, what data is coming out and how is it being manipulated and undergoing. But we're also setting up a lot of dashboards so that we're keeping our human overrides, so to speak, in the system. We're logging every change that's being made and then we're kind of competing against the machine. Right. So every human likes to compete. Right. So we're challenging ourselves to see can we still be better than the machine? And so far we're at par, but we're also starting to see that the machine is actually it's going in the right direction. Right. And that's really creating a lot of trust. And actually, there is a question that basically alludes to that. So what did Maitis do to convince the demand planners, you know, assuming they're still responsible for the demand planning KPIs? Yeah. So I pretty much had to convince myself. And I think the vision from or the goal from the sea level was that you can't have any more human manpower. So there was no way around looking into automation like I really had to just accept this is the way we're going. And it has been a challenging journey. And I think doing these competitions and continuously showing that the machine is better than the human actually is what's doing it. And there was another question. Yeah. Maybe I'll follow up on that. How long did it take to implement? The tech part was just a couple of months actually. But that's also because we had very structured and clean data to work with. So all our data was in Databricks and it was in nice columns and rows and everything and nicely labeled. So it was easy to work with. And then we also had a lot of experience from refining our own forecast model from the past eight years. So we knew what attributes affect things. And that made it a lot easier. But a couple of months. And I think the adoption change is going to take at least another six months until we're like completely satisfied that it's there. And it's going to be routine work looking into it. So if I just grab that comment because it makes me think. So you're saying you have another sort of six or so months on the adoption side. Can you envelop a little bit on some of the activities that are actually ongoing? So what is your plan right now to actually sustain this adoption? What are you doing for your team members and the people working with the tool? Yeah. I think we set up a pretty basic change management plan. So we looked into skills, tools, and then how are we going to continue focusing on the change? And I think it's important to cover those three pillars. Because skills, when you have a team that has to take on a new task that they have not done before. And you put more responsibility on them because you now have automation that can support them. Then you really need to train them. And we set up the physical workshops with prepared use cases that they could go through. And what we're doing now is that we're also doing, spending a lot of hours doing one-on-one sessions in real-time data. To make them confident in decisions and also to teach them about what exceptions to spot and what actions to take. And then the tools. So it's very important to set up the dashboards that can help the planners and not create pressure for the planners. So we've created dashboards that have the thresholds to meet for the day. And also to make an alert when there's something that's odd. And then just the change plan. And so we're putting in weekly reviews and monthly reviews and looking into doing quarterly deep dives into what has gone wrong and where we're moving to. Any sort of surprises, let's say. Something that hasn't gone according to plan that maybe you want to share with the audience here as a lesson learned? I think a lesson learned is I was surprised that, as I said, that trust was going to be a challenge. And even for myself. So focusing on how you can be transparent and make it an easy process for change. And then a good lesson is, again, as I said, it's having control of your data before you start a project of this size. Like it will make things run faster. And clean data, in this case, is really real. Like the saying, garbage in, garbage out, is absolutely true when you work with machine learning. So maybe just one reflection that we heard about Ethan talking about that entire, you know, the lab and the crowd. And did you have any kind of similar approach or how did you actually get going? Because, you know, he kind of had this thought that the lab is a separate thing that tests a lot of things. But you seem to have gone, you know, straight into let's try it out. Let's see if that works. What are your reflections on that? Yeah, well, yeah, sort of. I think there's no way you will have these strict lines between everything. But I agree with his approach that leadership sets the vision and the goal. And they also did in our case. And they always do that, actually, in my sense. But letting us, the operation, actually deciding on what direction to then go and how to do it is what made it an actual success. And absolutely involving planning staff and warehousing staff and commercial employees really made the process smoother and understanding the data and the direction we wanted to go. And then having implement on, you would say, as the R&D to help us decide on what kind of machine learning to work with and building the model for us and asking us some of the difficult questions that we weren't able to ask ourselves because this was a new journey that we were on actually made it a big success. And now we're really talking about making this at scale, right? And we want, everybody wants to have also their own capabilities and grow with that. And I'm just wondering, so did you experience that? You know, Ethan talks about the innovation part that is becoming like the integral muscle. So now you have a new tool. You have new ways of seeing things. Can you see already people trying to embrace or trying to try new things or coming with new ideas? Or where are you on that? Oh, God, yes, yes, yes. So Metz is like any other company has had AI on top of the strategic list for a couple of years now. And we have machine learning was the first step. And we started last year. But what I'm actually seeing now is that I have a team now who's OK. We can do machine learning. We understand it. Let's Vibe code. They are Vibe coding like crazy now. And they are coming in with, can we alter the algorithm so it can do this? And they're really like excelling in trying out new things. And they're less scared now to just try out the tools that are out there to help them. That's amazing. Let's see if there's a couple more questions from the chat. I can see they're popping in. Yeah, there are a couple of ones. But I think there is something around, you know, how do you fix the quality of the historical data? Do they have issues with data quality? I think you answer part of that. But no data set is perfect, right? So I think probably there was some work there still. Yeah, well, we had pretty good data. At least we got a lot of praise from Dean Cosanto that helped us. So I'm assuming that it wasn't. But I think there was some mess in it. So especially like the way we match up revenue to a campaign and order lines to a campaign. And that is not something that we had really done so much in absolute detail. And we had to work with that and do spend some time cleaning up that part so that we could use it better in the forecast model. Okay, great. I think sort of based on everything that you've shared now and now your personal experience of also having been through this for almost a year's time. Is there any advice out there for the audience or others who are about to embark on a journey on this, let's say, a slightly more nerdy side of AI with machine learning and so on? How do you start? I think you just jump in. And I think start by setting a vision and don't stop yourself by seeing limits. So what I think is that the AI is now so refined that you can pretty much do anything. You just have to be very clear on the vision and the outcome you want to see. And then I think there's going to be some kind of automation that's going to fit you in between. And I think don't hesitate to do it because it really is game changing in the amount of time you get to spend on real value adding work. That's amazing. Thank you so much. Thank you very much. Thank you for coming today and for sharing your story with us in the audience. You're welcome. Great. Now for something slightly different. Also something that we've been extremely excited to introduce to you. We're going to talk with NVIDIA and Sheffra. Basically how NVIDIA is working with an insane big company with plus 100 plants across the globe and helping on their journey in a completely different way than Metas on actually utilizing AI to bring efficiency into operations. Let's welcome NVIDIA. Amir, welcome. Thanks. So absolutely pleasure to have you here today. And I think for the people who don't know who NVIDIA is and what you guys do, can you maybe give a little bit of a glimpse on that? Well, let's do that. So I guess that NVIDIA is mostly known for GPUs and our stellar performance in the stock basically. So, but that's maybe not the full truth. So we are more of a software company these days. I guess that we are now closer to 40,000 people, employees, and 90% of them are engineers. So we are very engineer focused company. And 70% of those 90% are software engineers. And that might come as a surprise. And well, the task of those engineers and those developers is to solve use cases. And those use cases are sold in the lines of businesses, against the lines of businesses. So we find really, really tricky problems to solve. So usually manufacturing operations or any industry vertical. And we go about to build a stack to solve that problem. So that's what we do. And that's maybe just a great segue to the case today. Yes. So how are you guys working with Scheffler? What is basically that you're trying to do together? Yeah. So maybe we should set the stage by talking about Scheffler and what they do. So they've called themselves the motion company. So anything with motion, they're probably integrated in some way, shape, or form. So it can be powered trains, powered like anything that powers engines, gearboxes. You have robotics, components in robotics that make arms move. So joints, that's their business basically. And yeah, so what you can see the core objectives behind. And that's what we are aiming to work against. So at the end of the day, it's about integrating half of their plants. So they have 100 plus plants. So into Omniverse. And that's basically digitalizing the whole thing. So any assets within that, the whole processing, new processes that you invent in that factory. All of that basically. Okay, but I understand that. But if we just take a step back and think there is a variety of things we can do out there. Right now, there's a lot of our clients that are super pressured on the margins, on cost. There's labor. There's a lot of, you know, things happening around the world. Why invest on this? Why do this? Yeah. So I would say that this actually is part of solving the problem. So we talk about, for example, labor, labor costs. Well, if you can automate stuff, then probably you can digitalize, automate, and simulate and do all kinds of magic in that sense, right? You have, for example, supply chain problems all over, especially in this crazy days in the world. So you can go about and try to build stuff the old way. Or you can simulate all of those things and basically just see what goes, what cannot work, what will work. Let's say that you build a new factory and you cannot get new components into that. You cannot get the robots that you aim to have. You need to source them from somewhere else. Do you go and pilot that, like live and do that? Or do you simulate that first and see what works and what doesn't? So all these kinds of leverage you can do. So basically, we're talking about building a certain kind of infrastructure, right? Correct. But if we then approach that to a few concrete operational problems that they had, maybe you have an example or two to give to us. Yeah. A factory, a warehouse, or whatever that you're doing. So we have basically three very good ones. I would start with a Brownfield example. Yeah. So a new facility needs to be built in China. How do you go about doing that? Easy way, right? Well, you probably have a factory that works quite well. So if that is digitalized, you have all that like printed out in a blueprint. Why not take that blueprint, put it into where it's supposed to be, in this case China. And then you can simulate and emulate all of the various factors that are different. And by that, let's say humidity, for example, it can be different. And these machines are very, very complex. Like they have many moving parts. So will that disrupt the process flow of manufacturing? Can you compensate somehow? If you can simulate all those things, you take away a lot of risk. And of course, that reduces downtime and, well, increases efficiency. So are they actually using to say we have a flow from point A to point B in assembly of motors or whatever it could be? Yeah. To say how this assembly should look like in China and do it in a different way because you have digital assets? Is that correctly understood? That is correctly understood. And also if you want to implement or improve like incrementally, you can also introduce those digital assets in that new factory and see how it would work out. So it's both and more. So, okay. Now I see I'm starting to get the idea of here. I'm a bit slow. But if we basically look into how they traditionally been doing that, we're going to touch that in a minute. But I think a lot of us are wondering, it must be quite an effort to set this up. So what are the impacts? What are they getting out of this? Well, if we go back and see at the core objectives. So basically that's the objective. And the idea is to, at the end of the day, provide a digital ecosystem for all of these factories. And, well, we talk about efficiency, of course, driving down downtime, introducing faster implementation of new products, new product line, new processes. All of that is very much better done. If simulated first, you take away all the errors. And you can do it also in scale, in parallel, basically. So you can try out many, many, many things. And then you choose on the things that you like. And you focus on those. And then you start to prototype those. So sorry to drill you down a bit on that same again. But I just, you know, we are from operations. So I'm really after the hardcore numbers here. Hardcore numbers. So if we're looking at, I don't know if it's lead time or process time, have you seen any results on doing this? So benchmarked up until now then. And this is a continuous effort. We have a couple of other use cases we can go into. But we see basically a operational OPEX goes down by 60% the cost of that. 60%? 60% up until now. So it's probably increasing. And then also introducing new assets. So completely new assets, be it whatever, goes up by 40%. So the speed of innovation or the speed of implementation. All from like processes, but also into like proper new warehouse. So if I want to basically look at my manufacturing footprint. And I'm looking at a new factory. You're saying that deployment time for this new factory is basically decreased by 40%. Up until now and increasing. And it's quite easy to understand why. So you can go about, for example, programming a robot in a new factory in the old-fashioned way. You run into those problems of environmental differences, let's say. That can happen. And then you have, and that process might include 300 assembly points. So how do you go about that if you want to automate that and get that increase in speed? Well, you simulate it. And you simulate it as much as you can. And you automate everything that you can simulate. Of course, you will run into that 1% that's really, really hard. But then you use reinforcement learning, for example, for those things. So you would have a human being doing that. You have AI looking at what's happening and how to do it correctly. And you feed that into that 1% that doesn't work and you can automate that also. And this works especially well in prototyping. That is very interesting, both, I guess, from an R&D perspective when you start to introduce a new product. Yes. But also from a running production. So now maybe thinking about a company that is, you know, very traditional. There's a lot of engineers that work in a certain way. Yes. And now introducing something that for some it might seem completely, you know, from another world. How do you basically go about that? Yeah. How do you do that? Yeah. So, I mean, there are, it's very, very easy. Not easy. It's never easy. But for Schaeffler, for example, there is a clear guidance from the leadership basically saying that we are going to introduce this to half of our plans. And the idea is to introduce a digital ecosystem. Now remind me of that because we need to talk about that also. So if there is a clear vision, and I thought that we heard the word vision. Yeah. That kind of takes away the complexity of people and process and organization. Because if you can automate ideas and production lines working against that idea, then you will have the company behind you basically. So you want, and that takes away another problem also. So I think we heard the like perpetual pilot phase or something like that. That is a very, very big trap. So you do this proof of concept and pilots. They solve a technical problem really, really well. Everybody high fives and are happy. And then what? So where's the business outcome? And I think that that was also a theme. So work against that business outcome. Automate as much as you can. And even start with the complex problems. That's quite okay. If you think that you can automate those, just go for it. But have that clear vision in mind. I think that's the key. Okay. But then you touch on something, right? Because we just discussed with Metas that one of the advantages that they had was the data quality. Yes. That they had from the start. And now I'm thinking, okay, we've been to quite some factors that that might not be the case. So if we touch on data sharing, data quality, how do we even go about that if our data is not fantastically perfect? And how do they handle it? Yeah. So data quality in that case. So data usually sits in silos, right? And those silos can have various quality of data. So either you have good historical data you can lean on. So you have sensors, you collect those things, and they are accurate. All good. And that's especially, especially important when we talk about the digital ecosystem. So I'm taking a little bit of a sidestep here. No, that's fine. So you need to have that data quality in place because you would share digital assets with your ecosystem. So let's say that I make these robotic parts. I feed them into my robotic maker, which sells them back to us. It would make so much more sense if we use the same common language and same common platform. And we can try out and I can trust the simulation that's coming out of them to implement in my new processes, in my new factories. So are they doing that? You mentioned a bit on the ecosystem. Yes. So that is the whole foundation basically. And that's the Omniverse. That's the whole idea with Digital Twins and Omniverse in this case. So we kind of, we give to our customers and they give back to us in that sense. So we leverage each other in that sense. So maybe just now that I had the opportunity to ask a lot of questions, but the audience also wants to ask a few ones. So I'm just going to read two of them for you, but just start with the first one. If an operational leader in this room wanted to start tomorrow, what are the main capabilities that matter more than buying the right software? That's, you can answer that in a two way. Basically, I think the most important is still the vision and you work against that vision. And then the other one is, of course, automation. Can it be automated? If it can be automated, then it absolutely makes sense to digitalize it and use AI as much as you can. AI is very, very good in automation in that sense. It's kind of the purpose of it also. Okay. Then for the second one, for operational leaders here who are curious, but somehow may be cautious as well, because they have other things that are pressing in the budget right now. What is one smart and first move they should make and one common mistake they should actively avoid? Yeah. So one smart move, first move, and then one mistake they should avoid. So, yeah, this goes a little bit back into the people and process side. So, again, working against the vision, start small, but have that vision in mind, scale out in waves, and then you walk from there to that ultimate goal, basically. Do not do the perpetual pilot POC mistake. So have a buy-in on what you're trying to do. And when you do it, you see the clear business outcome that comes from that. I think that's the more important part. Okay. So now forward looking, right? So looking three to five years ahead, maybe let's skip to three. Which part of operations do you expect will look fundamentally different because of physical AI and other things that you're doing here? Well, I think for Schaeffler, but for the whole world, all of it. So I think that the projection is $5 trillion in new investments upcoming three years when it comes to manufacturing and heavy manufacturing. So that's new investments coming up. You can do it the old-fashioned way, and you can spend time like 80% in doing it, like losing 80% out of it. Or you can use it like in a new way, in a digital way, and using AI to do all of these magic things that you can see behind us. So what is this $5 trillion investment on? Is that capacity? What is it? It's new investment in manufacturing. So new plants, robotics, processes, that's the project that's within three years. So if you're going to do that, I mean, that's why we like that industry quite well, right? It makes no sense to do it in an old way if there's a better or new way to do it and enforce it with AI. Again, let's take that prototyping example, for example. I think for Schaeffler, we went down from 600 hours into half a day, so let's say 12 hours, from classic prototyping into new ones when it comes to robotics and part into robotics. Okay, so basically you're able to take one robot that you were supposed to integrate into the next factory, and you're going from how many hours to how many hours? 600 to 12. 600 to 12. Yes. I guess that's a pretty good return on investment, considering the cost of system integrators and other things in that. Absolutely. Absolutely. And that's only for one thing, and then you can do that and scale that out also for other parts. Fantastic. Well, I think we're running out of time, so thank you very much, Amir, and thank you so much for your time. Thank you. Yeah. And now we, of course, need to say thank you for NVIDIA for this amazing presentation. And let's hear again from Ethan. And once more about how AI will reshape work, of course, if we let it. So it starts with that productivity gain at the individual level, but if they're not sharing how they're gaining productivity, no one else can learn from it. So you have a couple people who are getting huge advantages and no one else is. A second set of problems that end up happening is you build a very small product that's not very ambitious, and you spend a lot of effort doing it, and it's successful. But now you have to spend a lot of effort maintaining a system that often is older when newer AMI models can do much more off the bat, and you're not evolving with it. So what the real challenge ends up being is how do we start thinking about the organization differently? So the org chart that we have today was invented in 1855 to manage the railroads. It was designed so that you could pass information in real time up and down a train line and have the right people answer. The time clock at a subway line was invented in the 1910s by Henry Ford, agile development in the early 2000s. And we keep doing these same methods of organizing ourselves because until now, the only way you could apply intelligence and management to a project was with a human. So if I added people to a project, that's the only way I could add intelligence or management. But now we have a different way of adding a quantitative intelligence, right? I can actually use AI to add intelligence of all sorts, monitoring of all sorts. They have video and audio capabilities. We have to redesign organizations around this. One of the big issues I see with operational leaders is not getting that AI is aimed at them, that there are key people in this process, right? Either they tend to use an IT problem because it's made of code. But it turns out coders are actually quite bad at using AI because they don't understand the subject area and they want it to work like code, deterministically. Shared of AI doesn't work deterministically. It works statistically. So if you're actually good at thinking about statistical outcomes, you actually do a pretty good job of thinking about this. If you're good at managing people, you're often very good at working with AI. Then on top of all those pieces, right, the other thing that they get wrong about this is not thinking enough about the fact that they have to innovate now, right? So you actually need to think about how to use these things because the systems aren't perfect. But guess what operations people are used to doing? Dealing with imperfect systems and figuring out how to deal with those flaws. And I think the third big problem I'd say is somebody else, believe somebody else has solved this problem. Like if I just go to the right vendor and they'll have solved the problem for me, right? But instead it's going to be working with people who deeply understand your process, whether that's internally or with consultants who really know what you're doing. And starting to think about how do I innovate? Where is the opportunity here? Rather than waiting for someone else to invent it for you. The examples I have are mostly from, you know, consumer-facing organizations. But they're ones that have, you know, the same kind of problem, 10 or 12 different data sources. And what they do with the AI is they actually have the AI agents literally go and do lookups on each of these, like on its own, following curiosity, to figure out what information it needs. If it finds what it needs, it looks for, you know, it doesn't find what it needs, it looks for something else. Follow leads and actually combine that information together and give you the report you want that's actionable, right? Like you don't actually want data. What you want are insights. And now you have an insight engine that can generate insights for you. You should be using that a lot. So it's not really much of a problem if people are innovating on their job, right? Because they're already doing that. Like that's what people do all day is they're like, what if I do the data enter this way? Like experimenting on your job is cheap and what people do all the time anyway, right? So that's not really a problem because it shouldn't be distracting them. It should be helping them. And if it's distracting them a bit and there's a little bit of that effort, that's okay too. The question is how are you harnessing that? So somebody comes with a breakthrough idea. For each one of those, two other people are using AI and haven't quite figured it out. And another three aren't using it at all. So the question for you is how do you take the person, what does the person with a breakthrough idea do with that? Who do they talk to about? How do they show that to you? Where's the lab that they're presenting this information to? So the problem I have is not so much that there's not talent there or that you have to balance this out, it's so much as how are you even harnessing the innovations happening. And then at the leadership level, you have to start making choices. Like you can't just say whatever happens, happens. You have to say, okay, this looks like this has really good effects on procurement, right? We should absolutely be, everything we do, we should do a test on this. We should let this help us with negotiations for each procurement thing. We should let it actually, you know, we should talk to it about actual quality of our products coming through one way or another and have this dialogue back and forth with the AI and let's invest effort into making that work. So you're going to be looking at the innovation happening throughout the organization. You'll be using AI yourself to understand what's good or bad. And from that, you're going to get an intuition about where you want to really invest. That's going to require a reinvention, right? It's an operational, strategic, and leadership problem. Not a technology problem, ultimately. Let's just zoom out a little bit. If you ask the AI labs, right, they think they will achieve within the next five years, actually within the next three, AGI, a machine smarter than a human in every intellectual task, artificial general intelligence. They genuinely believe it, right? They believe it privately. They believe it publicly. You should believe that they believe it. You don't have to believe them, but they believe it, right? A lot of words of belief, but you get the idea, right? We don't know whether that's right or not, but it is absolutely what they're predicting. So I think it's worth, first of all, thinking in scenarios. Assume, you know, that we have machines that are very smart in five years. But I think what the labs underestimate is how real work gets done. There are real organizations. There are interdependencies. There are traditions. There's unwritten context. And, you know, there's humans, right? And organizations. And they're messy. And process is messy. So I think that what you're going to see is even as AI gets better, there'll be a set of firms that try and embrace it and change everything that happens. And there's going to be a set of firms that sit back and say, this is too complicated. Someone will solve this for us. Then we'll figure it out. And I think what's going to happen over the next five years is the companies that figure out how to become AI first and AI forward and build the capabilities internally to do this are going to start to pull away from the others. You suddenly have, like, what, you know, operational organization doesn't need more management, doesn't need more brains, doesn't need more code, doesn't need more analysis. And the companies that figure out how to apply it are going to get the advantage. And you're going to start to see a gap widen and that may not close. All right. Thank you so much for the cliffhanger, Ethan. Important takeaway here is that AI will not just change systems. It will change work and the way that the work gets orchestrated. It will demand leadership choices. And you can't wait for someone else to figure it out for you. You might find yourself lacking significantly behind competition or colleagues within your industry. So with that as the pivot before we move on to our next case, we'd like to ask for your input. So within the chat, you will find a poll. And please respond. Where does your organization struggle most when trying to scale AI in operations? You will have five choices. Executive ownership and prioritization. Lack of clear repeatable processes across sites. Data readiness and quality. Skills and roles to run AI at scale. Or technology architecture and integration. And we would like to invite you to pick the one that is the most significant blocker in your own context today. Consider whether the barrier is structural, technical, or even behavioral. Because this could give you great insight as to how to start this AI transformation of yours. We will share the results of this poll with Samad from Stenaline in a minute and get his perspective and advice for you as well. And that's probably a good segue to introduce Samad and Stenaline in our next case here today. Stenaline, as you know, it's a leading global ferry operator. In this case, they were facing a lot of challenges in the claims handling. 6,000 claims annually. Paying outs rising 18% per year. 1,000 claims pending. And plus 80 days to resolve accepted claims. Let's see how they used AI to tackle that and how they came out of it. 1,000 claims pending. Samad, welcome. It's a pleasure to have you with us here today. Thank you. And let's get going then and start with something that I think there's a little bit of a loaded question maybe. But, you know, when you start talking about AI introducing this technology in a very traditional industry, when we're talking about shipping and transportation and so on and so forth, that must not have been an easy ride or at least the start of a conversation. Can you put a few words on how that started and how that process went? Yeah. I mean, the idea was actually a deliberate experiment. So you're absolutely right about the traditional nature of the business. And suddenly when you start to think about very, very traditional legacy processes, you think, oh, there are too many resistance factors in this. But the objective around it was, let's try and take a very traditional legacy process. We have many of them. So let's try and not create preconditions for processes which we should select, which are maybe further along the line into the digitalization. And then let's try the best of what AI, as in sort of decision support and automation, can offer and see how we can disrupt and sort of take maybe a broader leap. So obviously objectives around the process matter, i.e. progress matters. But also there was an interesting element of learning as well. How far can you push it? Was then there a specific objective before you got going? Like was there a target to achieve that? Absolutely. I mean, you talked a little bit about some of the challenges we talked about. We had significant volumes of claims processes that we needed to handle. Payouts not necessarily in control under increasing significantly. Lots of in-process claims processes that, you know, the backlog sort of grows. So naturally the process in its sort of traditional sense, in its legacy sense, had a lot of clear things, KPIs that you could do. And then you could maybe look at a very long-term stretch around it. But you could also maybe think about some of the immediate things that could happen that would be early proof of whether this is working or not. So, yeah, I think that was important. But then let's focus first. I'm really curious to hear about, I'm a really numbers guy. So I'm really curious to hear about a bit of the impact that you got out of it. And then maybe we can focus on the how right after. So if we start with that. Yeah, the obvious part is you can look at rejections. So rejections, we reduce rejections by about 50% in the regions we've rolled it out. Payout. So we talked about payouts increasing by 18%. Now payouts are more in line with inflation and what the damage valuation is like. So consistency, transparency, what's in the backlog. The backlog is reduced significantly. That's in the previous world or in the traditional world, we weren't maybe looking at it from an overall perspective of how big or long the backlog is. And I think, well, albeit anecdotally, but the customers who've been exposed to this new process suddenly realize, oh, my claim was processed very quickly. Oh, this is good. It was a pleasant, positive surprise. So we're getting qualitative, subjective feedback as well. So which is obviously useful. And in our business, if the customer says they're happy and they're not that easily impressed, that's a good thing. Okay. Well, that sounds like a great solution. But how do you basically, what is that you did differently than it was before? What was the actual solution? You know, what did you do with this beautiful process? Yeah. So, I mean, as I said, a traditional process, there is an element of estimation or estimating what a damage is. And then there's an element of automation. There's an element of sort of decision support and automation of the process. So, obviously, for that also, we had quite a lot of regional variances. So, documentation of processes in a very traditional industry are not necessarily there. In some cases, there is no actual digital footprint of the process either. So, the idea was to use AI to maybe look at sort of claim value estimations and then use automation tools to try and focus on getting the chain working and making sure that there is a macro view of the process and dashboards to follow up how we are doing across the board. Didn't you get any resistance locally for many of them when you look at this? Very much so. I mean, you know, legacy processes have deep roots. Yeah. So, it's, I think it's, you know, and in some cases, no documentation, no digital footprint either, no data to assess and look at. So, yeah, there was a lot of resistance. And, in fact, I think one of the interesting things was that maybe with the power of the tech that we have today, AI and automation, maybe in the sort of attacking the problem, we spend more time in actually addressing that, which I think is the big nugget for us in terms of learning as well. Can you try to put a few more words there? I was going to ask exactly the same thing. No, no, it's okay. So curious about how you orchestrate that change when it's like full of so many variables and so many different sets of legacy processes. I think in this day and age, I think even people who are not so exposed or not so digitally savvy kind of see AI as this very powerful thing. So, the question around proving technology as such, which used to be the old paradigm, oh, come and prove your technology works and then we'll see. People have this sort of understanding that this thing is powerful. So, then we're able to sort of shorten the cycles in terms of what solution we put out there. Then a lot of the focus is building trust and I know a lot of your earlier guests talk about trust, but I think that's the whole game in terms of implementing these solutions. You're spending a lot of time trying to build trust and maybe there there's a lot of learning and a lot of excitement, but I think the barriers of just or the days of just proving whether technology works are, we are a bit beyond that, I think. Okay. Okay. And, okay, but then there is one thing here, right? Because now technology changes. We've just been seeing, you know, the velocity of how that happens. You chose a system. You chose a variety of systems probably. You have your architecture. You have a lot of technical solutions out there and you're basically deploying this at scale. How do you ensure that that's also sustainable in, you know, six months time or nine months time or whatever it is, that you're making the right technology choices? That's what I'm trying to say. Yeah. And I think this is where also the nature of the technology is very interesting, the learning element of technology. So when we say, well, we've chosen a system, well, actually the system needs to learn the assumptions, the model needs to learn, the process needs to learn, the data context. So in some cases we don't even have a data context or a process context. So you're kind of building process context, data context, and you're also maybe improving your model over time. It's not a reductionist. We start with this system and just implement it and roll it out. But I think there's that, there's this, there's this beauty of learning in the process. And we should extend that when we talk about, well, machine learning or tech, you know, AI developing, the process context, the data context, and the sort of organizations kind of learning across the path as well. So emphasis on that is interesting. Okay. Again, if I can just grab that, because there's a very interesting, I think, question just popping, popping up in the chat. So when, when you're integrating AI at, at scale and across many regions and many different, let's call it process maturity levels, for lack of a better word. The question here is, how are you sort of constructively managing the risk of layoffs by both upskilling? Do you make mobility choices internally for people to swap or are you sort of maintaining the status quo for the people you have working in the process? How does that work? Yeah. I mean, upskilling is a, is a, is a big question. I mean, for us, the reality is that we are going to physically move people and goods. People are, are, the game is obviously operational efficiency. And, and in that, how do we measure efficiency? What do we do in that? I think there are, in our, in our example, there are people who've never even seen a iPad or a dashboard in their physical daily work. They don't, you know, they're not seeing sort of AI photos of damaged trailers or anything. And suddenly you've got dashboards and things. So it's a, it's a, it's a big leap, but we also think that even if we're very traditional or industry, a lot of people are very exposed to in their personal lives to these, to these things. And I think that's important. Um, when it comes to, um, the process, the, the, the, the belief that we need to upskill people. Yes. In operations, maybe our focus is not like maybe the traditional R and D manufacturing companies or traditional software houses. But I think the necessity to upskill is as relevant, as important. It is also a question of survival, also a question of, and I think just a personal reflection on, um, use of tools internally at the company like Standardline. Every time people go away on holiday and come back, suddenly the use of internal, you know, these, uh, tools goes up. And the reason is that people are actually externally influenced. So there is also pressure externally to learn. And so our job is to just, uh, encourage that and, uh, obviously give people access to tools to learn and, and play, but also maybe make it concrete for how this can work for a very operational company like, uh, Stenaline. Is there a specific, you know, we, uh, we talk a lot about the learning organization. Is there a specific learning program that you guys are focusing on or is it kind of a learn on the job, learn as you go? Do you have a, you know, a structured approach to that? Um, we are traditionally very, very good at experimenting, but we experiment fast. And sort of like, because we, we're not a manufacturing company, so we don't have a long, long R&D cycle, but we're very good at experimenting. And, and, and I think we're maybe, maybe leaning a bit on that. Uh, I think sort of like in this case, you, you look at sort of the whole development cycle and you look at AI and learning systems, learning processes, learning. I think the methodology of how you develop systems and how you interact with systems, how you roll them out is also changing quite a bit from thinking about standardization at the start and locking that in from a reductionist perspective and trying to roll it out is not so good. When you learn, you build trust, you more, you tweak your algorithm, make sure the regional nuances are taken care of. So I think a lot of that process is almost like, it's almost like it's infusing into what we need to do. You just need to be conscious of it. And I think traditionally, um, we should sort of also, uh, maybe rely on and back on what we are good at. We are good at maybe quite a lot of experiments and just trying to move forward. And maybe that's useful. Uh, you just have to be aware that sort of as, as in Ethan's challenge, um, everything's up for, up for review and you need to open your mind. So I think that's important just to remind yourself. But I just want to, just would like to grab that a little bit. There's another question that came up in the chat, just coming up on how have, how have you learned and like all these things that you are actually setting out on a giant experiment kind of. Then there's the question here is, have you made any changes to the way you were organized? Like roles and responsibility or even line leadership has, has anything on the sort of organizational structural layers changed as a result? Yeah. So the fundamental thing is, you know, the, for an operational company like us, we are an asset heavy operational company. So our, our, we have long-term multi-decade assets like boats and vessels and, and ports. And then we have the here and now the operation. We need to sail the vessel in a safe way. So there's this big dichotomy of, of, but I think, uh, because we don't have that engineering or, you know, background, maybe perhaps people are not so data driven or, or, or tech savvy. So there is that risk that all people are, uh, we need to sort of have some sort of separate innovation engine to drive that. But I think in this day and age, what we need to do is just make sure that the tech teams. And I always like to say the, the cycle of business development is equivalent to digital development is equal to AI development. So business development in our business is actually all about AI development. So it's almost putting organizationally putting all those teams much closer to the business. So the claims team needs to know how are we going to learn if the IT department is separately running on a different schedule and, and we are not developing the process or data context. So proximity is important. So I just, you know, in that one as well, and it's only a 30 second thing, but, but there is something around the learning organization because everybody's fighting for talent. Right. And then Ethan said, listen, we don't have a lack of AI talent. You just have a lack of people who actually understand that they can do it themselves and you don't need a hundred thousand data engineers for every single company. What is your take on that? I agree. I think if you, if you do an executive, I always say to our executives, all our leaders, if someone gave you a piece of code today, you can do something with it. Actually, you couldn't do it maybe two, three years ago, but now you can do it even if you've never coded. There is a lot of power in that. So you've got to, the tools are out there. I think. Yes. Then let's go to the poll. Yeah. I was going to say, cause the last thing we just want to do is touch back on this poll. And there are two that are so close to each other. One is the data readiness and quality. And then funnily enough, actually this, this relates because it's the skills and roles to run AI at scale. Those are the two top scores at more than 50% combined that people in the audience are saying, this is what's tricky for them. What's your perspective on that? The data readiness. I know Matha's great example for us. Some of these processes, there is no digital footprint. There's no, there was no, not even a digital footprint on the. So, I mean, if we, I think that we've kind of also proven to ourselves that we can apply some of these things to, but I think the skill is to actually, as you say, yes, did garbage in and garbage out is true. But in this case where we don't have a data context, we can actually make sure that the learning process continues. So we build the data context that will help us in the future. Sorry. What was the other? Skills. Skills. Yeah. Yeah. I mean, brave new world. You've got to approach it with sort of like that learning curiosity. Some of it, as I said, maybe you have to dig deep into your innovation culture in your company and maybe build on that. Not sort of, it's not necessarily a question of abandoning. I think a lot of the skills for a really old shipping company like ours, our curiosity, our ability or our ability to take on risk is something people can bank on. But it's about how much you learn and how curious you are about what's out there. And finally, maybe just one remark. What have you learned in this process that you didn't know before that you said, okay, I really wish I would have known that before I embarked that? I think the biggest learning is to not be set about standardization. And the process now and the tools are so powerful that it allows you to maybe give time to those nuances. And it gives you time to build trust around systems, how you work with systems, how systems spit out decisions and how you do them. I think you have the time and space to actually use that. I think we need to think about that and see how do we build trust with these machine learning models and how do we let it thrive and let it maybe go past our own judgment. That's the interesting learning part. Fantastic. Thank you very much. Thank you so much. Thank you so much. Thank you. So, Chris, a lot of things, a lot of learnings, but unfortunately we're getting towards the end, right? Is the end? I'm sure a lot of people in the audience have a Teams call or two, perhaps waiting after we're done here. You think there's going to be full of these kind of discussions? I hope so. I really, really hope so. Yeah, one or two for the rest of the day will do just fine. So wrapping up here on some super interesting conversations, operational excellence 2026 onwards means making AI a part of everyday decisions. So now we've seen how organizations move forward. They combine platforms, governance, and really, really bright minds turn these ideas into insights and actions that generate impact. MESAS has showed us the impact of forecasting at scale. NVIDIA are demonstrating how you can turn a really strong strategic vision and ambition into a scalable enterprise capability. And Stainalign are proving that hyper automation will improve speed and consistency over time. So a couple of things on where you go from having been here with us for the past 90 minutes. Own AI in your area. Lead the change for your processes. Progress beats perfection. And you need to design for continuous upgrades. The next model will come along just fine. Don't worry about that. Rethink workflows and not just tasks because the long-term impact will come from the end-to-end process and across your company. Sounds good. I'll try to remember those. Yeah, let's do. Thank you very much for joining in. We will share with you materials. We will invite you for follow-up events to go deeper into some of the cases. Have a great day out there. And thank you very much for joining us today. Thank you. Bye. Bye.