AIAW Podcast Episode 058 - Daniel Tidström

But then it's quite nice. My girlfriend has family in North Vescona and relatives also close by. So it's quite a good location and also quite close to the West Coast where I'm from. So is it that in more of the eastern part of Smoland or where is it? I might be confused here. Yeah, my geography is not in the eastern part. It's not close by the coast. So it's more mid-sweedom. Close to veckers or something. Yeah, I think it's southwest from Vekua. I might make a fool out of myself now. But it might be that way. But I think it's like one school. Yeah, it was crazy. But it was an amazing opportunity. It's a nice house with a called Kokilungnara and a lot of the character. A big building as well. Yeah, quite big. 310 square meters I think. Like 20 rooms or something? No, no. I think it's eight rooms. But then they have a big school building, a school room where they had a classes. So it's quite big. I think that's like 60 square meters or so. And then on the up floor they had the gymnastics. So it's like one big room, nine to six square meters. Yeah, really. So we plan to have a ping-pong table and place to hang out for the kids. But that sounds really cool. But that's a little bit bigger than the summer. Some invitation for the big house. Yeah, for sure. And I'm from small and as well. That's how you are. But how can you put a school building? It's very strange to me. What did you do to it? I think it's a cool idea to fix it up. It has a lot of character and a lot of potential. A lot of things to do, of course. But still in good shape. So we don't need to fix it up from scratch. I'm going to rent it out in the park or what's going to use it for yourself. I mean, that's the main idea. And I think also like long term, it would be good to have first of all a place where all children can have their own rooms. Potential also when they sort of get some partners or get kids on their own to be like a place to gather for a long time. So looking at the prices closer to Stockholm is quite you don't get that much house for for the price we paid. So awesome. And you haven't bought it yet or you don't we have paid the down payment. Right. So we have sort of committed to it. Any date when you can move in potentially? Yeah. 25th of March. Really? So quite soon. So we're sort of planning on what to buy first. Like clean it up and buy beds. Are you going to have like a housewarming party or something? And I'm invited. I haven't planned that much yet, but you will be definitely invited. I think we'll have more sort of open door policy. Everyone who wants to come by is welcome. You need to have some big computer like cluster room or something, right? It's like several hall or something, right? I have planned for an office slash music studio in one of the rooms. So that's planned music with what like pianos or guitars or what's your one piano was included in the house and also one of those school organs is also there. So we got a lot of old stuff with the building. Yeah. So it will probably be piano in that room, but then also I mean I'm doing some music production electronic music in my spare time. So that's a good place for I can actually put my spit carries on. Yeah. Sounds awesome. Try out things. You have to extend an invitation especially during the summertime. If you get some nice weather during the vacation or something and come down to small and my old home area, so to speak, that would be awesome. Where is small and where you're from? I Kalmar, Oska Shams, or Eastern part of our small land. Yeah. It's not that far. You usually have to go by train through Veku or something anyway. So yeah. So we got welcome here, Don Hiltistrum. Thanks. We know each other for quite some time I think. Is it like it's seven years now or something? Yeah, I think so. First date innovation summit. Right. You're one of the original people from being first in the inaugural. I like the originals. Yeah, the originals. It gets like 10 shares for some videos. Yeah. And of course one of the experts in AI and data science in Sweden. So it's a true pleasure to have you here. And yeah, let's start with that. How will you describe yourself? You know, we're stone Hiltistrum. I'm now a days a pretty old ex-Gothamberg person. You've had a lot of them. You can't hear that though. No, I think it's unfortunately sort of worn off. I haven't lived in Gothamberg since 94 I think. Okay. Then I started a young shopping. So also small-land connection and then moved up here. But I would say I'm very curious person, serial nerd. I always have some of the people that that mean. I get like passionate, interested in new things. I just want to sort of dive deep. But then sort of my my patience probably runs out when I get to certain depth of it. And then I'd rather sort of expand horizontally. So I'm probably not the best at anything, but I've a fairly broad perspective on things. And a bit humble as well, I think. I don't know. Cool. But if we move back a bit from the beginning and what's your field of study to begin with? Business administration. Right. So it wasn't the technical. No, it just a bit odd. What is really business administration? Can you just elaborate on a bit more what that means? For a technical nerd like myself? Yeah, I can try. I mean, I studied natural science at Yunosiet. But at that time, I couldn't really connect like the use cases for all the maths and physics. And I was also retired of school. And sort of had quite an easy way in school up to that time. And then all of a sudden I had to study and I didn't have the energy for it. So I don't know if this is administration, which I think I don't know why really. But it's I think what you learn is it's basically two sides of it. I studied international business administration. So more like our organizations and companies run and how do the markets where they operate on like behave. So it's both like economics, micro micro but primarily like how do you set up an organization? How do you set up business control systems like APIs? But then we had a lot of leadership and entrepreneurship in in young shopping. So that was quite quite fun as well. So every course block we had a company that we were attached to. So we could do like practical work at the same time studying, which I found quite interesting because then you can see not only learned the theory, but also try it out in practice, which is I think also a trait that I have. I want to see or understand like how can I use this? Yeah. Excellent. That's a sense. Okay, cool. But then after your studies, what was the next step? That was management journey at Scania. Scania, right? So that was I mean, there is sort of a red thread. And if you look at my server, you don't see it clearly. But I was I was interested in data numbers already back then. But then I was totally naive and thought that okay, if I go to a professional company like Scania, that is the best in its business, more or less, they probably have a lot of data available and I can sort of work with controlling and take decisions and like understand that thing. But then it was like a brutal awakening when I actually started working and finding out that they didn't have all the numbers. And there were a lot of broken processes and things like that. But I started out in 99 where in the middle of the sort of IT boom, everyone thought I was stupid because everyone started working either as management consultant or with the tech company. But I felt at that time that I didn't really believe the hype at that time. It felt that it was a bubble coming up and fortunately, or sort of almost right in the end. But also I felt I had so much more to learn and going in a training program to really get like a deep understanding of how the company of that size runs and why was quite interesting. Was it some specific part of Scania? I had a lot of, I started out on industrial and marine engines, so like trying to look at how is the value increase of engines in production. and how do you do that accounting? So I was also trying to figure out how information flows between systems worked. And then I did a portfolio on R&D, portfolio analysis on what are the most important projects, and how do you manage them, and how do you work between product organization and the line organization? Like where do the handshakes take place? How do you organize that? What is the right setup to have both the line organization and the product organization? I think it's actually still a valid topic for today. It is. I think it applies quite a lot also to our domain. Yeah, where I mean, it takes different disciplines and different skill sets to deliver on data and AI. And a functional organization can somehow sometimes prevent that unless you sort of figure out a way to work across silos. But I don't know if there is any right answer, but what we did then, I think, I might remember wrong, but trying to basically move people from the line organization into the product organization and have clear guidelines on basically scoping and stuff like that. So it was-- But I guess another question could be more-- you still want to have people-- if you take more like a product organization, you want to build something potentially software or it could be hardware, I guess, in the scone-ass case as well. Then you can argue that you can have a line organization where the people report in some kind of line. But you can also have a different organization in how people are grouped according to product teams or something. Exactly. And Spotify, of course, had a way of doing this with squads and chapters and some kind of reversed matrix organizational shape in some way. Yeah, exactly. Do you have any best practices or favorite ideas or how to best set up? So you have both the-- I guess one way to phrase this is, you want the product teams potentially. And please disagree if you don't agree with it. But you want potentially the product teams to be cross-functional in terms of the need to have the competences necessary to build some product. And no single person can have all the competences. So you have to have a mix of these kind of skills to be able to build their product. But at the same time, you want to have people that can communicate with people of their own peers, in some way. Exactly. And that could be more of a functional kind of dimension of it. How do you manage that? How can you find both the cross-functional grouping of people while at the same time having like a functional way of communicating or grouping people? It's-- I would say-- I mean, it is a trade-off. But I think you can manage it. I think there is an evolving way based on maturity. I mean, normally it starts with centralized analytics and data-term. Normally, from what I've at least experienced, is that the data-term is split to going into tech. And then the analytics remains within product or finance or something. But then after a while, when this one accompany scale, it will be hard because you're depending on requirements. And to get good requirements, you have to rely on people actually understanding what the potential is in using data analysis. The domain in some way, right? Yeah. So I think that's a problem after a while. And then from what I've said, and what I've applied myself is going into hybrid mode. So you may be retained some central resources. But then, as you say, embed analyst and data scientists into terms. So I've done that in many places. Like, can be and epidemic-sound. Spotify as well, where I was as a consultant. Right. So I think we had a clear cross-functional terms with product managers that were actually trying to drive improvements on user experience and things like that. So then, of course, connecting at the level is through practices. So I've normally said that what needs to be done and why? That should be those decisions should be taken in the team that has all the context. But then, sort of, how and the who perspective, like hiring, I can probably do, well, the best from the professional discipline level. So given-- I mean, you work with a lot of companies, either as an employee or as a consultant as well. So you have a large, diverse experience from how companies are set up. What would be your recommendation, though, for someone that haven't really figured this out yet? Or I guess no one really has in a great way. But you mentioned something about the hybrid potential organization with some kind of centralized part and then some kind of embedded part as well. Is that your-- Yeah, I think so. I mean, if you at the Primax Sound, for example, I was hired to set up a new embedded organization called Product Analytics from the beginning. But then we still had a central team. But then we had different managers. We had some coordination issues because basically, a data-term at one manager, I had one and BI had one. And that is hard. So we tried to group that under one umbrella, in the end, to at least get the linements and so on. And I think it's also often why there is a trend that people-- our company's a point-see level, like shift data officers, so we shift analytics officers. Because after a while, when you get multiple teams and you get more strategic value, you need to have quite a holistic perspective while still retaining autonomy and local decision-making as much as possible. Right, because that's one of the core things, I guess. I mean, if you do have a cross-functional team that have some autonomy and can move quickly, and it's very quickly on the product, that I guess is one of the objectives you need to have. Definitely. And also, I think, at least from my side, as-- I mean, it doesn't make sense. When you reach a certain scale, when you have multiple stakeholders that you work with, then you have a team that is a certain size. If you, as a manager, become the central point, then you do triage everything and assign all works to all people. That is a bottleneck. That's true. Now it's exactly doesn't scale. So I think, eventually, you need to figure out how to solve scalability. And I think that is-- I think trend in many ways, also with data meshes and stuff like that is the same premise. How do you scale a monolith, basically? And you can't just do it by throwing more people onto the problem. You need to probably think, think it without side the box and figure out how to do it without sacrificing other things. So just to summarize a bit, I mean, you can think about, I guess, two extremes here. Either we have just a single line organization. Everyone, everything is going through some manager. And that will have, as you say, some bottleneck issues and scalability issues. But it will be super aligned, I guess, everything. But the speed will probably be bad. And autonomy will be really poor. And motivation probably will, for as well, if everything is going top-down in that way. And then I guess, please do disagree. The other extreme could be, if you have only embedded teams that don't get some critical mass, if you take data science or analytics, and you want to have some kind of more infrastructure support or some way of building system or deploying system or doing that. And if you just have a single person working in a product team without support for many kind of centralized team, potentially, that knows how to do things and can build these kind of supporting systems. They will never scale as well, right? Yeah. I totally agree. And I think that is also why I sort of tend to go with a hybrid model, maybe in the same organizational structure, because then you can look at-- for example, when is the right time to invest in structuring your data, and documenting your data, and making sure that lineage is available, and that you add logging of errors and changes and stuff like that? I mean, it happens when-- I mean, for one person, the cost of error is probably quite small. But if you extrapolate it on a big a team, it becomes substantial after a while. So I think it's also finding that balance, which I think also is one of the nice things I like with the whole sort of platform thinking in general. I mean, what do you mean with platform thinking? But I mean, after a while, most companies probably end up with a platform team in tech that doesn't deliver customer features, but deliverers productivity features for the engineers. - And then, when you start to get a lot of reporting systems, exactly. - Exactly. And I think it's the same for data, for example. When is it worthwhile to have a few more data engineers to actually improve the productivity of the data science team? I mean, when you start scaling up and running many AB tests, it's day or each week. You need infrastructure to do it. So I think it's the way we normally approach it all. also from when we give advice is to do it first, maybe the cheap way and find value. And when you see that you want to run it again, but the cost of achieving that value one more time is prohibitively sort of increasing, then you need to invest in making it more efficient. So I think it's iterating on that and trying to find some kind of balance. Sounds great. And if we just imagine some company out there and they now want to invest them in becoming much more data and AI driven in some way, then they don't have any any kind of budget restraints here. They just want to do it as quickly as they can. I guess you can think about two ways to do it, you start off with three ways to do it. But okay, you can start off by building embedded team support. You add a lot of people to different project teams that you have. And then later perhaps you start to see that you want to build these kind of supporting systems, you can scale them or you can do the opposite, which is to start with the centralized system. And then later start to speed over, so to speak, into some kind of embedded organization or hybrid organization. Do you have any preference there? Start with the centralised, to start with embedded? I think a bit based on mistakes I've made myself. I think it makes us to start with the centralized term first until you get some level of maturity and then scale out. But you don't have to make everything perfect. It's again finding some kind of balance. But I mean, if you have no idea about what are the core KPIs and core metrics that you're working with, you have no documentation on the data set up. It tends to be a bit chaotic. There is some governance that I think is best at least to have first. And I have made that mistake. Good to go too soon. And also, I mean, if you have, it's also, I think, leadership and like a previous satisfaction. If you go into, as a single data scientist, into a team that isn't very mature in how to work with data, it's easy to feel alienated in that set up as well. And that's a learning that's quite recent as well. So I'm probably becoming a bit wiser every time. Well, you work with Spotify. And so I can, especially from those years, remember those times, if you had a single data analyst or something working in a product team and you have a product owner that is just thinking about the next iteration. Yeah, exactly. I think it's very important to have, how should I frame this? And I mean, product teams that are accountable for business outcome rather than product features. Because if you are accountable for business outcomes, it becomes sort of self-preservation to the measured things and log things and learn things. If you're only measured on producing features, exactly. I mean, then running an A/B test will maybe add a week or so to the time to market. So I think it's very important to really know why you're building what you're building, what's the purpose, what's the reasoning, what are the hypothesis that you base it on and try to be really sort of true to yourself? I think that's a big, big gap. We can probably iterate back to that. Well, I'm very glad that you said that in that way. And you prefer to start in a centralized way. I think also that's the core thing to do. You need to have some critical mass to be able to do anything that you can later scale. And I tried not to buy as even anyway, but you show the same thinking that I have. So very happy for that. So don't have to start off with the fight. Yeah, well, that's great. I think that's actually a very important learning that a lot of companies haven't done that you need some kind of critical mass to get things working properly. Otherwise, it just will be prototypes and they will never really show off any kind of business outcome or positive outcome that you can measure. Okay, that was a bit of a rant on the line versus product organization. But it's an important topic, I think. I think so. I think it's one of the most critical ones. I think it's a source of a lot of problems today, especially on the data side. Cool. Well, that was Scania. And some other classes. What happened if it's Scania? Maybe I should sort of go through to just keep the red thread. What happened in Scania? I think after the training program, I got the because that's when I got into data really. I got an assignment to basically support the smaller markets of CDRIs with the financial systems support because they didn't have that. So they basically sort of reinvented the wheel every time it was a new CFO. So we looked at basically benchmark like how's the best way to structure the financial model based on the outputs that we wanted, like little guild and fiscal reporting, management reporting and group reporting made a good information model put an into the system that implemented that. So that was basically my job at that time going from country to country implementing that. After a while, we found out that we had a lot of good information in the system, but we couldn't get it out in a good way. So that's when my sort of data journey started. I just had like, we need to solve this because the value of the accounting is not realized because we can't get the data out in a good way. So that was when I got into actually just opening my laptop, finding a SQL database at that time, building some views to just translate column names and table names. Then we've found some problems, so we developed into data warehouse and then all upcube down, reporting layers. And I had to really use her interface to manage the thing because I was sort of every beginning of every year, I was hassled by all the companies that didn't support to sort of roll the years. So I had to learn coding to figure that out. So that was like my starting point to my data journey, really building a data warehouse before even sort of really having the notion of that it's called like that, or solving a problem. And what year was this approximately? This was training program study 99, then was going all the way through end of 2000. So I think this was like between 2001 to 2005. I guess data warehouse in wasn't the concept at that time. It sort of was, but it was quite rudimentary. It was like Microsoft stack that we had tools were called differently. Cognos was the coolest BI tool out there. So yeah, but some good learnings. It was a good school. Finance is really a good data warehouse in school because it's like a star skill, as it is. And transactions are immutable. You cannot change a transaction. You can only count return transactions. So it's like in event log in a modern system. So it has a lot of good. I thought about that. But that's true. I mean, the finance has a lot of similarities to the kind of OLAB kind of structure. Yeah, that's a good data structure. And also like write the head logging as well that it's immutable structures. Append only. So it was quite good. And then I moved on to actually try to to actually work with what I started. So I took a business control job. Found that I enjoyed the sort of more strategic things of that. Setting up like proper KPIs, measuring things. But also like still that that Scott Scott. Now it was marking a marsh industrial company selling marking equipment. Marking what does it's like this bar code and best before dates and laser engraving in the like absolute vote cap bottles for counter for for fitting things. So like a very hidden product. But we had all the major industrial firms in Sweden basically. So it was a lot of time trying to figure out. I mean, first the finances and then I did a post sales work. So I was off the sales manager trying to just sort of what happens off to you by an equipment. How do you how do you make it run? How do you finance it? How do you solve things if they break? So I was the first management job. But after that I had like the first big career crisis where I felt like okay I don't want to do finance. I don't want to work with after sales. I kind of like numbers and I kind of like data and IT and I'm always sort of I always end up in IT projects because I think at this I hope it's because I am fairly adept at that and I haven't used a way to learn. But then I started working in BI and I've sort of BIA analytics and I've stayed there since that was 2010. 2010. And that was right. Was that when you were at C-DIS well or was that later? That was a bit later actually. This was at X Open Systems, small BIA/performance management firm. Product company and consultancy. So this had a small basically from the beginning a small just Excel plugin that had a big benefit. So it basically was like you could write SQL queries from Excel to database connection and then you can could like pinpoint information where it goes into, for example, income statement or balance sheet. So it was quite effective solution. So yeah, I started there as a bit of consultant, did some solution architecture and pre sales. So kind of like having discussions with customers about solving problems. And then I ended up product manager for the company as well. Probably because the product manager when it quit, I was the person that was nagging him all the time about things, new things we needed. So what does the product manager do? Trying to set the strategic direction and market positioning of the products working with pricing. And then basically trying to figure out what the users need, somehow prioritize that, and describe it into features and try to deliver. Product features that they want to build in some way. Awesome. And perhaps we should skip ahead a bit here as well. Or it would be fun to move into a svenska spiel, et cetera. Or is it anything you want to mention before we move into that? Now maybe not. I think the year at Connect, as a management consultant was a bit of a sabbatical for trying to figure out what to do. So I can skip by that quite quickly. But that was when I was really getting into advanced analytics. And also I think the embryo of our company, I met one of the co-founders at Connect app where we basically tried to connect the sort of the management consulting part of the business with actually delivering solutions. Because I think we saw at that time that I mean, I had an interest in numbers and data from a long time. But when sort of everything started going digital and the mobiles were coming and social networks were coming, it just made sense. Like now we have so much data. Data is really the sort of the voice of the customer in many cases. So it just made perfect sense looking at sort of what the big tech companies were doing and how they were solving problems of dealing with that data. So I was like really, really interested in big data, started running Hadoop clustering at home and learning advanced analytics despite my finance background. While still at Connect or? Yeah. And for people that don't, I mean, Hadoop perhaps is not as prevalent as it was at those times. What is Hadoop and the why did you, what's the advantage you can get with having that type of cluster? I think there are multiple things. I remember a customer meeting where I actually just did a whiteboard exercise where I equal like traditional databases with one of those child toys with the lid on where you can put like different shaped boxes in. That's like a typical database. And if you run out of space, it's really hard to grow. You basically need to buy a bigger box. And also if you, oh, sorry. Also, if you get something that is not shaped according to what the lid allows, you can't store it. So hard to scale and very rigid. Exactly. So being able to just add nodes to cluster, to handle volume is one interesting aspect and also being able to to apply some schema, schema on read or at least be able to store files that are not always as you expect to receive them is powerful. And then I can argue now that maybe it was taken too far with data lakes without skill mass that it created a lot of problems. But that's of course one learning. So that is why I got interested in a Swedish game because they were one of the, I think earlier at least sort of main mainstream players on had up. I think I mean, Spotify and King obviously had it since a couple of years. But I think sort of Swedish game and klar now was sort of about the same stage. So I got the data architect position at Swedish spiel, which I found quite interesting, not that I'm like super interested in betting and gambling, but it's a lot of sort of famous brands are well known. They have huge customer base lots of data. Yeah, tons of data and also both digital data but also like points of sales data and casino data. It's like a very very interesting business and a lot of touch points and what should I say business areas. Can you mention some of the text that you were having at Swedish spiel? Yes, we had we had a had up solution later on going with Hortonworks, which was the distribution. We had, no, I forgot the name, we had one streaming solution because Swedish spiel was actually quite cool in that case. I don't think that realized that I didn't for sure because I thought it was a state-owned company, probably not that advanced, but they had since 2006, I think, basically an in-house Kafka solution that they built themselves. Yeah, so they stored an immutable event log from 2006. So when I went in there in 2014, we could just sort of load eight years of data immutable on event level into the system, which is quite cool. So, but that was built on C# and we needed to translate it. So I had to translate it, feeding into Kafka. Kafka loading it into, if I remember correctly now, both HDFS, which is a distributed file system. On a loop, yeah. And also into solar, actually, to index that. I think we actually asked why when we met first time, but that was to basically have a real time transaction index so we can search on, if we forget support ticket, we can search on an event or a customer and we get the real time data. And yes, we'll elaborate a bit more because you have so much knowledge about different open source tools as well. And Kafka, of course, is more of a message queue or how do you describe Kafka? So real time streaming, I don't know, event bus, maybe. But it, you can send events to it and you can subscribe for different topics based on that. Yeah, exactly. Publish, subscribe. Absolutely. I still will call it. Exactly. But also that you can retain the data as well so you can, you can consume it multiple times. And if a consumer goes down, the data sort of just builds up in Kafka. So you don't drop data. You have like a lot to wait, stream event. Exactly. So I'm not to Kafka expert by any means, but it's a great system. A bit complex to operate as the rest of the head up stack. But it's much, much simpler today. But the animation is solar as well. And I guess another alternative is elastic search as well. Yeah. Do you have any preference or thoughts about in a solar versus elastic search? Not really. I haven't sort of used that much. I have done some tests and I think elastic search is super user friendly to set up. I haven't worked that much in production, but we use that too. I think the, how do you describe what solar and elastic search does? Full text indexing or? Yeah, exactly. Indexing and like searching out fetching and video records really, really quickly. I think we had some, we used it to serve content to the websites. So basically everything that had like really low latency requirement, were basically in elastic or solar in that case. Actually, I made a mistake we had before we sounded to HDFS. Actually, the story is in HBase as well. Oh, HBase. I have a lot of memories of that as well. Like it's HBase and Cassandra or to like a similar kind of solutions that as well, right? Yeah, exactly. We had Cassandra, the main two outside the data stack. But I think HBase was part of the distribution. So that was the way we went. That was in place before I joined. So I still have some horrible memories of HBase when working with this price runner founders second endeavor, which was called test weeks. We had like, we were launched throughout the world in China and US and like 30 different countries and then at one point HBase just crashed. Yeah. And we could not recover it. Really? And it was a large number of nodes with so much data into it. And we were trying to hack the source code to try to get around the problems we had. We called every kind of expert, someone from Germany that goes to help us. We're sitting there like day and night for two days trying to solve it. And finally got it back, but it was horrific times. It's nice when it's working, but when it actually breaks, oh Jesus Christ trying to solve it. I didn't experience that luckily. And also we had that the good thing was that we had what was called, let me say. I remember another mind, but I mean we had a persistent data store in the back. So which was also a super nice thing because if we wanted to do a change that was complex, we could basically truncate our later platform and just rerun it. Since it was imutable, we didn't have any state to consider. It was like super nice. It was basically a set of the set of lines of code that we just needed to run. So if solo and less search is more for one like working with text and indexing that and finding stuff very quickly and in real time being able to retrieve things, how would you describe page based in Cassandra? What's their difference to perhaps perhaps a solo and elastic search? I mean the way we used, I was a bit wrong before, but we stored the data in each base, but then we had to retrieve it by key. So we couldn't really search for it. So that was what we used solar for. A key key, could you use the each base if you have the key? Yeah, exactly. If we had a key, but the key was just some, I might be off now, I have to talk to you. You won't get the son that tells us how to do that. But I think the key was not, it didn't have any business meaning. It was just a key with some salt to get the distribution of the data. So to figure out where it was, we needed to use solar to actually fetch the right things. I see. Interesting. Cool. And of course Cassandra is a big part of the Spotify text stack as well. I can imagine. But I mean it's super fast. If you know, if you know the key to the record that you want, it's like amazing. Yeah. And we can have a lot of coliums for a single key and that row can have families of coliums and so much more. So it's more of a like a no-sql kind of quick retrieval of key values in some way. Exactly. Awesome. Yeah, so many interesting stuff. We're growing a bit technical here because Henry is not here. So we can actually go a bit techy. So that's kind of fun. At least for me. With some limitations. Yeah. That's great. Okay. So, so what's the space and they had huge amount of data and what were like the main products or type of tooling that you were working with there? We had multiple things. It was quite exciting because it was in the early days. What we wanted to do, I mean, setting up the actual big data stack that was mainly in place when I joined. So what I tried to do was basically replace the current data warehouse structure that has had become outdated both from a technical perspective and also from a business perspective. It wasn't relevant anymore. It was like a new company putting the customers in center instead of like the stores and retail site. So I think basically my, I think it was like the fourth week at work. I had to do a presentation for it for the CEO telling them, telling him that we need to rebuild the whole thing. So that was quite impressed with my manager at that time, Christian, that actually allowed me to come with those conclusions. Yeah. But I think it was like a fairly strong case that that was the most effective solution. So then we tried to figure out like how to run that and we had some consultants in as well. So basically going with a data lake in HDFS with a hive data warehouse on top to try to sort of figure out where the data is. So we had like one transformation later. But then taking it to more data warehouse structure that was needed. We did it in many iterations. First we were looking at different ways of doing it. I was a big proponent of Spark at that time. But we ended up with here was this by the way, sorry, which year was this? I think 2014. That was. Yeah. It was. Yeah. And that's kind of early. Spark was just at the first version. Was it even one point or with that time? I don't remember. But I know what really made me hook into it. We tried a lot of things. We tried like cascading. We tried scrolling, which is a scholarly cell on top of cascading. So I had like a lot of communication with Twitter to try to make it work. But then we looked also. We had a lot of I don't remember all the tools and frameworks that was. That was the problem back then that there were so many frameworks and so many ways to solve the same problem. It wasn't really like a clear leader. But I hooked on to Spark when they released the data frame. Right. From RDDs back to. Exactly. It's a great. Because RDDs is like still. I'm lazy. I like to have like a sort of type data set where I can call columns by name. And it makes a similar to pandas in much more familiar and to more as well. And it had also like the brilliant function that I love the most. We would add dot to pandas function. You can actually like you can work, which is also what I presented on the data innovation summit. So we can basically put. We put a Jupyter tool on top so we can actually run from one interface code fetching data from Kafka, HBase, the data lake, Hive or external other data warehouses with one coding framework. And we can process the data on the cluster. So that was sparking and you could connect to all these kind of data sources. Exactly. Right. And when we had massaged the data because one early conclusion as well as you don't want to work with big data unless you have to. You want to get to small data as quickly as possible. And then we could take it offline with the two pandas with just one function. And then you can go with all the cycle learn and math plot label, whatever you want to do with the data. So that was pretty amazing. I think really cool at that time. Really quite obsolete now. And I'm not sure if Spotify, I was a big proponent for Spark as well at that time. But no one at Spotify really was actually. No, I mean, now it's a CEO. Yeah, exactly. And on Dataflow, which is Google's managed thing instead of Spark. Cool. So once again, a lot of different tooling here and that is more for I guess like big data processing in some way. So Spark, Scoding, Cascading, I guess Dataflow these days would be counted in the flint. Flint, I guess. Yeah. Have you ever tried flint that way? We did try it. We had, we used to flint a lot for the streaming, gestion from Kafka. I didn't throw on it myself. That was the other day 10 years that it's I actually switched after a while from the data term to the date Science Ten. I was found that that's where the sort of pain point moved. And also I was very curious to actually try out building things and crunching logs and trying to build some models. Cool. Okay. So, but can you give some examples? But did you try to build in the Svenska Spell? So you used a lot of tooling here for processing big data with HBase and Hive. I guess it's just trying to map an SQL query to some hard-to-job that can run. And then you have all those Spark and Scolidings and whatnot. But for what purpose? Mainly reporting purposes. We did some work to try to enable basically, and we had some models, like propensity models and cluster models that were running on SAS at that time. How would it describe SAS, by the way? I don't know. I'll probably just make a lot of enemies, I guess. But to me it felt like at that time I was a strong open source proponent that felt that this will be the future. SAS will not be the future Python or R will be what wins. So that is why we, because we had the situation as well at Svenska Spell where we had log in on my Fentallopumbling, I guess you know all about that. Unfortunately. So either we had to buy more advanced tooling on SAS, which would take a lot of pains and admin work and be quite expensive, or we could just deploy Jupiter, which we actually did just as a scan work basically from Monday to Thursday. We set up a full working solution with Jupiter, Jupiter hub, spawning, docking containers with local instances and attaching all the frameworks that we needed on it. So it was quite just as scan work with the DevOps in Gotland. So DevOps running on. Yeah, so I think basically I just wanted to show them that this is how fast we can run if we want to rather than sort of going into a nine month procurement cycle. then we did some things but it's Vi gjorde inte riktigt en spark jobb i produktionen, för jag har fått en ny ventjärs. Så vad var det med den här tulten som vi har för att rapportera? Hbäs och höjstav? Vi var från en stag, där vi hade en fjol, så vi var med i oracal-kognos- -och Microsoft-analysis services cubes. Vi hade nog en kognos, vi hade en fjol att få en fjol, eftersom vi hade en fjol. -Tablo. -Tablo har små. -Ja, det är en fjol. -Tablo är väldigt fast. -Ja, vi har inte gått till fjol, men vi gjorde oracal-kognos- -och det är en fjol att få en fjol. Vi har förkaldit det som vi hade upp och kallade det med höjstav. Det jobbade att det var en fjol att få en fjol att få en fjol. Och då var det inte en fjol att få en fjol. Det är det som vi började se på i Texas, för vi skulle- Så som gällen som ShoELon gör genomgohr Kingdom, det är sleevesvullent på Москв conscious, nowadays. Vid-Vara har en brukiga fall av på en tiro, främst att lada erfaren i bäst som behöverоворstern tips på att limitations och på falla på. Det där går inte in till det. Men jag vill märka fick i, för exempel är det en som som fokuserar. Det finns andra som vi måste fära ut. Vi var också att vi hade en responsable gambling side. Så som om man är tillbaka i gambling, så du vill se det så. Ja, exakt. Och då var det väldigt mycket business reporting. Jag tror den första fallet var vägastmasjens, som var också mycket av det här. Det var en trönskansk, som bästet på alla märken var strömd i en väg, en väd i en kaffek. Så det var en biljön av eventen som var en par satt och en report. En vägastmasjens, så man är ju som en bandigt omgörna. Det är ju att de har stämt med det till långrådning. Eller som de kommer att kläna. Jag har inte det. Jag vet inte, inte även, men kanske. Jag tror att det är väldigt bra att du kan spänna en par satt och nog tillbaka det i en mån. Ja. Okej, är det väldigt intressant? Det är mycket av tekniklösningen. Vi miser mig om det, jag har en av du. Jag har inte gått till att öka det med oss med en. Det var en bäst, så vi skulle spälla. Då är vi två och en annan. Det var en liten uppdämpande. Vad var det här? Det var min 2-karrierkrisis. Jag skulle trycka att få ut. Jag vill gå till den här satt och sätta det på. Jag tror att det var väldigt fann. Jag love det typ av jobbet, men i samma gång jag har att jag inte har det med en matematikl. Backgrounden skulle vara så jäkligt. Det var en bit av pain. Jag hade spännande många timetöjringar som så här och så. Linnar Algebra och så här. Det är ju att vi skulle röra dokumenter. Jag förstår på vad som är en förkanskare för att följa om en av de modeller som har varit. Intuasion förhörsade. Och också, det var en mannatchment, jag är ganska generik. Det har en tekniklösning, men också en av de sammanhangsvisning, koncepts och hur valier har varit. Det är ju en av de här är en av de här som är en av de här. Jag har jag inte gjort det för att jag är jämt. Men jag har också en konferent i vad jag inte vet. Det har jag faktiskt varit på jobb på ett sätt. Jag har varit på ett sätt som är på ett sätt som är på ett sätt. Så jag var tråd att det var ju en av de 6-7 interviews med King med en bra person. Det var ju en av de här komputerna som jag tänkte. Så det var så bra. Men då var jag en av de här. Nej, det var inte bra. Men som andra väljerna och matta. Så, förväntligen har jag varit på staden som var lite välkomstats, som de var ju helt rätt. Och då var jag bara fråga om det skulle vara en av de här komputerna som är en av de här komputerna. Så det var ju inte minst en av de här direktynna, och sådana, men jag var ju en av de här komputerna som är en av de här komputerna. Men då var jag en av de här komputerna som är en av de här komputerna. Så det är det där sådana, jag älskar det här. Jag tror det var en rätt fråga. Ja, ja, ja. Ja, ja, ja, ja. Och du var ju stådde väldigt glömt på våra pelter i en office, sådana när vi. Okej. Jag tror att du var på ett flor, sådana, sådana. - Det kan vara ju. - Ja. - Att valgen gott om. - Det är ett cool valgen gott om. - Det var ju ett flera. Jag tror att det var ju en av de här komputerna som är. - Ja, ja. Okej, så där var det som var lite. Okej, cool. - Ja, så det var. Det var nöjborste. Jag tror att när vi medde, Basically, en office var nära till Alofredik i en. - Ja. - Ja, så där är. - I en filterion är det exakt det här i en Alofredik. - Ja, det är väldigt klokt. - Ja, det är så. - Just en flora bello, rola bov. Ja, ja. - Ja. - Och så, kan vi börja kanske få det här? - Kan vi. De är att börja sätta en sportsbok, som är som. - Ja. - Det är ju en av alla sportsbättingen som är inkluderande, utgörna intervjärns. - Över till eventen, och översa, och alla bakgränskomponenser, som som riskmanagement, och så. - Så vad är det att det är att det är det som är som är som för att veta? - Det är en bästa operatör. Det är ju. Det starting point av Cambie var att det var ett par av juni-bätts först. Juni-bätts så att det ska vara mycket svårt att kompeten med B365, som har så mycket revenue. Så det är som vi har foundat i en ordentlig kompeten i sportsbättingen, vi måste ha en kust. Det är väldigt spännande och väldigt labor och kompetent att kolla med sportsbätting. Så det är ju en kambinstead. Juni-bätts först, det är ju en kust. Det var en annan annan. - Det är en annan. - Det är en annan. Autogia, för exempel, är det en av de mer risker, och det är inte mer risker, men de har ju tagit på en plattform. Sweden var det regulerat i en vettningarket. - Okej, men vad var det som har man ju inte förutvisat? - Ja. Inre ännu, det var ju fulle-manage, som vi skulle se med i alla kallor. Men också med de customiseringen och också SD-käs och ABIs. Vi kan ju ju liksom se hur mycket vi vill customisera om det är att det är fulle-käs, så det är ju en kall, eller om vi vill kasta det i alla kallor. - Men du har ju också hos det i service, eller kan du faktiskt ta en öppensorsfasjon och vi är SD-käs och rannit och hos det. - Ja, jag tror att det är ett bild från den komponenser, jag tror att det är hos det där. Men det baken och infrastruktur var centralt. - Ja, okej. - Och du rar på en kallor, eller var det. - En prem. - En prem. - Vad är det du tycker? - Kompliades en reglatorie-risesens. - Allright. Jag vet att jag har börjat gå till en kallor, som jag tror att det är för det data och analytics-stak. Jag har inte följde att jag hade lansk med våra data-arkitets förra månnspäck. - Det är intressant. Okej, ku. Och kanske du kan också säga att det text-staket kan vara med att göra det så. Jag tror att det har varit mycket rönt. Ja, det var mycket rönt, men i minst var det mycket poäng. Vi hade en data-initiative som vi kallade med Kafka och Hadoop. Jag inte ser det där, men det var plan. Men i det här var vi som har som en data-house som vi kallade med Vertica. Jag kallade en kantaho-interfis-att-tidigt som vi kallade. Det var lite slow, inte en liten skäl. Så vi kallade en liten chang. - Det var inte så mycket rönt. - Jag var inte så mycket rönt. Men jag tror att det är nog en vertiga i den kvartiga. Jag kallade en liten rönt, och det var en väldigt performant. Men det är en problem som jag skulle nog minna till en annan kall. Jag var sådana på en annan lytx-tidigt. Och vad gjorde en annan lytx-tidigt? Vi gjorde mycket av det. Vi hade inte så mycket som vi kallade på stannionale data-tidigt. Vi hade mycket av data-analytics för det var mycket som vi hade centralitat. Vi hade en annan lytx-tidigt, vi hade analyset som vi har deployed i det här många partier. Vi hade analysen som vi hade med en annan lytx-tidigt, en klant behaviour. Vi hade en annan london som vi hade med en annan london som vi hade med en annan lont. Vi hade en annan lont, en annan lont, en annan lont. Vi hade mycket av det här och det var mycket som vi hade med en annan lont. Det är en superdata-driven-bist. Det är också en stor skol. Det är en stor skol. Det är en stor skol. Det är en stor skol. Det är en stor skol. Vi hade mycket av det här och det är en stor skol. Det är en stor skol. Det är en stor skol. vi gjorde inte så misstone och bör nånontin, där vi sono wurden allt blank att få ur soi. ppt oss allaobrikets sk Ravensen och t 버�ridir, Brand slaffning men inget inte för det s Mayh ut 상 Usen. En A slow je SS 8 I fjolvar. Jag förbi Circular T juare 노력 offsetjedom, så ett. Det är en lort av fokus om trying to capture things as they happen, but also understanding them afterwards. And learning from them. So a lot of the things we found, we could fit back to like the quantum and give feedback to different traders and it's like that. Cool. Awesome. And yeah, shall we move to Newden perhaps? What made you use to take the jump to Newden? I. It was a couple of things. It was a sort of an interesting business to go into. I think I've never been particularly found of a fond of finance as a domain. I'm not really interested. And I find it also bit like sports betting that you sort of have to be a bit of a nerd to get the value out of it, you sort of served with a million options. And then you have to figure out what's the right way to to approach it. So I think it was they were brought from by Nordic capital and taken off the stock market to try to rebuild the company, focusing on data AI and user experience. So it was quite quite an interesting setup where it was really going into scaling the analytics and its science team really quickly and try to get value out and support it, support the business with creating new products, like RoboSave, an algorithmic sort of RoboSave. So that's trying to is it for stocks or for funds or for funds? It's basically an auto trading fund like like a Vansar as well. Yeah. Also, I'm trying to see the time is flying away here a bit as well. So I'm trying to move a bit quickly ahead and anything that you like to share, you know, or some main highlights from your time in Norden that you'd like to share. I don't know. I think it was. It was it was the challenging. I think we had. I don't know if we had any. We did some recommendation systems, but then it was a bit sort of trying to find the path of how to build things. So what I think happened was some in first I sort of left for like personal reasons, but it was we delivered some recommendation systems we started to go with some sort of data driven serum campaigns that actually creates some values. But then it was realization that we had to like rebuild the architecture and prioritize certain things. So it was. I think the right decision for the company, but it sort of postponed the data journey a bit. Awesome. And then you move into another very interesting company as well, epidemic sound. Yes, right. Can you just quickly elaborate what does epidemic sound do? Epidemic sound provides and sells access to. Royal to Fred music, more or less. So basically, if you want to use music for something, they have the tool for it. And I think the premise was parts of the founders were coming from the TV industry, where they found that producing TV shows like soul, the done, I think is the creation from of those guys. It's easy until you come to putting sound and music on top of it, because then you get to deal with a lot of lawyers to have to pay extensive amounts of money to do it. And basically big hassle. So that was like the problem definition from that perspective. The other was that it's sort of music music industry income is not like fairly distributed. It's very centralized around the big artists from the record labels and somehow diluted by a lot of collecting agencies and Royal to holders and lawyers and stuff. So the other founders came from the music industry where they wanted to allow for a way for musicians to make a living living from their music. So they started that and basically the whole company is like more of sort of a platform company where it's curating somehow like Netflix. So they buy music from musicians. They buy also the license pay for it up front. And then they sell the licenses to to TV broadcasters. It's one of the biggest tech provider of audio for YouTube videos as well. If I yeah, I heard someone mentioning internally that around 20% of YouTube is epidemic sound music, which is enormous. It's enormous. And I think it is one of the biggest them. Yeah, I think it is the biggest at least for, yeah, for that type of quality music. Did you do that because you have a personal interest in music as well yourself or partly, partly it's music is a very positive thing. It's like coming from the betting industry where people have a lot of mixed mixed opinions and feelings. I mean, nobody dislikes music. So it's like a very positive thing to work with. And also I found it really an interesting company to go into the business model is amazing. It was scaling extremely fast. It was a chance to build up a new team from scratch. And basically apply sort of all learnings on the blank slate and really build it up. So super fun. Yeah, can I mention? And yeah, scaling fast, I mean, it became the latest unicorn last year, I think. Oh, they are a unicorn now. Yeah. So one billion dollars. Yes. And there are how many employees do you think they have right now? I think right? I think four or five hundred now. I think when I joined was like one 50. So we sort of out grew our office had to move to another one. Still in Södermalm or yeah, yeah, thought pevron right now. Nice. Södermalm is nice. It is. Awesome. And then you took a big leap suddenly and what you created your own company right? Or was this something that already existed when you moved to date age? It already existed. I was sort of invited to join quite early, but then I felt that epidemic sound would be a good experience to do. But then I joined in October last year, which is, I don't know, I mean, I wanted to do something something different going into another head of analytics job felt like not that challenging or I mean, of course, it's challenging, but not in the same way that I wanted. So having my own company or sort of being a bit more independent and trying those wings, I've been sort of a dream since I was kid, I always said that I would have my company when I get old, I guess I'm old. So it was time. We all still 27, I think. mentally. Yes, mentally. Okay, so Okej, quickly describe what this date is and what this their specialty, potentially. - Yes. We have, I think we're based on our situation awareness was basically, I mean, first of all, there is a lot of value in working with data. It's proven over and over in many places. So that's the first one. The second one is that we see also supported by a lot of academic studies and so on. That a lot of companies are investing heavily into data analytics and AI. However, quite few of companies say that they are successfully becoming data driven or establishing a data driven culture. So there is a bit of a gap there. And that is what we try to focus on. And sort of our experience and my experience has been that there is a gap in the market. There are a lot of management consultancies that, I mean, some of them are quite good at this, but it's not really the main core business. And on the other hand, there is a lot of companies providing tech services like data or analytics. But there is a gap between those two from, which was also the realization that connect that we had management consulting them delivering power points. And then it was up to the customer to figure out how do we make this happen for real. So that is basically what we want to do to have part of the company. I mean, there is a strong management consulting background to really understand the business sense of things. - Indicator, I mean. - Yeah. But we also have really strong competencies in analytics and data science and in data as well. So we think that with those abilities to have the data in place have the understanding and the analytics in place, but also figuring out how to act on it and how to actually create value from it. That is what we try to do. But it's not the product company, it's mainly a consultancy company or. - Pure consultancy right now. I mean, we will say it's something that is on the table maybe, but also I think we try to right now be very agnostic from platforms so we don't have any technical partnerships or anything like that to just be able to say with like high integrity what is best for the client right now. - How many people are working at the data right now? - Right now we're 10 and we have one person coming in. So 11 right now. - Cool. - So we were growing from four to now from five to 10 last year and hopefully we can continue growing. This year as well. And I know you've been working with a lot of very well-known companies as well at that time already and like Spotify and Vault, et cetera. Anything that you. We shouldn't go through everything, but is it some part of those that you would like to focus on and give some example of what you have worked with while at the data edge? - I mean, that's one benefit of being a consultant as well. You get to try out different things. Your diversity and breadth of knowledge there, I think is for few people have the same kind of understanding in a more holistic sense, I think. - No, but I think that's of course what I say is my strength as well. Then on the individual parts, there are experts that are much better, but I have a fairly good holistic perspective. So I mean, I've been doing everything from head of data for trusty, then head of product insights for Spotify and SoundTrap, the SoundTrap team. And now head of analytics for Fish Brain, which is also with my former boss from Epidemic who's going in there. - Oh yeah, okay. - And more just strategic advisory for Vault. One of the companies in Iqt Ventures portfolio. So it's quite fun to be able to work on like different levels from very strategic matters to answer on things, do sounds. - If you're still worth to try to mention some examples, some highlights, something perhaps that made you a bit surprised when you started working with some company of those that you worked with, anything you can think of, this was a bit surprising, from a positive or negative point of view. I would say one realization that I've made, at this particular, I mean, quite a red thread last few years is that I've been working with scale ups quite a lot and I think it's quite interesting how similar the maturity journeys are. Most companies follow the same pattern pretty much from the startup phase to the scale up phase with more mature phase and sort of becoming data-driven. I find that quite interesting and somehow rewarding, even though the business models can differ and environments can differ, it's surprisingly similar. - Okay, but let's try to elaborate a bit more on that. So give them your experience in a number of companies. What is that journey like? How do you become more data-driven? Can you just give some kind of, yes, best practices in how to take that journey in the right way? - Yes, I can give it a try. But I think, I mean, starting out, I would say, I mean, as we touched on the beginning of the talk, like having a centralized setup where you focus on getting the basics in place, just getting the data in from the right sources, sort of trying to get an understanding of what actually matters and what's important. Basically from getting the basic reports in place to understanding your metrics and defining the outcome studies relevant. Then I think the first thing splitting up data into tech and giving BDI or analytics, going later into a hybrid model when, basically when you get too much of a disconnect between the central team, and you start losing context of what's happening in the business. And I think it's like, as you mentioned yourself, like there are two ways, like two generic ways to approach scaling problem, either you can apply it like command and control. So you take more decisions top down. Or you need to-- - You say command and control, I'm thinking about completely different things these days, but okay, please, come up. - Yeah, maybe we should chip it at that. - It's okay. (laughing) No, but I think it's not a scalable way. It's a way to force alignment, but it probably doesn't work also, I guess, in real life. So finding a way to align with having clear understanding of where we're heading and how to align the teams. And I think that's normal. I think that both the BDI and we can Spotify that became quite clear. I mean, we had a clear idea about what was important. What is the most important metric for the-- - So the understanding of being day driven and the values for doing so was clear, you would say for those companies as well? - I think it became clear. It's of course an ongoing work. But understanding what are the outcomes that you're aiming at, I think is really important. And I think that's a bit of a divider, I think which is also supporting literature that there is often a gap between the top management of a company that wants to have some kind of impact often measured in financial terms. - Yeah. - And the rest of the company is managed on, like resources, people or money, activities or outputs. And there is a disconnect between producing something and creating value. And I think there is somehow an assumption that as long as we build things, we create value. And I think that is not necessarily true. And I think a lot of support also from literature states that maybe 60% of what a product them does is waste. So assuming value is very dangerous, I think. So I think it's good to be informed, take informed bets because you need to deal with uncertainty. You cannot go away from uncertainty because then we have to do nothing. - Yeah. - But also trying to prove, I mean, even if you take a decision based on assumption, you can still measure the impact in clear terms and try to connect that. - Well, then becoming data driven means so many different things, I think. And some people think data driven just means they be I, like trying to have the proper understanding of the business itself to make some decision on what to continue with. But it can also be more product driven, I guess. So you actually start using it for the products you're building or it can be for the manufacturing, if you have that or the marketing or the sales and whatnot. What do you think the right way to do this? If we speak about the data journey in general, should one, is it correct to start with a BI journey to make sure that you first have like a data driven way to understand how the business is going? Or should you start also with the products or with the sales and marketing or manufacturing or what not logistics and so many things? I think that, I mean, you can't do both, but you cannot, you can do both in the sense that you don't have to like start with BI for the whole company. You can, in a vertical go from nothing to maturing quite quickly. The rest of the company can still stay behind. I think that's valid. But I don't think you can start with, and I think, I mean, it's a big problem. I mean, a machine learning model or like an AI solution often tries to optimize towards some kind of parameter. If that parameter is not aligned with the outcomes that you want, you risk sort of automating yourself into the wrong direction. So I think having a good notion of what are the key success factors and how do you measure them is very important to be able to go advanced. Right. So, I mean, that's, I don't know what your perspective is. I mean, of course, there are some more AI things that solve other problems. I mean, something that we speak a bit about here is the so-called analytical ladder, which means that some people claim that, you know, you need to do the full BI journey before you even move into more advanced analytics or you start using data or AI for some product. And potentially that can be a bit dangerous if you need to take, you know, if you wait with using data and AI for other things than AI until you have, you know, everything perfectly working for the BI reporting kind of things. That's correct. And, you know, moving stuff centrally, if you take some kind of logging of the data that you have from whatever product or from the sales or from whatever part of the organization, it's good, of course, but it is actually quite a lot of work to get, like, centralized, like pipelines of all the data to have the proper KPIs that you mentioned for the business reporting purpose. And sometimes to include something in a product can be completely separate from that kind of centralized BI reporting pipeline, you know, even, you know, not even the same kind of logging or data that you're using. The one example that we, I know we spoke about from Peltore was this kind of manufacturing company and they had the product team for some kind of machine that did some floor grinding kind of things. And they just want to think for predictive maintenance kind of purposes, you know, we want to understand when it's breaking and you can as a domain expert understand very quickly that when I just hear the machine sounding a bit strange, you know, as a human, as an expert that now it's time to do some service, otherwise it will break down. And if you think about that and you can simply put a mic on it and click some data and do some annotation and build a system or some AI model on that and you can have some predictive maintenance, you know, rather quickly. And that has nothing to do with the BI like pipelines, right? So I think one, at least that's one of the thing that we've spoken a bit about, you know, the BI is super useful of course, but don't mistake that you have to do the full BI journey at least to start using data and AI for other things. No, you agree? I agree. I think the results, I mean, I tend to read quite a lot. One thing that I think was like 2014 or something that an MIT study basically like plotted like two journeys to being data driven, that one that they called the generalized path, which is very much starting with the BI layer, adding some diagnostics on top, maybe some predictive and then sort of going up the ladder. And a specialized path where you can go in verticals and go where deep while some other things might be lagging behind because it can or because you just have to run fast. So we applied that when we set the strategy at Svenska Spiel where we said we found that some areas we need to be able to go more advanced. For example, marketing was way ahead of many of the other functions because they were higher maturity and more business value to be derived. So I very much think so as well that there is not like one way of doing it. But I think it's you need to be quite clear with that the model you apply, do you have the context for it and do you know that what you're optimizing for is actually a meaningful metric. Right. Yeah, optimizing for the wrong metric, you know, that can be super dangerous. And I think it happens quite a lot. And it's high risk even if you do a lot of BI and analytics because it's hard and the metric to optimize for is moving around as well because it's not a stable thing over time. Yeah. And also remember some discussions that, you know, if you do an A/B test that you've done so much during years as well. But sometimes, you know, the data you use for doing that A/B test is perhaps flawed. It could be noisy. It could be that actually measure the wrong thing for the A/B test. And one term or sometimes we have used this, you know, data only versus data first. And what we meant with that was, you know, if your data only, you only trusted data. Yeah. So if the A/B test says something, you don't even question it. You don't even think about, you know, is it because we are logging the wrong thing? Is some of the data missing? Is it noise in there? Is it some bug somewhere? Because in this, and you just say, data says that we don't care about anything else. That's potentially super dangerous then. So data first means, of course, you want to look at the data, but then you need to take the human into place to actually try to analyze, you know, what does this really mean? Yeah, definitely. Right. But I think at Canby, we were very deliberately call it data informed. Data is one point for the decision making. And there are many other data points that you need to consider, not only like quant data. I mean, at Spotify, we had in the product insights team, we had also a lot of user researchers. Or a lot, but we had user researchers as well to get also, because data will probably not give you the explanation of why a user acts in a certain way. It's only a sort of that it does and in to what extent. But the motivations or the actual problems you solve, I mean, don't forget to talk to the users. It's like we had a discussion in a customer meeting today that it seems that analog companies are a bit reluctant to sort of look at the data because they're used to talking to the users. And I think as a tech company, it might be also dangerous to just look at the data instead of actually meeting and talking to users. I think that's super important as well. So data driven can be very dangerous and can lead to like as we discussed, when we met last time as well with local optimizations. Exactly. I mean, I guess the balance is the solution to many so many things going high with both in terms of the organization, in terms of how you interpret or make decisions from data. Yeah. Right. And I think in general, just, you know, if you think AI versus machines, if you want to go that route, I think that the best thing is to have an hybrid there as well. You should be data driven, but perhaps not fully automated for at least some more advanced things. If it's super simple thing like a recommended system, which not always is simple, but at least that's something that could be potentially fully automated. But for so many other tasks, it's just, you know, one source of information that augments the understanding from the human. And that can, and then the human has to do what that's what they are good at. And if you combine the two in a hybrid way, that potentially is the best way to go. Yeah. Would you say so? Yeah, definitely. We talked about that quite a lot that can be where, I mean, you can of course, I mean, I mentioned in the model, probably give you some probabilities of it being right. I mean, how sure it is. I mean, you can let it decide up to a certain threshold and then you can add a human interface on top for the specific case, so the ones with most or least signal. I mean, it's, it's all about trying to, I think, be like focused on the situation you're at the context and the impact of things as well. I know we had a lot of talks in the management and soundtracks about type one and type two decisions. Okay. And I move to that. We have to continue that discussion. Okay. Cool. But okay, what do you mean with type one and type two decisions over time? I mean, it's Jeff Bezos, staying from Amazon, I think, from the beginning. Basically, is the, is the decision easy to revert if it fails? Or is it really permanent and so on? I mean, depending on if it's something that is very easy to change. I mean, if you do like change in interface, you run an ebit test, it doesn't work out, then you can just revert. It's like no consequence, more or less. But if you take like a big pricing change, for example, that might be much more difficult to revert, so you need to pay much more attention up front. So I think finding the balance of like, what are the consequences? And I mean, that also goes back to sort of product management and like Silicon Valley product group things that I conspired and those books from Articagan with. validating the product risks before you do something and taking informed decisions where I actually looked at the problems and understood them ferret intercontext and also the probability and consequence of something happening. I would say is the thing to do so what we have tried in many ways is to try to enforce that into business reviews. What is the data right now? What kind of conclusions do you draw from that and how does that change? Like the deep framework that you talked to Henry Glamigry and about. How do you update your world view based on the situation now and the data you have? And then looking at, based on that, how do you assume or how do you analyze and decide what you think is the right way to the right thing to do right now? And just let people know the deep framework is based with the data insights and beliefs. So you like refine the data from the raw data you have into some set of insights and the frames and beliefs to make it efficient. And the bets that last bit as well. And I think that was a big learning from sports betting and I'd sometimes sort of go back to that. I mean, if you, I mean, if I would bet something against you, if we were talking about AI, I wouldn't be very confident in my chances that I would win. So I would have a very low probability of winning. Meaning I probably shouldn't spend set a bet that is that high. But if we were talking about, I don't know, schools in Elmhulth, then maybe I could think that maybe I have an edge on that so I can go bet bigger. And I think that sort of valuing like the stakes you put in versus the probability of success. This game theoretic kind of thinking. Exactly. So I think that is very much what decision making boils down to and also I think you just need to void like analysis paralysis as well because you will never have all the facts you need to deal with uncertainty. I think being too risk a versus also a very dangerous thing, which is also something that sort of I think is the case also for companies just starting out. And then they don't want reports on everything and then they don't really know what's important. What do you think about like your Facebook comment like move fast and break things? And let me just perhaps put that in a more context. I mean, you can think about the old style kind of waterfall planning, thinking ahead like a couple of months before even start working with something. That is, you know, very safe, low risk. And you have considered all the contingencies that they can think of. But on the other side, it will be moving very slowly. On the other hand, if you move too fast without doing any kind of thinking, you get a lot of stuff out potentially in a product. But you will breaking things all the time and you have to revert all the things all the time. So the value will still be slow. So that's the value that you provide for for the company will be slow in both extremes. Like either you move too fast or too slow. What do you think about this? Is this basically this kind of game theoretic thinking, you know, thinking about in finding the right balance of taking risks sometimes to. I could change. I definitely think so. I'm probably tilting towards run fast and break things myself. I think that's how I'm geared. But also at the same time you need to value the risks and also the type one or type two. Can I revert? What happens if I'm wrong? And like, what is the consequence in relation to the likelihood somehow? But I think it's super important. And I think. I don't know. This is my pure theory. I think we are risk averse. I mean, we are risk averse as human beings. I mean, that's also the scientific fact. We rather not lose 100 pounds than winning 100 pounds. Losing pay sort of hurts more. That's like a sort of biological fact. But also I think your cultural trait is. I think we don't deal with failure very well in Sweden. It's my theory. And a lot of companies perhaps don't have this kind of forgiving kind of culture either. Yeah, maybe. But I think. I think that's definitely a case where we're sort of expected to be right all the time. But that's just. I mean, to be fair, it's a fallacy. It's just an imaginary world. Because we all make mistakes all the time. Right. It's just like how big are they? How do we handle it? Do we learn from it or do we try to hide it? Yeah. And I think having a culture where you focus on the problem rather than assigning blame, I think is super important. Because the realization when you start running a bit, for example, is that you get black on white. That most of the stuff you do is probably fail. Yeah. It doesn't add value. And that realization can be quite hard if you feel like, oh, that's hurting. Yeah. Because I thought I was right. But if you can. But if you can. But if you can. That you think this is going to be an improvement. Exactly. The ability of being right is 25%. But the upside is enormous. Then maybe it's worse, the risk of failing. So I think that is a critical part. I'm sort of very passionate in trying to get that as right as you can. I think sometimes you should even have a KPI. The person that admits to being wrong the most should have a higher salary. Yeah. It's really. Yes. I think you will have a future guest in Mikhail Shilkin here. He told me, he was a date scientist, worked with Atkambi. And he told me one thing that really stuck. That his job as a date scientist is to prove people wrongly. Because that's the valuable outcome of his work. Because if you're right, everyone will know about it. You will have built a future, it will be celebrated, it will be right. But if you do something wrong and hide it, it's very likely that it will happen again by someone else. So learning from failure and like a Spotify, I guess, is a celebrating failure a bit is. I think it's really, really. It's actually. It's really important just not to create a nice culture, but it actually adds a lot of value. Admitting noticing and learning from failure. That's value, but also I think it removes the whole. The scary part of admitting failure, which is so dangerous. Because then people start to fake success since that. Super dangerous. Success theatres is like really an anti-pattern for being later than I was saying. I think you'll need to put a very, very fat line between different. What is called work positions where you can be actually rewarded if you do wrong. So what if a fireman does something wrong? So maybe we should limit it to research. It's type 1 and type 2, right? It's hard to. It's innovation and research. Or a politician making a wrong move, which is. Or like a surgeon, you know, making a decision. Or a surgeon, right? It's kind of the most person that actually fails the most gets rewarded. In a research maybe, we're not in a research. But a politician in this conflict. I think even a surgeon can admit to a lot of mistakes. As long as not like type 2 kind of things or these things can't be revert. Right, right. So let's make a term for type 2. We should be rewarded, but not for the type 1. Which one is which, by the way? The type 2 is the permanent thing, right? The type 2 is the. I just Google it actually. It was like this. Type 2 decisions are life. Type 2 decisions are like walking through a door. If you don't like the decision, you can walk away. So it's type 1, the permanent thing. It's the permanent thing. And just sort of. It's also one of the benefits being involved in analytics, because here's sort of. It's part of your job to be very. fact oriented and just state things as they are. It's sometimes always not valued that. state things as they are. But I think it's really important. And me personally, if I hear someone say, "I was wrong." This acts so much credibility for me. I remember one person that's qualified called Adam Kava. "Ah, I'm working with him now." "You are?" "Yeah, I'm working with his company at Waltz." So I'm involved there. "Get in data." "Get in data, right? Of course." "I'm going to Warsaw next week to meet up with those guys." I mean, he made these kind of awesome talks. And he said, "Oh, did so many mistakes." "Oh, that was so wrong." "Oh, that was so horrible." And when a person goes on stage and say that, it adds so much credibility. I wish people understood that this is actually how you should speak and think. And if everyone were doing that, you would have a culture where I think people would be so much more productive in a good sense, so to speak. I actually listened to. I don't think of us Adam, but it was some colleagues from him when he was a Spotify. I think it was like a Duke Summit in Brussels, like 2012 or 2013. They were standing holding a. presentation about how they killed the whole cluster at Spotify when the board was having a visit. We fucked up. We fucked up. We fucked up. We learned from it. We implemented things. And for me, it's very refreshing. Because as you say, I mean, it will make mistakes all the time. So that's how we learn, right? I think a person that does no mistake is doing two things wrong or one of two things wrong. Either we're playing super safe all the time or they're hiding stuff. Exactly. And neither is good. Exactly. Right. Cool. And I see the time. But you spoke about type one and two. And I first thought you meant something else. So I like to move to something else. I see where you're going, I think. So this is more the cannon man kind of thinking fast and slow kind of things and moving more into, I guess, the difference between, you know, what human and the human brain is and what AI systems are. And let me just elaborate a bit more and see what you agree with them potentially does not agree with. And if we, I think we can all easily agree at least with that the type of AI that we have today is super narrow. Sorry. And it's trying to do very specific things and the human brain is much more general. Agree so far. Right. Yes. And, you know, we can also say that AI is actually better than humans in a number of tasks, like of course, playing chess and doing perhaps some image classification or even, you know, some understanding text in some way actually can be better for some very specific tasks already today. So AI is certainly bad at some things. You know, one thing is being able to have a more general general kind of understanding and reasoning capability. But then I think I would also like to say that humans are bad at some stuff. Oh, yeah. And I don't hear people saying that enough. And I get really annoyed when someone claims that a goal for AI should be had to have human level intelligence. Yeah. Because humans are really stupid sometimes. Human are really bad at some tasks. It's very easy to see. I mean, try to ask anyone to try to multiply two large numbers. Try to have an human trying to memorize, you know, 20 things. I remember it. A computer can do that so easily. So we have to recognize that humans are really, really bad at some stuff. AI is really, really bad at some stuff. And I think it's better to focus on what they are bad at. And then if we say that, then we can say that, okay, let's see that AI is probably really good at going through large amount of data in some way. We can be large time series data for some spotting, sports betting kind of purpose or it can be like a huge amount of images in some hours of video that you want to find some kind of thing that's happening or thousands of pages of text. And for humans, it would take a huge time to do so. But for an AI system, we can do it in a fraction of a second. But they don't really have the deep understanding of it. They don't have the general kind of knowledge and the reasoning capability to be able to understand what it means. But it can find it in a superficial way really quickly. So now, okay, after that round, would you agree for one that there is a big difference between the human brain, what they are good and bad at an AI. And potentially what we see really seek if we want to date to find a value from data and AI is to find the examples where humans are bad at something, try to automate that part in a semi-potentially automatic way, then have humans coming in helping out with the part that they are good at and having a hybrid, as we said a number of times now, the solution for that. Yes. I agree 100%. And I mean, there are many layers of that from problems to solve all the way to I think existential questions like singularity and so on. That's a find quite interesting. I mean, looking from like a really high existential position, I mean, you can argue that are we doing good or not as a species. So I mean, I don't know. I think there is a human's heart severely flowed in many ways and I think we need to accept that and try to remedy it and use the tools that helps us. Yeah. I mean, I think we should spend more time building AI that are good at things that humans are not good at. And I think it's really weird that so many people are saying that we should build AI that can reach human level intelligence. I think that's complete the wrong way to go. Yeah, exactly. It's the wrong framing of the problem. Yeah. We should use AI for the to be a complement to humans, right? Yeah. But then you spoke about singularity and since we're moving a bit towards the end, perhaps we can move into a more speculative philosophical kind of set of topics then and the AGI kind of things, you know, general intelligence of me, we may reach some point and it may be a singularity when we have lost the control in some way and things are moving ahead quicker than we think. But, okay. So we spoke about human level intelligence potentially being a bad way to at least guide the development of AI, right? So do you have any favorite definition or things, you know, how would you say that this is the right direction of where we should spend time improving the intelligent level of AI systems? Yeah, perhaps not a perfect definition, but you know, what is the right thing to make sure that we spend time in research and in companies to prove the value for AI? I mean, first of all, I agree very much with like we need to evaluate what activities and what work is adding value in as opposed to letting a machine do it. So I mean repetitive tasks that are fairly straightforward and simple at scale. I mean, we shouldn't do that. That's just a waste. Rather focusing on the very complex things where we need to connect dots and stuff like that is harder for a machine learning model or an AI application to do. I mean, if we could do that, that would be a massive productivity boom of everything. It will have ever sort of a societal consequences that I think we need to be prepared for and act on. But also I think, I mean, we're talking a lot about the problem with AI and machine learning being biased based on what the data we feed. But also all those biases are coming from us humans. So I think if we can, I saw you posted about like what's the optimal tax rate, for example, like those kind of decisions, what is best for the population? I think that concerns me quite a lot in the world today that is so polarized and there are so many data driven. Exactly. But I mean, but I'm ill-pre-importent societal question. And I thought I'd also like to talk to you with Henrik Langevin. If you can't agree on the facts, I mean, it's impossible to have a discussion about what's the best course of action. And I think if we can use AI for those things, I think that would be highly valuable for humanity, I think, because we are not. I mean, there are so many flaws in humans like greed and stuff like that. Now I'm getting sort of really political, I guess, and philosophical, but I think. I mean, if you take the tax example, as you mentioned, I mean, people can't really think. I mean, humans at least is very anecdotal in the way they're thinking. If you want to do some kind of simulation with a million people doing a set of action a thousand times, and then seeing what the change in the tax rate will have as an effect on it, there is no way that the human can do that. No, I would call you. The only way you can do that is by having some kind of data-driven approach for it. Definitely. And I mean, there are also, like, I think we discussed that last time, we met as well, like, economical theory as well, that. I think based on the political spectrum is Sweden, I think, to sort of left wing tend to go with higher taxes, or was better, and the right wing tends to go lower taxes are always better. But the truth and the theory is somewhere in between. There is, like, a theoretical optimal as a high-grade. Tax rate is a hybrid one. It's always a hybrid. But I mean, if you go with ideology and beliefs instead of facts, then it will always be a biased discussion, and it's, like, more of a rhetorical debate about who sort of takes the decision in the end. I mean, there is an optimal tax rate everywhere that gives the most money per capita to the state. Yes. Then you can have discussions about, okay, how do we then distribute that money from that state path to what should we find? So, it should be. That should be. in other objectives, I guess. Yeah. prolactivity and obliquity. or. or the equality and productivity, I think in that case, was the two objectives that they want to balance. And if you don't balance them, the productivity will still be suffering in the end because people will be really upset and it will cause some kind of. Yeah, that's exactly. I mean, it's like called a Lafayette curve. I think if you have zero tax rates, you will have zero income for the states. Yes, exactly. If you have 100% tax rates, no one would work. Yeah, exactly. Those extremes are quite a lot of stumps. Yeah, it's that simple. Okay, so let me test you on another thing. It's just a theorem, I mean. But there are differences between the human brain and AI, of course. And we can take the obvious thing. I would say it's like four or five different main things. Let's see which one you agree with here. I think one is obviously generality. The human type of brain is much of general. It can handle so many more tasks than an AI system can today. But of course, we see all the latest type of models that we have in AI starting to be more and more multi-task, multi-modal and multi-lingual and it's starting to become more general but still very, very far from the level of generality that AI, or sorry that humans have. So obviously that's one part, right? I guess you're not arguing so far. The other I would potentially say is reasoning. And then I think people also sometimes speak a bit about this in a wrong way. I think we have so many examples of systems that do reason in some way. You can even take a chess playing AI system. It has some reasoning thinking if we do that action, what happens then? We retract, we backtrace, we try to find the best set of actions in a rational reasoning kind of way. But it's not high level reasoning. It's not thinking of objects in a high abstraction level. Like this is an object, this is if I were to hit your head with a hammer, it will cause some kind of effect in a high level reasoning kind of way. And that type of reasoning we don't have at least, I would argue as well. Yeah, probably agree. Or agree. Yeah. And I want to take this self-driving core example just to make this a bit more concrete soon. But let me just continue. So the second thing is potentially plasticity. And what I mean with this is that the AI systems we have today is they have a data driven approach when it comes to learning the weights of it. So instead of manually programming all the rules, you just see data and it learns the weights. But the architecture for that network that we have is very manually designed. Yes. We're potentially have neural architectural search, but that requires so much data to be trained. So it's not very plastic, basically. The human brain on the other side, it can change and rewire itself surprisingly efficiently. So even if you have some kind of stroke, you know, you can quickly rewire the brain to handle things in areas of the brain that you didn't even do that before. So it's surprisingly plastic. And that's certainly not the case. I would argue with AI systems today. And the last thing, I'm speaking too much, okay, this is the last thing. The other thing is, you know, we are so like battery oriented today. So instead of, you know, the human brain is both changing the weights and making inference at the same time. Whereas AI systems is very much, you know, you first train the thing, then you make a lot of inference. Potentially you retrain or, you know, prove it a bit, but it's not done in an online fashion, which the brain is. And I'm very hopeful for like the neuromorphic kind of computing, which actually moves these things together. Have you heard about the neuromorphic architecture? I haven't actually. Okay. So tell me about that later. Yeah. We can take that after after work. Yeah. But I think these are the four things I would argue that this, at least to me, one of some of the top differentiating factors, you know, generality, high level reasoning, plasticity, and an online learning of the system as well. Do you think anything is missing? Would you agree with those for anything missing from those when it comes to, you know, what's the differences between humans and AI? No, I would agree with all of them. Maybe a bit sort of the plasticity. It's not my kill domain. So I trust you on that. Yeah. But I think it's, it's, I forget the name. What's the, what's the, that wrote the live 3.0? Oh, Max Tigmark. Yeah, exactly. Because I think that's quite an interesting thing. If, I mean, we have, we can learn, but we cannot change our hardware. Yeah. I mean, an AI could potentially do that, I mean, to an extent, if we figure that out. Yeah, that's true. Which is also maybe the part that scares people, I guess. No, but I agree with your premises. And also, if you just make it a bit concrete, like it's, you know, everyone is speaking about self-driving cars now and Tesla, what a pilot and what not. Would, would you be so, if you take a car, the latest type of Tesla, what a pilot, I think it's up to version 10.2 or something now. Would you be comfortable having your hands tied, not being able to steer the wheel and just let the car drive you around in Stockholm today? Um, no, not right now. Yeah. But then again, I wouldn't know how good it is because I haven't sort of started it on it. But I mean, of course, I guess you're referring to, does it have to be perfect? Yeah. Are we perfect? I guess it's still an improvement, or is it a worse driving experience? Or compare it to sitting in a cab with a cab driver, that's human. Exactly. That could be drunk or very tired or, you know, playing with his mobile phone, I don't know. Yeah. Which one would you trust the most? I agree with you, I would still trust the human drivers more today. But do you think, if you were to make an estimate, you know, when you think the self-driving car would be an improvement to the average cab driver, can you give some kind of, what's your thinking there? When will AI be better than the average cab driver? I mean, it depends. I think there is a bit of a disconnect between feeling secure and being rational, in that case. I mean, the break point is probably from a rational perspective, probably lower than what we think. Because I mean, I guess I'm probably also biased in that way. I sort of tend to trust myself when I'm driving quite a lot, not because I'm the best driver ever, but at least I believe I have a fairly good take on what my limits on boundaries are, so I can adapt. That does my bit. It's probably not like a way to prove that. But trust someone else, even to let you drive, I mean, it would probably be probably the best driver ever, but it would still be a bit of an uncertainty. So I think that's a very interesting question. And I don't think I have a clear answer, but from a societal perspective, we're probably already now at the point where we should ban human driving, I guess. Soon at least. Yeah, I'm sure we're there yet, but because I mean, at least a machine will not fall asleep. The machine will never be drunk. The machine will be probably quite predictable, I guess, in how it drives. So I mean, there are a lot of things safe as well. I mean, it's rather stopped, you know, than being, you know, it's very safe, much more safe than sorry. Exactly. But it's like the coexistence done between drivers that can be totally irrational than emissions to be able to adapt to human irrationality. But now's the problem. And also a lot of bugs that could occur, you know, in a software, but we probably can't trust that much. Blue screen. Exactly. That would be, oh, you can't even take the wheel break anymore. That could be kind of scary. Cool stuff. Okay, let me see here for some potential final questions. We spoke a bit about singularity, but I think we already covered that. If we just try to end a bit with, you know, a more forward-looking kind of question and eager to go in a certain direction, but I'm biting my tongue a bit here. But AI can be used and abused for a lot of things. It's happening a lot in the world these days, and we can easily imagine at least AI being used for bad things that could hurt humans and society. But it could also be used for a lot of positive things, like climate change or whatnot, or medical care and who knows. What is your thinking here? Are you afraid of AI being abused for malicious purposes more than your hopeful that it will be used for good purposes? I think that it will be mostly positive. Men det är ju en ljus med en ljus, AI, och ML. Men det är inte. Det teknologi är inte ivillt bär sig. Det är ju humanstadshuset för badpropses. Så min igen är. Men jag tror. Jag menar att. Vad konserteras mig väldigt mycket är. Hur är det som vi tycker är att vi är så kritiga? Jag tror. Jag har förstått mig att se på socialt delamma på Netflix i alla fall. Det är en väldigt tiktok-user och youtubeuser. Det måste vara en kritig utgång där. Det måste vara fäk, det måste vara en riktigt. Vi måste ta en ljus med det. Jag tror att vi är väldigt kläkt att vi inte har det mentalt och tränning, eller vad som är det som är necess att det är en fönstriga i det informationen vi har idag. Jag tror att det är mycket av en bärd konsekvenser som vi ser nu och vad vi ser i den här elektionsen i USA och så. Det ska vara lite mer. Men jag tror att jag har hört som att att det är ju humanstadskommendet och stupit. Jag tror att det är ju en stor strö. Jag tror att du sitter och tog till som man. Du kan komma till en konkurrens av en kärd worldview. Men det polarity och extremism av opinions idag, speciellt om internetboken, höjer det lite. Du inte vill ha ett konfront på person där du gäller att jag. Jag tror att det är en bra, menn, fossel, men det ska vara lite mer. Jag tror att vi måste educateda oss i det här. Det är mycket mer än vi gör idag. Jag är så tryckat att gå in till Facebook, Gädd, Alema och. Just go for it. Okej, jag är lite mer än. Jag är inte sure att du tar den konkurrens här med Mark Zuckerberg 2 års av det. Ja, det är en part. Jag är trying att få en bra väg. Jag inte tycker att Mark Zuckerberg har en hel intensions. Jag är inte sure att du inte vill ha en av det. Jag tror att du inte vill ha en av det. Man är still stupid, vi inte vet hur man kan göra en kallan som är en teknologisk AI. Så det kan vara konsekvenser som är unexpected. Jag är inte sure att du är en av det. Jag tror att du har en av det där man har så mycket power för att få en bra, men det är mycket mer än att göra. Det är en bra förslag i den här interneten. Det är en superdificka question. Jag tänkte att jag är med med dig, jag är inte en beroende. Jag tänkte att det är mycket mer unitang att vi ska ha en av det som är ju med dig. Men jag tänkte att det är ganska intressat att vara kallad i en system där du är rådd på det som du inte skulle ha. Du kan göra något och bli rådd på vad du gör, för att göra det så att det har en konsekvenser. Och då är det som hans har en bra pojta som är fattig och det är det som det är det som kallar kan vara renforskt med. Det är det som är hans för att få en bra pojta som är att göra vad som är att ha en av det. Jag tror att det är en problem med Facebook och många de kompisar att ha. Jag tror att de har kontrollat över de diskussioner av algoritmen och att de inte kan se det. Jag kan inte vara så far. Jag tror att det är en väldigt intänt av det. Jag tror att det är en väldigt intänt för de kriser. Jag tror att det är en väldigt intressat av Facebook. Det är en väldigt intressat av Facebook. Det är en väldigt bra uttidigt. Jag tror att det är en väldigt intressat av Facebook och vad som inte. Ja. Är du fredag om det? Är du trying att. Manage det in some way? Jag tror att jag vill manas i en term som är kvantiti. Sorry, kvantiti. Jag tror att det är. Jag tror att det är en bra styadianca projekt som drygerna�, men svagare terus�대 membrane. Jag har sowent någon dom och är över därför wieder annual för bilen. Jag tror att han börjar med kuk som har bara tagit upp och inte varit. Så nu är det superpassionat i kuken, som är ett problem som kan bli så här. Men han gör det så här, så jag tror att det är ett ena bra. Men jag tror att det är ett bra stort uppe, som jag tror att det är bra, men jag tror att det är bra. Och det är ju inte alldeles om att det är en kritisk eller vad du har sett. Och vi har mycket av det här. Det är det som är det som är det som är det som är det som är det som är resnöbbar. Så det är mycket jag har sett att jag måste educateda dem. Det är också att jag är förstånden till vad Social Dela Man, det var så här exeption för ett törsdag med någon skrivns. Det var så att jag ska ha ett törs. Och om vi bara gör det till exempel på det här informationen, potential och problem som vi har. Och även på den här polarisen som vi har i en kallan som är om att man är tryggt att ta en image eller något som är att göra det i en liten sida. Ja, och det här är en minst en hybrid av oss, som vi har en stor nummer. Så vad tycker du att det är en solution för det här? Hur kan vi trycka att det är det information som jag kallar oss för att vi är. Jag skulle berätta om du har sett en av det här volupten. Ja, jag tror att du har sett det här. Jag tror att jag inte vet det. Men jag tror att det är kanske hur jag har sett en av det här. Det är en av det som är demokratisk i det här är det som är i termen som du har en demokratisk system där man har en person som är informerat och är för att sköra. Jag känner det bra. Det har gått meget interesting. Det också är en wi tournament den naturen händer discardна. Den är vosselig, så mycket cose available. De känd man uppskriver Aah Emperor, їch argumentar komparen att få compreh��는 att enṇa som är jättebra, men det är behöver vara som instit기� nosotros anväxt men det är mycket lambida att l further argsa att det är en bra förslag. Jag har bara varit just också underställda min av ingåra om det. Jag har varit rörelse och har varit i en kantig om det är svårt att ha det. Jag tror att det är jättest. Men jag tror att det är inte att det är väldigt svårt att att få att sätta bilder och att du har bjudit det. Och det är bara att ha ett konkret exempel, att få att sätta bilder och det har varit en stor fråga om det är jorgon. Ja. Hur har du fått jorgon? Jag har inte, men jag var att sätta bilder i en tid. Så jag har en kvart i ennsattning, också en interna debatt av det. Och jag tror att en fråga är att vi alltid har kontänkt att vara publict på en samt planform som är spodifatt. Och då är också en fråga om att du har en önsattning eller responsabilitet för vad som är publict på en kontän. Och hur. När är det en plan för att gå in i en medie-house som har en person responsable för all kontänk att vara publict? Det samt kan vara argon för Facebook eller Google eller vad som. -Do you think Spotify made the red decision in keeping your broken on? and not censoring him in that way. But I think so. I mean, I just disclaimer that I haven't listened to your organ's podcast myself. We had a discussion actually with the Spotify term back in, I think it was in October, where we already discussed more or less like, is it not his existence on the platform? That I don't think anyone argued about, but that Spotify actually bought the, right, exclusive rights. I think that, of course, needs to be evaluated, is that aligned with company values and stuff like that. So I understand a bit of different opinions, but I think equally dangerous to allowing sort of bad things being circulated is also like the sort of cancel culture. I think that's also very dangerous, because of that as well. Yeah, because that can be, I mean, it can be applied from both directions. Exactly. So I'm also very afraid of that. And I think some, some things that Elon Musk said that, I mean, he allows now internet to a certain country that doesn't have access to it anymore. But he refuses to drop the other countries. What was his quote? It was something he's being in absolute, exactly. I'm like a fundamentalist when it comes to freedom of speech. And I sort of think that someone had to have him at gunpoint to be able to censor things, right? And I think I would like to think that that's the way. And I hope that people are, or at least could be capable of dealing with conflicting pieces of information. But I think we need to, I mean, we're like the first generation born this way. I hope that I mean, perhaps in a Twitter has a nice approach, I think. They are counseling so many things in Twitter. So I think that's perhaps going a bit too strongly. But they at least have some kind of like warning sign when something that they argue is potentially not accurate or factual correct. Yes, I mean, using an AI for fact checking, I think would be a great thing. Exactly. Yes. But then of course, agreeing on what are the facts. Yeah. Super. I mean, it's a, it's an AGI problem. What's the target variable? We have to wait until we have AI system as far as competing, yes, succeeding the humans and have much more than human level intelligence to be able to know what the facts are. Oh, yeah. Awesome. Daniel. I, yeah, let's stop that. It's a really easy to continue forever on these kind of topics. But yeah, let's continue that on the after after work, I think. Definitely. Don't let it. What's next in your life? What's happening professionally, privately? Coming months. Coming months. A lot of things. I mean, we're getting to school, obviously. So trying to figure that out. That's one of you might be living in there in coming month. I mean, we're not moving in, but we will prepare at least to live there. But I mean, the plan is, of course, to spend a big part of the summer. Yeah. So it's a lot of things like fixing, fixing the walls, checking up everything and doing some fixes. Right. Other than that, I'm starting up, I mean, two projects, fish, brain and ramping up at the wall. So that's super exciting. Some good progress. And also, of course, running a startup company is quite fun as well. Like what's the next big fire to try to put out? Because that's, do these years of us more of a startup person rather than being employed at a bigger company or seer thinking there? Yeah, but I think so. I think I found, I mean, I've tried my way around, really small companies, really big companies. And I think maybe my sweet spot as an employee is somewhere like, I would say, between 200 and 800 people, where it's still fairly nimble and lots of things happening. I'm not the maintenance type of person. That's a realization that I made. I love approaching and like trying to define new problems and try to find solutions and build things. But I'm not the maintenance person. So yeah. But now, of course, building a new company is fun. It's like so many things I have never done before. Like, how do you run marketing? How do you work with, I mean, sales? How do you formulate an offering that is understandable by the marketplace? I mean, so many things. And I hope you do it in a data driven way as well. We're not. Not at all actually. But I mean, it's like, again, that's probably, I mean, we're trying to find, and I guess that's same for all startups, like trying to find some product market fit, scale and speed. Then, of course, a great data driven way. So I mean, of course, we measure things, but it's, I mean, the volume stuff we have right now is like, I can keep them in my head still and for some time. But we have, I mean, we have a very nice Google Cloud setup built on Terraform, like best practices. So we have a super advanced data platform. So a good base is that. Yeah. Awesome. Do you have any people that you would recommend to come on this podcast? That's super hard. I was, I was thinking about that this morning, and then I dropped the ball. So I think, but let's, let's go with like a cheating bet. But I think you will have soon. I mean, Mikhail, Shilken. I think it's a person that I find, I mean, super knowledgeable and interesting to talk to. I had a privilege of working with him at Canbey and now he's at Arsenal Football Club. Oh, interesting. And I am being interested in sports as well. I think it would be interesting. I mean, I will meet up with him soon and I think you will be on the channel soon as well. Right. So I think it would be interesting to hear his thoughts on a lot of things. You should ask him about sumo wrestling. Oh, cool. Yeah. Okay. Yeah. Interesting. Cool. Well, it's been a pleasure to have you here, Daniel. We will continue talking and speaking a lot of interesting things, I think, in coming hours. But thank you very much for coming to this podcast and gain. Yeah. Thanks for having me. Thank you.