Ep14 - Casper Sønderby, Co-Founder, CEO at twig.energy

Hello and welcome to episode 14 of the Nord Search podcast. Today I'm super excited for this episode and I think the listeners are going to find it really interesting as well. I'm joined by Casper as Sonnerby who is the co-founder and CEO of Twig Energy in Copenhagen. Twig Energy have a really interesting angle on the short-term power markets. They're fully automated trading platform and asset optimizer. Casper has a really interesting AI machine learning background but I'm sure Casper's the much better person I am to discuss that but Casper, welcome to the podcast. Thanks for your time. Thank you and Casper as we usually start it would be great if we could start with your really interesting background like what you've done in your career and then we'll lead on to Twig if that's okay. I've got I'm a trained engineer like a long time ago almost and I actually trained as a biomedical engineer and then afterwards worked a little bit actually in like scene development but then I changed topic a little bit and went into machine learning and AI and did a PhD in in probabilistic machine learning and kind of caught the kind of earlier wave of deep learning. So that's back in like 2015 I did that then I had a startup for a little bit like a few years didn't kind of scale as we hoped and then I like I got an opportunity to join Google Brain it's kind of a long-term research division in Google and joined a method research scientist on AI and it's like method development on AI and actually that was kind of interesting and a very super good opportunity and but eventually I started working on after doing some like more like method development AI I worked on on data driven well of forecasting so started that effort in in Google and and kind of was one of the initiatives of that and basically we were like trying to swap out like kind of a physical model of the atmosphere with a with a machine learning based approach so I worked three years on that and of course I was kind of my entry into to interest into energy and and have very very interesting I think we had some really good early results in that and you have actually seen recently that there's a lot of very good results coming out in that space now but after three years in Google I'm back to Copenhagen and then I kind of took a little bit of the vehicle and thought about what I should what I should work on eventually a home day non on on the energy sector and how we can use software to to be a part of the transition into to a carbon neutral grid so that's really the kind of where we're to again he just started and you started that with your co-founder who's your brother right yeah so with my brother yeah and he's also got a similar background to yourself yeah he's also a technical background he actually worked for Apple for in in gubattino for a few years and then I think co-witted surface school and then he was a little bit I wanted to move back closer to like out of the lockdown and closer to family that was really the good timing for starting a company together that we ended up in Copenhagen at the same time awesome awesome so you've both got this deep expertise in machine learning or deep learning as we call it an AI yeah that's yeah very much like from a technical angle either of course we are like I mean learning a lot of other things now but really really the kind of a what we're trained in is it's AI machine learning and it was called machine learning back then it was actually my hobby yeah so everything's called AI now but yeah it was was machine learning interested so so Google brain is that like the research arm of of Google like for people who don't yeah so Google had like kind of it too too long-term research divisions so one is called Google Brain and the other one is deep mind technologies so they're kind of a little bit like similar but they kind of have different kind of cultures I would say Google Brain is this kind of sprawling a little bit like a university I almost call it with my very flat hierarchy and deep mind is made a little bit more kind of focused on some very particular topics like alpha fold and alpha go and all those methods they develop research so what are the areas of research for Google Brain was weather forecasting and then you were one of the researchers on that side yeah so actually it's kind of this amazing job where you kind of you are you are hired to kind of make sure that Google do not miss out on something so actually there was no weather forecasting effort in Google Brain when I joined and then me and a few other people started their effort and built that team eventually when I left I think we were like 15 people working on it right and kind of at the at the level where where we were like 80 testing life on Google and trying to replace all the providers and still there's a really good team that have taken that a lot further in the last three years and awesome so basically they get some of the smartest people in the world in different groups and just say gun research different areas and and see see what you can call at least when I was there I kind of joined I think a little bit like at the pinnacle of the AI hive almost but it's really just kind of shortage of people trained in that field so I mean I was just it's just pure luck right incredibly lucky to be in the field at that time so we had this incredible amount of freedom to just do what you think is kind of interesting and we'll bring you to the company and you just do that no that does sound amazing I was doing some research this morning so was it not now is that what you were yeah that was the kind of published like the the public across week we read some more like internal models like deploying them in Google and yeah well there's obviously a lot of people that listen to this podcast it were in this power trading space and we all know how important web for customers for for power trading could you give like a brief summary of like what met net was designed for and by what it did yeah so I mean basically yeah what it does yeah we were kind of targeting something that's very useful for for Google primarily which is like precipitation forecasting so you can imagine right if you are trying to plan a route in google maps for example then one thing you really want to avoid is to get soaked so we tried to kind of make really short term like precipitation forecasting data so that's the kind of I mean where we started out and really the what what you do in data driven well for casting I think it's like all the approaches are similar in the sense that you you kind of need at belief about the state of the atmosphere so it's kind of like thresholds of humidity and also like song and wind or like solar irradiation and wind and also precipitation and then essentially you have like ground truth and then instead of like running a physics simulations that kind of like simulates the atmosphere is basically like the third dynamics of the atmosphere you kind of take that piece out and you put in a machine learning model that tries to predict the next state of the atmosphere like one hour into the into the future for example and you run that forward multiple times and you can then like to use one hour forecast or like we actually ran in two minutes intervals we're kind of predicting two minutes into the future always so two minutes four minutes six minutes and so forth up to about three days I think was the in the end when when I stopped but but really now they are like a week or 10 days but really the main benefit I think of this is that these machine learning models I mean they're incredibly flexible but the other thing is that they are really fast to run so you can run like a like a forecast in like a few minutes seconds maybe perfectly parallelizable so you can kind of deploy them on these kind of parallel accelerator that like people use for training deep learning models now and that just means that compared to like a physical model that takes maybe like six hours to run then you get this kind of if you're talking short-term right you kind of you kind of can collect data up to very like just now then you run your model and you immediately have your forecast so but for the physical model you have to kind of stop collecting data and then run the model for six hours when starting predicting that actually means that like even if you want predicting like one hour into the future you are kind of working on data that's six or seven hours old okay so it's faster and getting more accurate like short-term prediction yeah and the other thing is that you are I mean we can write up the physics of the atmosphere in the case where we could kind of that's actually too hard to do it's it's it's about like seven equations not super relatively simple but but but the problem is that you cannot really compute that on a kind of granular enough level to be able to simulate it accurately so right now the kind of level the most kind of high resolution model solve it at a great that's about one by one so kind of each then there's that I get cube of one by one kilometer and then one hundred meters and you know this data the atmosphere and that one but everything that kind of happens at a scale that's smaller than one kilometer it's not really resolved that well like for example cloud formation that's really like on a kind of a micrometer level and that's then you still need to do some kind of approximations of to to make the physics works out then you kind of do this it's called the parents are my anyway it doesn't really matter what's called it's it Yeah, but it's really like you make these approximations where you're kind of already used a little bit of machine learning to to kind of a make kind of solve problems that happens at a scale smaller than your kind of a grid. You can imagine you have chucked up the atmosphere into these kind of cubes that are one by one kilometer. And really machine learning is just great at approximating stuff that we don't fully understand. It's super interesting because I speak to a lot of different training houses and they stay like weather forecasting is like the big edge. Like everyone's trying to get more accurate forecasting because that's going to help obviously when it comes to trading on the short term. Super interesting. So, could we talk about the reason you see obviously left Google and then you started up twig like why did you decide to get into this space or actually could we talk about twig first then we'll do the why after that. So, who is twig? I mean, we are a power trading company and I think what sets us apart from most other companies is that we are trying to automate power trading completely. That's kind of the internal how we work in terms of markets we're active in we acted in asset optimization and proprietary trading. So, on the kind of asset optimization we we acted in the Nordics right now and basically optimize like batteries flexible assets and renewable assets in general. And we offer all services they need to go to market and that's basically we like to get in as early as possible. So, can be building the business plan with with asset owners kind of helping with the financing and then of course after cd and to the actual market optimization and bidding that they need. And then the other arm of the company is is the proprietary trading so so there we I mean we trade for profit in Europe and in the US and we kind of like to I think they they go a little bit hand in hand in the sense that we like to. So, we enter our market as like with proprietary trading first then build all our technology kind of understand like in depth the market and then get all the operation of running and then I still to kind of follow up with with the asset back trading. So, that's super interesting approach where you do the property trading first in a market and then you follow suit with the asset optimization so you're property trading in Europe and the US on the short. And then yeah, but and then asset optimization in Nordics so far yeah so far we're explaining into Europe on like like other countries with the asset optimization I mean like anything that involves hardware just takes a lot of time right you know from people kind of start talking about a project and you kind of develop the relationship usually take like maybe one or two years right for them to actually go into operation. And so it just takes a better kind of to know what time it's just slow. Cool awesome so I want to get into detail in a couple of those areas but so that's the who is twig so could we just also cover like why did you get into this space like why did you you got all this amazing expertise in machine learning or AI why did you decide to come into this power trading space. I mean of course there's some elements of kind of just chance but really after I stopped working at Google I had the kind of luxury of to be able to take it like someone's off and just kind of think about what I would like to work on. Also had a couple of kids and I kind of changed my perspective on many things but I kind of had this thing that I wanted to I mean do something that's useful in a more like broad sense then for example optimizing using engagement or whatever you can do with machine learning. So I had like a few things I was interested in that like one is kind of the major transitioning of everything into a like a carbon neutral economy and the other one was more like. Which is actually it sounds very different but but self-driving cars was my other kind of interest actually almost joined a self-driving car company but then for family reason ended up turning down the offer. But really I was thinking about how to do something useful really something I could tell my kids where you know I spent my time on something that that kind of makes sense when I look back at it. That was the kind of major really that the why I would say after we started like thinking about energy and energy markets I started visiting like commodity traders and kind of also consider taking up a job as some of them. And one thing I was kind of super biased in this but I mean I think the state of AI or data adoption is was fairly low I would say compared to what I was used to. So obviously I said I'm biased by because it's my my my hammer is and you know machine learning hammer you're looking for nails right but I thought that you you know I could not see any reason for why we could not make this not more efficient by automating the whole like short term powertrain business. And I can talk about why I think that's a great deal but really that was the starting point so we kind of at this three step plan. Well first thing was to build a platform for fully automating short term powertrain so it's anything less than a few days of of kind of the time. Yeah and then this after we have kind of done that we wanted to do something maybe more like tentably like kind of more focused on the green transition or like directly impacting that and just really like starting to optimize as it's and make sure they actually get the revenue they need. And then the third step for us is where we are I think we have we have found a well underway on these two steps but then the third one is for us is basically more broadly make sure that we actually build renewable infrastructure as fast as we need. So that's you know that's the more like kind of the vision of but how do we innovate on on everything around like there's both like the optimization but there's also how do we actually make sure that the financial works and they actually kind of have the right material for the business plans and so they actually get funded the projects. Yeah that's the kind of a what's your goal like maybe more long term to go here and cool. No it's great because it does have a real purpose it's not just trading for profit is it like you are helping the energy trading transition. I actually find similarities with with why I sell just to do powertrain it obviously completely different angle but you know young kids what to help you know I couldn't recruit in any market but why to do something that's actually helping. And also on the quantitative side like like you mentioned like it still feels relatively early compared to the world you're from machine learning is I would imagine what the expertise you're bringing there's not many shops out there that have that degree of sophistication on a machine learning side I talk shops though I don't. Yeah I don't think I wouldn't know actually because one thing that also struck me when I kind of you know I'm from this maybe at least back then when I was in in tech it was very open and kind of you know you publish your results and you kind of share. And then I you know I joined this kind of a try to bring into this kind of maybe more like commodity world and I mean then maybe secrecy is your modus of operation what. So I was a little bit struck with kind of like how many people actually wanted to tell me I think it kind of maybe tickles something in my mind a little bit like why are people so secretive. For me that's the kind of sign that there's something going on here that maybe it's not like you just need to have the right knowledge and a lot of people can can participate. So I thought it's not I mean if you know you do something that no one else can do right then you're not so afraid of sharing it. So that was at least the way I thought maybe I was a little bit you know you press or whatever you call that right. No that is definitely like I said I don't see many PhDs from Apple and Google in this space so I think you can bring a wealth of knowledge into this area. So talk about the fully automated part of how you trade because that's super interesting to me. Am I right in saying that you have no human power trader in the loop it's 100% automated. I mean we have certified power traders of course and we follow all the requirements that you are like all the regulatory stuff you need right. But from a kind of background point of view I mean we don't have a trading desk in the sense that we man a 24 hour trading desk. In the same way as you're like and I mean we we man something else and that is kind of a surveillance of everything. And of course we have all the like you know the red button that you need to be able to press if things go wrong which they likely have not done. But really it's like on the inside right we're basically truly machine learning people and proper like soft engineers and that is served up well I think. But I think it's kind of culturally a different company right from most commodity traders so I think some things are easier for us and some things are definitely harder. We are much more focused on developing software that scales and can kind of be used for many different things right. But of course there's a few things right where we really had to learn it the hard way how you like all the maybe regulatory stuff. And there's some kind of all these things that are hard to read from a kind of the market entrance manual of course we had some catching up to do there. So I think we're lucky now like I mean we're up to speed on everything right but that was a lot of a hard work where I mean it's always easier to write something down as a little plan on a piece of paper that actually executing it right then all the details actually starts to matter a lot. Yeah but you're what four years in now as a company so obviously you I guess you know every day you you've built all that experience of what you need to do to be able to trade in these markets. And I guess would you say that it's much easier to do it that way around like bring the expertise on a machine learning side and then learn the industry stuff that you need to learn. I don't know. I'm still right because I mean I think that's that's marriage to both approach approach is right because I should say maybe what we. Well I think we have an advantage is more that we kind of focus very much on like systematic insects. Like stuff that kind of like broadly affects the markets or kind of like you know obviously the weather but could also be like it. inter-connectors or like outages and stuff like that and prices and different zones and how people live and all that more like being like broad view of the market and then we try to like have a very good model of of what it then happens in each market. What we don't focus on where I think other people have been much more or like have more knowledge than we have is more like focusing on some very particular kind of thing where you know something about like for example the swish to friends interconnect or something and you just know in certain situations that this is kind of mispriced or like not very efficient. That's where I think you need a like a human I can have some knowledge there right that we just don't know but I think on this like systematic side I think a computer is just the right to like an algorithm and you can go into all the reasons for why I think that it but yeah well I'm really interested in algorithmic trading so if you could give like you know a brief overview of what I'm not going to be cool. Yeah so basically we started out at the like Interday markets in Europe that's the kind of genesis of everything we have like the company and now we have branched out into more things but really building a software platform that handles the real time data from the exchanges so there's like all the auto book data and more like all the bell forecast that stream in and then we every time we get an update we kind of have a belief about what we think the price should be of all the contracts in Europe. So right now I think like the kind of amount of data has been increasing a lot like the amount of bits and offers and the market the amount of trades has been I think like almost toppling in a few years. So I think right now we see something like like a thousand messages per second or something that we need to handle. That's really what I think our software platform does really well. It's really fast and really kind of a solid in a kind of a technical point of view that it handles all the like it's fast enough and it can kind of recover from outages and these things that you've struggled with in like exchange have like server problems and then all these things. So that's really the kind of the backbone of what we do and then we have like machine learning models that yes as a predict what we believe the price should be of all the contracts and then it's the deviate from what they are now of course we can buy us out and that's a very short term. So Interday market like under our time span or something. So on that then is that like high frequency trading would you class that as high frequency? No I would not call. I don't think anyone does really high frequency trading in power markets and there's also a couple of reasons for not doing that. So traditionally high frequency trading right to try to exploit kind of a arbitrage between two different exchanges and that doesn't really exist in the power markets because there's this shared order book in the Interday market. So everyone actually kind of targets the same order book if you use not fool or epex bought or it but I still think we are we are kind of fast. I think we have the fastest platform in the market and I mean is the computer right? You can put your computer where the link is low and you can I mean we are in the micro seconds right for update and we spend a lot of time on that. I mean it's also a little bit because it's our we have some people who are interested in it and think it's kind of fun right to kind of tune the engine a little bit. Yeah I mean I don't think it's strictly necessary it's more kind of interesting. So obviously you say it's not strictly necessary because of the shared order book but what advantages do you have of running a really fast platform? I think the main advantage is that you kind of when you backtest your models you actually get executed at the price you expect because you kind of know that there will not be anyone or very few people who will kind of be able to to front run you on on information. I think that's the main advantage right for us. So we don't have this kind of nefarious opponent who always knows something a little bit before us. Machine learning is kind of this is kind of more like on the technical side or like computer science side I would say we we have spent a lot of effort on kind of making sure that it's such kind of fast that goes as it can be right within of course you can always go deeper on that and do like really hardware optimization and I do a build an FPGA build system or something we're not there yet it's not like I don't think it's kind of unreasonable if we're not on kind of making it fast but I think we are kind of a you know fast on execution and processing of data and the other thing is a corporate modeling like can you run your models fast enough and actually make your predictions are they like of high quality and of course there's goes more like kind of knowledge into what type of data do you need what type of features do you need how do you pre-process it how do you train it and how do you make sure you backtest correctly and all that that's really our kind of the where we are have a strong background right both from from a kind of industry experience and more like academically as well. So that's systematic data driven approach to to coming up with models and you can almost feel right why we are kind of more systematic driven than some other because you to kind of who does all you need like some amount of data you just need a kind of a sufficient like kind of bulk of data to to make sure that you can you can kind of have a robust model and if you're kind of targeting some little like work in some corner of the market there's just not enough data to do it and on a kind of I think machine learning based model and that's that's really where humans excel right and I think some other people probably have done really well. Yeah cool and then obviously we mentioned about your weather background are you using your own like proprietary like weather forecasting technology. We see AI as a kind of our core piece of technology so we build every forecasting house and of course we also buy some from you know there's a few good kind of forecasting providers but that's more giving us knowledge about what everyone else believes about the forecast right and then we run our own weather forecast and power and price forecast and then both as like focusing on super short term which is kind of in the second minutes hours time scale and also more longer term ones so more at the kind of day head kind of timings so that's like 12 to 36 hours usually and I mean I think that's our kind of competitive edge and where we need to be really good so that's all in-house and and I think of course you need some people who kind of are trained in doing that and they're I think great to think about that is that you really know how the model was trained and what you know you know it's more like if someone else gives you their historical forecast right you know this is what I said last year of some forecast you never quite know if that was really what they published or if that was something else or like it's they changed something in between or something right. Yeah and I suppose everyone else has it as well yeah that's also yeah yeah I think that's true so this is in a world today where every company markets like powered by AI and they're all working with AI but it'd be really interesting to just define like what you what you use AI for and it's in-house like you're building your own AI and what do you use it for so we talked about whether that so that's using it for forecasting which is obviously a big competitive edge as you mentioned. Is there anything else that you use AI in the business for? I mean I think basically everything is driven by machine learning or AI and I mean that's what we do right so we of course also target problems where it makes sense so there's a couple of I mean on the kind of just purely forecasting right which is useful for many things I mean there's the weather path which eventually right you I mean we are not super interested in in weather but we know we interest in in power production power prices and that's more like a you know what you derive from the weather forecast right so that's the one thing we focus on right and then of course on the price level I mean you can talk about price on many I mean you can forecast the day ahead prices of the interday prices but really you can also I mean it's very it doesn't really matter how granular they are right so we actually forecast all prices of all products at every update that ticks in so it means that we kind of update our price I think about a hundred million times per day on on all power contracts so it could be like a you know quarter our product in like France at some particular point in time so that's the kind of granularity we're working on and of course it takes some kind of effort on on the infrastructure side to kind of make that work and and be able to run your bodies fast enough and all that but really in any of the right you're interested in yes as a power production and maybe also consumption and then power prices right can I sell it is price too high to low now and that's both for proprietary trading but it's also what you want to do for asset optimization right this is the similar thing you want to sell at the point in time when they choose to make like the best profit right and we can go into a little bit more like of course if you're talking about flexible assets that's kind of another layer on top but yeah kind of mega lots of life be your building everything in-house is that like yeah awesome can we talk about the multi optimization angle that you have as arm of the business as well because that's obviously super interesting as well yeah yes so basically asset optimization is that correct yes yeah yeah okay so I think like as I said we are kind of a I think that I said that into the introduction we are active in the knowledge power market on on asset optimization and it's both like basically we're targeting mostly assets that have flexibility and that most assets actually do have that today even so I'm going to have the flexibility they can kind of return off and then kind of optimize in that way but multi market optimization is basically to ensure that the assets owners maximize the profit by participating in all the available revenue streams I mean that's the kind of you know so there's not kind of any magic involved in that but of course it's it's pretty hard to do in practice because basically when you you have an asset you can put so there's the power markets like they hit and like the spot markets and the intraday market you can participate in but then batteries for example also have multiple other markets that they can participate in so these angelic service markets which like primary research and secondary serves and they come in various flavors depending planning on exactly what country we're in. And really the problem that turns into, and you kind of have to enter these markets, you can imagine there's a day ahead option at 12 o'clock on the day before delivery of the power. But then there's also these other markets that are like these ANSI service markets. Some of them happen like early in the morning or like at 10 o'clock and also some of them happen later when they're ahead. And some of them will kind of be mutually exclusive. So you kind of have this, every time you have to want to make a decision, you have to have a belief about what could I earn on something else later that I can participate in later right. And that's really the hard part that you need to forecast your kind of expected earnings on every combination of market share from participating. Actually the optimization turns out to be, once you have the forecasts, and you know your operational constraints of the battery, and most of the optimization into what you should participate in is not so complex. I mean, it's a little bit complex, but not very complex. It's called, it's the linear optimization, most of it, and we go into technical or land here, but really getting the right forecast is hard. So that's in the battery optimization space, which I know is there's so much going on there. It's probably the hardest area of power trading at the moment, everyone's interested in that space. So you will optimize clients' assets, and so you'll take the risk, and you're like, will you do like a contract where you get a fixed price? They get a fixed price, and then you trade. - Yeah, actually right now we, yeah, you know, there's an interesting question, because right now, until now, I think pretty much everyone has done this kind of revenue share model, which is basically you kind of get a fraction of the revenue or the profit of the asset for doing the optimization. That kind of aligns your well with the asset owner, right, that they feel the world more and better. And some do this kind of one, like watermark basically, I kind of have to hit some baseline, and then above it, you get there like a kind of a kicker. But traditionally, actually, it has been kind of also, both in Germany and the UK, and also in the Nordics, there has been this kind of, until they're a service market, that has been very lucrative, and basically you just wanted to participate in those, and then you were doing well. And then recently, those market are though, pretty kind of shallow, I would say, they're not super deep, and they kind of, they fill up pretty quickly, and that's already happened in the UK, right? But then you want to move into more, like kind of a bigger market, and that's where optimization starts to be really interesting. Because then you know, you have these primaries that are, they're not really using a lot of energy, they're using a lot of power from the battery, but they're not using a lot of energy. And then you move into energy markets, where you kind of, you know, you really use your battery for like storing energy, and kind of, well, selling it back, they're kind of arbitrage. Like what do you expect you would use a battery for right is charging, you know, during midday, when the solar, the sun is shining, and then more floated later. That's the kind of traditional arbitrage, but there's also kind of, until there's service markets that target that, where the market operator of the TSOs, they kind of conflict you for, for kind of stabilizing the grid on energy. So you kind of, you know, we send out a little bit of energy if they like some. So there's these kind of, you know, these kind of fast for service, then there's kind of energy as there's service markets, and then there's purely arbitrage, we just participate in the spot market with the battery, and kind of getting all of those to play together is, and some of them can be stacked and untold with each other, and some of them will block each other, and you can get into kind of a state of charge with your battery, so you cannot do certain things. So there's these kind of interesting problem, I think, and not trivial to get right now, when we have kind of all these like five, six different markets you can participate in. - Yeah, so how big for a technologist, so how big a technology challenge is doing the market optimization, the asset optimization? - I would say it's, I mean, there's obviously like a huge operational part of it, so kind of, you know, making sure you, so we have this kind of a little hardware box that we develop that we kind of put out at the asset, and then controls the asset, and we can kind of establish communication with it. That's of course, there's a lot of like stuff going on there, like, you need to encrypt and you need to be kind of, you know, it's kind of infrastructure, right? So you need to be, you know, do proper software engineering and cybersecurity, you know, that stuff. That's of course, technically hard, and you need to do that right. And then of course, there's the more like the modeling part of it, and I think that's kind of interesting, right? Because I think it's turning out now to be pretty hard to do. You need people who are really good in forecasting, who are good in optimization, and who also have some knowledge of the power market to kind of do that right and kind of regulation, and can, you know, you get this 100 pages of regulation from the TSO, and they say, "Hey, this is what you're allowed to do." And then you have to read it, and you have to understand it, and kind of be able to participate in a kind of a discussion with the TSO on, what can I do, what can I not do, and yeah. So I think it's increasingly complete from a technical point of view, and interesting. I think it's interesting, right? But it's getting harder. - Yeah, for sure, but what I would imagine for, let's say you're trying to win an agreement to optimize, and you tell them, yeah, we don't have any traders, it's fully automated, because a lot of the competitors, I think they still have, you know, they picture, we've got traders, intraday traders, we know the market inside out, and we can obviously make sure you get the best returns, and then your pitches, well, you know, we don't have any traders, we'll just fully automate it. Do you get a bit of, do you get some people saying, "Oh, that sounds scary." - Well, I mean, I, yes, sure, we do get some right, but I actually think most people buy the argument that we've made that, basically, I think the argument, the wee pitches that, I mean, power minus, traditionally right, has been running at this time, maybe like hourly schedule, right? You have these hour contracts, or maybe even four, I was like peak or baselo contracts, and then, you know, that's kind of, then you have a few, like maybe like, 20 contracts, or like 30 contracts per country, that you need to follow, which is kind of manageable, you have a little bit of trading, maybe like a few ticks per second, or like, but then, I mean, as we kind of add more and more renewables, right? And people, you know, there's a quarter hour contract that means that now there's about like 130 right contracts per sown, and also the amount of, and people are beating in small amounts, more amounts than in today markets. And that's just increased the flow of information, right? And I think we are at a point now where this is very hard, that's a human to kind of actually track the information, react fast enough, and that's really the pitch we are trying to make. I really believe it, that's true, and I think we'll just kind of accelerate this development, that it's just information overload from for human being, right? Either you need to have really a lot of traders that follow very few products, which is obviously expensive, and also traders, there happens to be quite a lot of churn in a little trick of, you know, in companies. So you need to train you once all the time, and I think that's our advantage is more that we build up knowledge that are kind of more stable and persistent in our, in basically in our algorithm, right? I believe that, and I think a lot of people buy that, of course, some people take a little bit more convincing than ours. But I think actually the major, maybe a more important thing is track record. Like people want, if you can show them track record, then they are, I mean, confident in you, or you can guarantee them a flaw, like a price flaw or something, then it kind of makes people more confident in you. - When we talk about track records, how would you say your track record stacks up against your competitors in the market? You don't have to give specifics, but like how well are you performing? - I mean, obviously, I don't know the details of how everyone else is performing. So actually the UK is the only one where you can see how everyone is performing. You can actually not see that in the Nordics on Europe and general. I mean, I think we do well, and I mean, we are constantly improving our algorithm, and that's just kind of, you know, it's just incrementally get better all the time. That's just building knowledge in the system, and I can see we win contracts, right? So I think that I take that as a kind of token that we have something to offer. That's interesting for people. - Yeah, and obviously you're in the Nordics. When's the plans again to Germany and overall? - Yeah, so I think we are about to enter the Eastern Europe more like Baltic's area, and then also in Northern Ireland and Belgium. - Okay, cool. - As the kind of next steps. And I think that will have been within a few months for the Eastern Europe, and then probably within this year for the other markets. So I mean, we need the partners right who have assets that are deployed and ready to operate. And that's, of course, takes some time. I think Germany is also very interesting, and it's the biggest market in Europe. Of course, there's some really great players there as well, right? So obviously, we're also focusing on where we think we have a customer base, and that depends a lot on that. We have some partners that want to expand into areas that we can then make the entrance for the invasion. - Yeah, so it's customer driven. - Yeah, it says that area. - The strategy. Okay, awesome. Could we talk about Twig's culture and values, going to that for people that think, oh my God, this sounds so interesting, I want to join. Like, can we talk a little bit more about that? - We are a commodity in the energy business, right? So we are, and from the outside, we're delivering commodity services and our energy trading services. I would call, from the inside, we are much more like a tech company on how we operate. So I think something like 80% of our people have like tech background, computer science, or machine learning. So it's kind of, I think, builds a little bit like a different culture. I would say, from kind of values, I mean, we are building a single piece of software that can do things we need, and that can only be done by a great team. So we rely on much more on teamwork, I think, and very skilled people to build on that kind of common project. And that just means that you know, you have, I think, we have very high thoughts on actually having a well functioning team and people who are able to participate in that, right? I think we don't have this thing where you have this kind of great human being that sits in front of a screen and run their own book and all that. Yeah, that's probably the opposite of the strategy we are following. That's just because we are software company, right? From kind of an operational point of view. So that's a kind of, I would say, how we feel and the teamwork is super high, right? And also, I think we have the, an advantage that you have as a startup. We have many tests. advantages for volume is that you start from a plane. One kind of advantage is that you start from playing sheet of paper, right? So we run a modern tech stack and it's solely focused on the things we want to do. So it's developed for that and that just means that it like we can do daily deployment of our like updates, we can like have extensive testing and monitoring and all the things you're having are kind of a technology company. Just it was built from the start that way, right? So that kind of makes it, if I should maybe go into more like value, I mean that's of course a little more like that to be a little bit more inspirational, I would call it right almost, but I think we ambitious right, I mean I mean I have this thing that we want to make a little mark in the universe right if we can and that's what we should aim for. I think climate changes our generational problem and we want to have that as like aim for having a sizable impact on that if we can. So that's a kind of you know one way of so even though you're kind of maybe more like I would say maybe stuffed in a way on the you know culture, I still we're very ambitious still so those two can still go hand in hand and then of course as a startup like innovation and speed is just super crucial, right? So we try everything we can to make sure we as lean as possible and as a value like everything we don't need, we will cut it away. And so that's I think move fast by this if there is a value that many startups have and they think we have it for good reason. And the last one is I think we also focus on it's basically as a company we are in this for the long run and we are in this is a kind of a long term goal and and and hopefully a common one so and I think we can only solve like making the power with carbon neutral if we collaborate. That's both internally like as a team but also with our partners and with with a risk stakeholder basically involved in this something we the focus on is basically making sure that of course we want to succeed but also thinking as a kind of like society we want to succeed in this and we have to succeed right. So that's the kind of values I would call that. So is that what you look for when you're hiring people? So people who share that passion for you know helping with that renewable transition? Yeah I think I mean we of course like everyone else we look for brilliant people right and I think we are you know companies just pretty much the sum of the people right. So of course we look for great people like it's they have to be technically brilliant right of course and those are of course hard to find those people who really excel at what you're looking for but I think on the machine learning side and the modeling we have a pretty good like given my background and and assurance background and we have a pretty good handle on on kind of evaluating people on that and I think that's one advantage for us is right we can actually talk to an expert in machine learning and really understand if they kind of know the things we need. So that's of course one thing we look for the other one is that people who can like I said this thing about like move fast and and think big right and I think that one thing that kind of people who can do that or excel that is usually people who actually excel at getting stuff done and I don't know where and I think that's just everyone is looking for those people who can kind of move the goalpost on the meters forward and then we'll do that. That's of course something we look for a lot and then the last one is I think team working right like we cannot hire someone like people can be great but they also have to be a net positive once you kind of put them in a group so they need to fit in right to the group and some people fit better in some groups and others right and that's just you know the I guess the beauty of diversity in and of course that's the important part as well. What things do you look for to make it clear that they can work in that team environment and operate well with you guys. So one thing is of course I mean when we do interviews right I think it's a little bit hard to kind of know if people will fit in well because I mean obviously you cannot observe people for that long and see how they agree. So one thing is when when we interview people I think we are general I think very nice people but I think we try to kind of have an honest discussion on kind of some technical topic and also do a little bit of like pushback on some like thoughts that maybe or like some project that are candidate spilled and then like just see if we can have a kind of a good discussion on that and of course if we do not manage to kind of establish a good discussion in like have a good discussion in one hour like then probably will not manage to have that over a long like extended period of time if you already can feel that you are kind of a little bit like tense or something then of course that's probably not a good fit for anyone. So there's one thing and the other one I think just for me it's just incredibly important especially on the technical roles that people save and there's something they don't know and I think it's perfect fine but I'm kind of like I'm kind of technically skilled enough to kind of know when people are just like we're soup you know yeah just kind of a decent they just kind of tell like probably finally does something you don't know and the kind of the worst is that someone tell you that they there's something they know and they're not they actually didn't and that's just much harder to fix them. Yeah no I get that and I can imagine with like super technical people sometimes that people don't want to admit that they don't know something but honestly it's the best policy way like you said just say I don't know this thing but I could probably pick it up pretty quickly. Yeah yeah I don't know it's take yeah I don't actually know if technical people are worse than anyone else of course there's also I mean to be fair to true kind of Kenya as well I mean it's also super artificial setup right you have during an interview right we tried to do these kind of we have a kind of a practical coding for technical hours and then we have a more like a kind of a discussion of of basically like theory around like math and statistics and and stuff like that. I mean I'm kind of aware right that it's an artificial setup and it's stressful and it's like there are other things right that are not nice but we have not found a kind of a a platform like personal recommendations we have not found a better way than actually doing like a series of interviews and some really practical ones so one thing we've also thought was on is like do they really like when you talk about machine learning I mean a lot of people can take the newest take like a high torch or like some machine learning page and build a little model and then run it but when you start to kind of question what is it actually you are kind of optimizing for what's the kind of treat or if you made in that kind of lost functioning use that's also stuff we look for right so kind of in-depth knowledge of what are you optimizing for why is that a good idea in this particular setting and what the kind of mistakes you would expect from this type of model versus another model and it's kind of in-depth right breath of knowledge also on the statistical machine. Yeah I like you said I think a big advantage you guys have is you're able to do in a discussion determine the level of a machine learning expert I guess sometimes it's hard when you don't have that big level of expertise. Yeah I can I can actually tell you the reverse also like the kind of office it also happens right we just recently hired some business developers and some financial controls and that's like challenging for me to hire like I mean that's super hard honestly like to see you hire for something way or not an expert that's just like really hard and and you kind of like I really don't like to go with my godfilling in general about I mean you really have to do that almost because is it true what they tell me or is it not true that's kind of like just do not have the knowledge to kind of know right and that's just oh like probably yeah that's a kind that's just the way this but yeah absolutely I think sometimes that's where a good head on second coming advantage. I agree I totally agree yeah no no I agree that's totally true. Something that I want to just ask as well is based on what you told me I imagine the people you're hiring them don't really have much or don't need to have much energy trading background like you just hire you know based on machine learning aptitude if it's a technical person obviously ability to work in the environment that you've created so if you see that happen like you don't need them to have industry experience you you can just if they're smart if they know how to solve problems if they've got a machine learning level they'll work out and they'll be good good hires. I think I mean in general people who I mean we hire a lot of PhDs also when I think I want a good good thing about I mean usually you know the exact topic that they kind of the dip was maybe not excited what we're working on but I think what they have learned is basically to solve problems and they have also learned to have a like basically used to things that is super hard to make things work in real life and also most things will not work so that's that's yeah of course and I think those people have kind of proved that they can like for example do a PhD or do some really good projects and like publish them super nice stuff and get hop for example and don't know like an extended period of time they have usually proved that they can kind of be go like to the kind of frontier of knowledge right and then actually contribute a little bit to it move the our kind of knowledge of some fields forward and I think they will just in general be good at most things right yeah I don't know exactly what it is but you know I think probably some people just really good learning strategy or like a tenacious enough to just like keep on hammering against the wall until it breaks I think and those people are obviously super great too. Yeah yeah absolutely. And on the back glass of course we now we run like internal training earth trainings already on internal training on our new hires so we kind of bring them up to speed on on how the market works and how the regulation is and all that stuff right so we kind of have built that knowledge in there in the organization now but be more higher for the technical background I would say we actually looked a little bit for some quants to join and I think maybe we will hire one at some point I think what that allows us to do is just to kind of you know kind of kind of get like a lot of knowledge that we just do not have right now like on for example if we like it's much faster to hire someone that have a knowledge of a particular market and then having to learn it yourself right and there's frozen cons and false right but I think yeah that's one thing we can we can do now as we are a little bit bigger. Yeah no no it's really interesting because that I've worked in this space in the quad building quantitative teams in the energy markets since like 2017 and I've seen like two different types of approaches your approach where it's just like let's just hire super smart people they don't need to know the industry they'll do really well and that has worked amazingly like one of the best hires I made for one of my clients he he had a PhD in mathematics but he came from drone technology like not anything to do with energy trading and he was like one of the best hires they've ever made because he was just super smart and got it and then I've got all the clients who only one people with quantum, with fundamental knowledge in energy. And that, you know, they seem to work too, but I think you have the blend of both. I think that's also one of two. I agree. And knowledge in any, you know, a lot of people all technically excellent, I think. I think a lot of me also comes back to like, I mean, we are out of the tick world, right? So I mean, our kind of network, the people we know are tick-force. And of course, if you are out of the power training, well, you don't even know other powertrails. And of course, that kind of buys us your own. I would say on the fundamentals, I think that's kind of interesting. I mean, I think we kind of try to let the data speak for itself and analyze the data well. But I think what I mean, there's also, if I, I don't know if I can say that without being too offensive, but I mean, it's not super hard to know what the kind of underlying drivers are of the power market. I mean, it's the weather right now, like pretty much. Of course, it's hard to actually execute on your kind of knowledge on the weather, like the power production and actually do good forecast. Right? That's of course super hard to do and know in what receives what matters and all that. And that's of course a challenge. But I think from a kind of very high level, I mean, there's a very few things where it actually drives the power market. I mean, it's just stolen wind and temperature. And that's, you know, and then maybe if the nuclear power plant actually works, it's also a big problem, right? But yeah. Yeah. Cool. Awesome. Well, thanks for that. And then just one of the last things I want to ask you is looking ahead, looking to the future, like for Twig, and maybe the market is a whole lot. What are the big things that you see in the next couple of years, three years? Maybe on a kind of the, I really believe that we can automate everything in power. And I think there's a couple of other companies that are working on the same. But I think anything that's kind of less than like two, three days, I think we can, we can automate that. That's basically what we're working on. We kind of start a super short term where I think automation is kind of already now is the what you should do. And then we are working towards like now, entering like the more like they had time frame, and that's where we are focusing also a lot now. And then we're moving to the future market, right? So that's a kind of, you know, maybe like on a kind of time scale of the market we focus on. Then maybe more as a company, right? We are, I think I already mentioned, right? We're expanding geographically under the asset-backed optimization. So that's like entering new markets. So we are in the Nordics now, and I took them in all the markets in the Nordics. But we're entering into the Baltics and Netherlands and Belgium soon, right? And hope to make it, yeah, basically bring on a lot of customers there. And then basically like power trading while we are in the US and expanding to cover all of the US as you speak pretty much. And then we are also actually looking at the APEC. And like in Japan, also Japan is interesting for us and Australia of course. Yeah, we're big a few other markets out there. The next big market is everywhere. But I guess what's interesting for you guys, because you're fully automated, it's a plan to do everything from Copenhagen. Yeah, so I think well right now we are, we have less of a problem about having to man like a fully staffed power desk like 24/7. Of course we need someone to kind of monitor all the time, but we don't actually need, like I would say the kind of, you know, the monitors, like the people at risk still need to be allied all the time, right? But the kind of execution, you know, the actual hands are just automated away. And of course that makes it easier for us to only have a single office right now. So I can see like a lot of our colleagues in other companies, they open like offices in Singapore or in Australia or Japan and I think due to the time difference essentially. Yeah, absolutely. I'm helping quite a few clients now open offices in those areas. I guess it's an advantage for you that, you know, it's a lot of time and operational headaches to do that. And if you can do everything just in one office efficiently, that's going to be a big advantage. I think you can keep this in mind. Yeah, I can keep my lean right, I think, that's of course an advantage. I think there's really a lot of the advantages into having people at the same location that can kind of check with each other, like kind of develop ideas. And I think most ideas kind of, you know, just kind of grow a little bit like bottom up and you just need people to bump into each other at a coffee machine or whatever that happens, right? It's a little bit harder to do one on like online on some video chat or something. Absolutely. So you all worked together in the office, that's your goal. Yeah, so I don't know if we are like old school in that regard, but if we actually prefer that people come to the office, of course. I mean, that's the, like people kind of causes they have a need to have some good reason for not being in the office. Of course, it's perfectly fine if they need to go to doctor or something. They won't come out, like they need to pick up their kids early. They know that a ton of good reason why, but really more like the kind of like we actually prefer that people come to the office. And I think it's actually interesting. I thought a lot about this, but I kind of like, you know, the people who can work alone because they have a lot of knowledge, they can kind of work independently. Those are actually exactly the people you want to show up in the office because they, you know, other people rely on them and need to ask them about things. And kind of, so I think it's hard, at least when developing something new and you kind of, you know, you're kind of trying to push innovation forward, I think it's hard to do remote, it's a hard problem to do, kind of, when you do not actually sit together and can kind of use a wipe order. Yeah, I think it's definitely easier, isn't it? Like, if you're fast, small agile starts being together and collaborating every day, it's easier to do in the office than obviously. Yeah, I think, I think big companies can probably like maybe like forward like one team on something and put them in some specific location together, right? But we try to hyper it. I think this is kind of also the kind of the energy in the office, it's just better when people, like you actually have a group of people that kind of work together, and it just gives more like kind of your collaborating, I think. Yeah, I always say like, you know, different styles can work, but it's whatever works for you. And also if that's your iron, then the people you're hiring, well, if they want my environment, then great. If they want another environment, then it's probably not the best place to be. So it makes it more simple than hybrid sometimes, I believe. Yeah, I think that's true. Of course, we have the issue of then we basically need to relocate people, and of course, that's of course a little bit of a hazard. I think we managed to attract some really good people and also from many different countries. I think companies are super nice places to live, so it's also attractive for people to live there. Absolutely, absolutely. To how many members of the company do you have now? Staff wise. So we're about 20 now, and I mean, steadily growing. But I mean, we try to be, I mean, everyone will say that right, but I mean, we try to hire great people right now. I mean, it takes a little bit of time to find the people who are like, as we thought about, like, good fit, and kind of can work on the things you need. It's very impressive that a number of people, for everything you do in actually, but I guess when you have fully automated environments, it does allow you to do more of what I can. I think we only do stuff we believe we can automate. Of course, that means that we can focus on specific things, but I think that's the good thing and right, you develop something, you put it into production, of course, you need to manage it, but you don't actually need to run it, like, operate in the same way as you would, and I'm kind of manual trading this. Awesome. We're a lot of customers. Really appreciate your time. Love this chat. It's been super interesting, or show the audience, or think it is so, as well. So yeah, appreciate your time and have enjoyable rest of your summer. Thank you, and thanks a lot for proposing. I think it was also interesting. I mean, I have actually not done any podcast before, I think. I'm actually so excited about it. Sounded on the other, like, on the museum. I'll keep a lot of people to see how that turned out. Sure. Thanks a lot. I hope so.