SDS 008: Data Science in Computer Games, learning to learn and A 40M Euro Case Study

Podcast Guest: Ulf Morys

October 29, 2016

Welcome to episode #008 of the SDS Podcast. Here we go!

Today’s guest is Ubisoft Finance Director Ulf Morys
Working in finance at a video games company can lead a curious mind to explore the realms of data science. This is exactly what happened to Ubisoft Germany’s Finance Director Ulf Morys!
We discussed so many interesting ideas, from Principal Component Analysis, to learning how to learn, to the uses of big data and data science in video games.
You will learn about what led Ulf to explore R and Python, and how and why he came to prefer Python.
You will hear Ulf share stories from his early career, including the clever way he managed to save his company €40 million through applying data analysis!
Ulf and I spoke about what he sees for the future of data science, and why you should learn more about the fields of unstructured learning and unsupervised methods.
Tune in to catch it all!
In this episode you will learn:
  • Data science from a financial background 
  • Principal Component Analysis 
  • Learning how to learn 
  • Data Science in computer games 
  • Excel liberation: R vs. Python 
  • Visualization: MicroStrategy vs. Tableau 
  • Case study: Saving €40 million through data analysis 
  • Using Data to make predictions 
  • Proactive vs Reactive analytics 
  • Structured vs unstructured learning 
  • Supervised vs unsupervised methods 
Items mentioned in this podcast:
Follow Ulf
Episode transcript

Podcast Transcript

Kirill: This is episode number 8, with Finance Director at Ubisoft, Ulf Morys.

(background music plays)
Welcome to the SuperDataScience podcast. My name is Kirill Eremenko, data science coach and lifestyle entrepreneur. And each week, we bring you inspiring people and ideas to help you build your successful career in data science. Thanks for being here today and now let’s make the complex simple.
(background music plays)
Hey guys, welcome to this episode of the SuperDataScience podcast. I am super pumped about this session because you won’t believe who was on the show with me today. Today we had a very special guest, Ulf Morys, who is the Finance Director of the German branch of the game design, development, and distribution company Ubisoft. Now I’m sure you’ve heard of Ubisoft before. They are the maker of such game brands like Watch Dogs, Assassin’s Creed, and Far Cry. So as you can imagine, this is a huge company, and it’s a huge name in the world of computer games. And I was actually very surprised to learn that Ulf Morys, who is the Finance Director of the German branch, is actually one of my students! It was such an interesting discovery, and you’ll learn more about how we met in this episode, but yes, Ulf is one of my students, and we instantly connected and I invited him onto the show, and he shared some very valuable insights about how he learns data science, and how he proactively uses it in his day to day role.
So you see, Ubisoft uses data science for lots of things, such as in-game analytics and monetisation. However, finance is not one of them. And Ulf found himself in a situation where he’s the Finance Director, but at the same time, data science isn’t used much. And through his own experience of seeing where the world’s going, of using the tools around him, such as Google, Amazon, or Shazam, he figured out that data science is a very valuable skill to have, and he started learning it on his own. Like a lot of our listeners, a lot of you guys, actually do. You’ve started learning data science just because you have that feeling that this is something so valuable, so important in the modern world.
And since he started learning it on his own, he actually then slowly started incorporating it in his role, and now they’re building this whole data landscape at Ubisoft for their operations side of things, and they’re going to be using quite interesting tools in order to drive these things in the world of finance that previously didn’t have this data backing. So a very interesting story of how Ulf is actually introducing these tools into his organisation, to that side of his organisation, and how he’s slowly building that data culture.
And in addition to all of that, in this podcast we talk about a lot of interesting things. We talk about learning how to learn, we talk about an information sharing culture, which is very important for organisations in our modern world. We talk about Excel and its fallbacks. Ulf gives us his view on Python versus R, and why he prefers Python. We talk about MicroStrategy versus Tableau. We talk about data visualisation, falling in love with object-oriented programming, using data to save millions. So Ulf, before his current role as the Finance Director of Ubisoft, he actually also had exposure to data science in his previous roles, and he shares a very interesting story of how he was able to save €40 million just by using data science, and that is definitely recommended, even if you’re going to listen just to one thing in this podcast, definitely listen to that. That is a huge success story, and it’s a testament to how data science can be so valuable.
We also talk about supervised versus unsupervised learning, structured versus unstructured analytics, and lots, lots more stuff. So you will get a lot of value out of this conversation. Super excited about this. Without further ado, I bring to you Ulf Morys, the Finance Director of the German Ubisoft.
(background music plays)
Welcome everybody to the SuperDataScience podcast. You won’t believe who we have here today. We’ve got Ulf Morys, who’s the Finance Director of the German branch of Ubisoft! Hi Ulf, how are you today?
Ulf: I’m very fine, thank you.
Kirill: Am I pronouncing your name right, Ulf?
Ulf: Absolutely. Absolutely.
Kirill: Wonderful.
Ulf: That’s because people always go like, is this a sound, or is this a name? But it’s a name.
Kirill: Ok. Great to know. So for those of you who don’t know, Ulf is actually a student on some of our courses at SuperDataScience, so I’m super excited about that, I was super excited to learn that Ulf is a student. Instantly invited Ulf over. It was actually quite funny how we met, because Ulf gave me some negative feedback about one of my tutorials, and how I messed up somewhere. Like a true German, he picked up the incorrectness. But hopefully, that’s all in the past, and now we’re going to have a lovely chat about how Ulf uses data science in his day to day life, right?
Ulf: Yes, looking forward.
Kirill: Wonderful. So Ulf, just quickly to get us started, how did you find out about my courses, and how did you start learning data science through our platform?
Ulf: You come naturally to it, basically. I signed up for a couple of newsletters, like from Kaggle and data science communities, and you see information about the possibility of doing online courses everywhere, and then I took a look first at Coursera, but in terms of price-quality ratio, and the possibility to determine the pace of your learning on your individual scale, I think Udemy is unsurpassed.
Kirill: Yeah, I totally agree. Coursera has its own merits, and I’ve actually done a course on Coursera myself. But I found that it’s a bit more archaic type of education, like Udemy’s more flexible, and these new learning platforms are more flexible in terms of how quickly you can absorb the knowledge. So I totally agree with you on that. And now let’s get to know a bit more about Ulf. Like can you tell us a bit more about what you do in your day to day role? 
Ulf: As you said, my official role is Finance Director of what we coin a “distribution subsidiary”, which means the company I’m working for, we are distributing the Ubisoft video games in the GSA region, GSA meaning German, Switzerland, and Austria, and we also do have on the financial level the responsibility for Central Europe, which includes Poland, Czech Republic, Slovakian Republic, Hungary, and so forth. That’s the major part of what we’re doing.
Kirill: That’s fantastic. So you actually are in the computer games industry. I remember back in the day, when I played games, I think — was it Morrowind? Is that an Ubisoft game?
Ulf: Yes, definitely. And it was distributed. It was not Elder Scrolls. It was the publisher. It was the publisher, but not the developer.
Kirill: Uh huh. So that was the publisher, not the developer. Ok. And so Ubisoft is the developer?
Ulf: Nowadays, we mostly distribute our own games. It’s really like the best analogy you would have is with film studios. It’s film studios who do their own productions, and those who just distribute foreign movies, for example. And nowadays, we have huge studio operations, like in Canada, in France, and in China, and everywhere. So they produce the games, and that’s what we primarily do nowadays. And we do not do what’s called in the industry “third party distribution” so much any more.
Kirill: Alright. And so what’s the latest hit game that’s coming out this year for Christmas?
Ulf: Oo, definitely — very close to the subject of data science, by the way, I mean without wanting to go into illicit commercial activity here, or marketing activity, but it’s Watch Dogs 2, which has a major subject vulnerability of smart cities.
Kirill: Oh wow, interesting! That’s very interesting. So for those of our listeners who are into computer games, check out Watch Dogs 2. That’ll be your little shout out to Ubisoft for lending us Ulf for a few hours here! Wonderful. Alright. Well how did you get into data science? Let’s talk about data science. Being the Finance Director, I’m assuming that that’s a bit separate to the data science division in Ubisoft. So how does your role actually relate to data analytics?
Ulf: One of the reasons why I’m so passionate about the subject is very simply the fact that the finance guys within each company usually consider themselves sort of the Lord of the Figures. The funny thing is, in the classical financial departments, which I would define really as accounting and managerial controlling and planning, people are completely oblivious, for the most part, to the advances in data science. In our company, I look into the use that we make of data science and machine learning, it’s primarily in the production units, which means the studios, and there you have some enthusiasts who are using neural networks, or other machine learning approaches, just to optimise games, or, I need to admit it also, the monetisation. But in the “financial space”, people have not, for the most part, the slightest idea what is possible with these sorts of new tools and approaches! Using a lot of digital technology every day, because your question was how did I get into science? On one day, I use Google, Amazon, Shazam, and — well, you name it, some other more digital services. And suddenly I came to realise I didn’t have the slightest clue how the mechanisms behind them worked.
I said, come on, this is revolutionising the way I’m living! I’m using it every day, I don’t have a clue, and that’s usually a situation I’m not very satisfied with, so I started to dig in, and well you very quickly, when you’re asking yourself the question, how is an Amazon recommender working? I mean, first of all, you need to find out that it’s called a recommender system. Once you’ve found that out, then you go, ok, but how does a recommender system work? Then you go like oh, Bayes… Native… Probability… you go, oh damn, I had this during studies some time, but how is it actually applied? Then, after a while, I just stumbled across a book that wrapped it all up. I read the book, and I’m hooked ever since.
Kirill: Oh, wonderful! What was the book?
Ulf: That was “Data Science for Business”, which was already, I think, quoted by Hadelin on your podcast?
Kirill: Yes.
Ulf: And it’s really an excellent book, very simply.
Kirill: Ok. Wonderful. So that’s a very interesting story. So it’s data science wasn’t part of your background, which we’ll get to in a second. So data science wasn’t actually part of your background, and part of your role, it’s something that you more realised that that’s where the world’s going, that’s how all the tools that I’m using operate, and I just want to know about it. And so since then, since you’ve learned a bit more about data science, you’ve read the book, you’ve started studying data science, has that in any way helped you in your role at Ubisoft?
Ulf: Absolutely. The knowledge I’m gaining from learning data science has so many connections to the daily lives of a “finance guys”, there is at least so many questions where I usually in the past would have said like, “you don’t have the tools to approach this”, or “it would be interesting to get an answer, but I really don’t know how to attack this sort of question”, and suddenly you go like, “now I’m not reasoning in terms of obstacles, but in terms of solutions, basically”, and say like, “ok, we need to dig in deeper there,” and “where do we get the data from?” And, “can we improve the data quality so that I can answer this and this business question in a more mechanistic way, because I think that’s what data science is about, it just mechanises our relationship with data.
Kirill: Very powerful sentiment there, that knowing data science and learning about data science has structured and altered the way that you think about business problems, that there is so much data, like I understand what you mean, there is so much data floating around, not just in our lives, but also in our day to day work in the organisations that we work in, and sometimes you get confused, or you stumble upon it and you don’t know what to do with it, how to approach it, how to extract certain insights. But knowing more about data science has helped you to understand how to tackle these problems when they arise. So yeah, that’s a very powerful sentiment there. I think a lot of our listeners might be finding themselves in similar situations.
Ulf: Yeah, and especially in, let’s say, not in companies which are “more traditional”. Obviously you depart from the assumption that in companies like Facebook or Google, you have very dedicated patrons. For example, in our company, it’s just happening now, really. You have what I would call or refer to as “islands of competence”, where people, based on personal enthusiasm, have built up some know-how, but now suddenly, the organisation starts to realise this can really be put to good use.
Just to give you an example, we just had a strategy meeting where we discussed about the challenges we are going to face within our local context here of this little subset. So it was not like Ubisoft overall. Our product, at its very core, is digital. It’s a video game. But all the processes around it, the selling processes, the marketing processes, everything gets digitalised. We need to get more knowledge out of this data because, very simply, it’s there and if we don’t use it, our competitors will. So one of the outcomes of this meeting was, we set up a task force to really map out our data landscape. Because everybody knows a little bit, but you don’t have really a map that by itself already will allow more people to ask business questions based on data. Because, very simply, if you don’t know something is there, you can’t ask a question about it. If you know the data is there, then you can suddenly start asking questions about it and try to derive knowledge from it.
Kirill: That’s very interesting. And what you mentioned about digitisation actually reminded me of a good book that maybe some of our listeners have already read, because I’ve recommended it before. But still it’s a very powerful book, it’s called “Bold”, that’s just B-O-L-D by Peter Diamandis, and it speaks about the world we live in, and where it’s moving, and he’s got a rule of 6 Ds: digitisation, democratisation, decentralisation, and 3 more! I don’t remember them all off the top of my head, but yeah, very powerful, and also very similar sentiment that you’re developing this data landscape, he’s got some comments about how organisations can leverage data. So that’s a good book to check out.
But I would like to ask you a bit more about the data landscape. That’s a very interesting concept, and it hasn’t come up in our episodes previously on the podcast. Could you tell us a bit more, how is the data landscape structured, and what does it entail?
Ulf: To give you one example, which is unfortunately not a good one in the end because I still haven’t come to a final conclusion on that, but it’s like in most larger companies, the data landscape is structured around enterprise resource plannings, or the big B moves like especially here, where we’re using Peoplesoft for example, or most of the time, it’s — what’s the name again? The German company that everybody’s working with. SAP! Yeah, right. Those are really, you would say nowadays, old systems, legacy systems. Yet, the information in there is, on the one hand, it’s incomplete, and even what is in there is not fully exploited.
To give you an example, we are selling to a lot of customers, which is sort of funny, because usually in asset distribution stuff, you’re selling just to what you call “key customers”, something like a Game Stop, or a Tesco, or whatever, and it’s just one customer. Your number of items you are creating, so builds, your [16:41] and things like this, is rather limited. In Germany we have a different market structure, we like it very decentralised. So one of the big market players here is completely decentralised, so we are in a position that we have much more invoices, and we should have much more information. Because we go directly with more retail stores. So we have a direct relationship with every retail store, and not just with a key buyer. And nevertheless, the only information we have in our official systems about these customers is basically their billing address. And then you go like, why don’t I have a more systematic approach to relay the figures or the revenues I generate with a customer with data about the customers? But I just don’t have the data about the customer! I really just have a postal address. And then I go to our sales department, and they tell me, that’s a good store because it’s well-located, and it’s in a demographic area whereby a lot of young people that like to play video games, blah blah blah blah blah. All of this information is not collected systematically, so I could not do sort of a cluster analysis on stores, and say like, ok, which are the stores which sell according to expectations above or below par? I can’t do it because I don’t have the data.
This is where I would say ok, this is basically a terra incognita on my data landscape. It’s a white spot on the map. And I try to fill it, and approach the sales department, and say like ok, can we collect more information on the customer, both on a qualitative and quantitative level? An example would be, how many square metres does the store have? How many square metres of that are dedicated to video games? Qualitative data would be, what’s the attitude of the multimedia buyer in the store towards video games? Is he a fan? Is he not a fan? What’s his attitude towards Ubisoft, as a publisher for example? If I had this sort of information, per store, and let’s say we served something like 2000 stores, then you get some statistically relevant data, and you could start to really do some cluster analysis, and say for example, principal component analysis. Which are the factors which really have an influence on the revenue, or on the return behaviour, or whatever, of the customer? Right now, I can’t undertake the question because I don’t have the data!
Kirill: That’s a very surprising suggestion. I would think a company like Ubisoft, the magnitude of the company, that all those processes and data structures would already be in place. But if they’re not, it’s actually good news. It means that there’s so much more room for development, so much more room for improvement. That if you get that data landscape in place, then you’ll have a completely different playing field next year, or whenever it’s ready.
Ulf: Exactly. But then again, I need to say one thing in defence of our organisation! Obviously behind everything there’s also a good argument. And the good argument is, if we do it in Germany, it just doesn’t scale to the international level, because the others just don’t have this information because they just sell to one central customer.
Kirill: Yes.
Ulf: So there are always two sides to the matter, and we have this particular situation. It’s just we don’t fully exploit it yet.
Kirill: Definitely. You’re at an advantage in that sense. Alright, and just one question. You mentioned principal component analysis. That’s a very interesting topic, and I’ve actually used principal component analysis on one of my projects when I was back at Deloitte dealing with customers and dealing with distribution centres. Just without going into any technical details, for the benefit of our listeners, how would you, in layman terms, explain what PCA, or principal component analysis, is, and what it does, for such projects?
Ulf: Again, as I’m not a data scientist, I’m currently working from two ends. I’m attending classes like yours, trying to get the theoretical knowledge. And on the other side, I’m still on the data collection, data managing part. It’s my plan, or on my planning horizon for the next year, to put it all together. So I can only repeat now on PCA what I basically learned. But I haven’t really put it to actual use yet in my daily role, but it’s taking a lot of data with a lot of features, and it’s basically like condensating data. You just try to boil it down to where it really matters on your output variable that you’re interested in. And some features of the data you are having are just more important than others, just have a bigger impact on the outcome variable, and you keep those, and you discard the rest. And this is like what I would call “condensating data”.
Kirill: Beautiful. I love that summary. And hence the name “principal component analysis” right? So you want to find out what has the main impact to really focus on those features rather than everything that you might have in your data set. Beautiful. And so moving on to some questions about your background, and what you studied, and how you got to where you are, can you tell us a bit more, like what you studied at university, and what was your journey towards becoming the Director of Finance at Ubisoft?
Ulf: I see one similarity with some of your previous podcasts, because there seems to be one common denominator, which is an engineering background. Because even though I’m now working in finance, by training I’m what is called in German “Wirtschaftsingenieur”, which is basically to be translated to economic engineer, and that means you are studying at the same time engineering with an MBA. So you do finance and engineering at the same time. I did mechanical engineering, something I have not worked in this. But you learn how to learn. And that’s the most important thing.
Kirill: Yeah, I’ve never heard of an engineering plus finance degree. It sounds like a very intense mix of knowledge. But I totally agree with the sentiment that it’s all about learning how to learn and you know, I studied physics as well. I don’t do much physics right now, but that degree, because it was so challenging, it helped me learn how to learn and how to retain this knowledge, how to find the most important bits of information and actually leverage them, understand what’s going to help me learn. So that’s a very powerful skill to have. I think there’s a quote, I think it was Einstein who said that learning is not about—or education shouldn’t be about just memorising facts, it should be about learning how to learn and something– I’m paraphrasing. But, yes, it’s a very powerful concept.
Ulf: And the concept of this particular curriculum is also very nice because they tell you basically already during your first semester how you always have the edge in any organisation afterwards because you have the knowledge both from the financial part and from the engineering part. It’s like—okay, if you are with your back up against the wall against the financial guys, just ask them how it works, and if you’re with your back up against the wall against the technical guys, just ask them how much it costs.
Kirill: So, if your back’s against the wall against the financial guys, ask them how much—how it works and then technical—
Ulf: Yeah, exactly.
Kirill: Okay, good. Good. That’s a fun one, yes, ask the opposite questions from what they’re good at. Next question I have for you is, so we talked about your background a little bit and how you got into data science. You mentioned that project that you are working on for the data landscape. Are there any other ways that you can share with us that Ubisoft—maybe not specifically in your department—but how Ubisoft uses data science in the organisation?
Ulf: Yes. Because I’m very curious, and luckily at Ubisoft, they really foster a culture of information exchange. We do have something like an internal social network and there is a group on machine learning and data science, and suddenly you get in contact with a lot of interesting people, and there is no problem if you just pick up a phone, or the messenger, and call these guys up. And as I’m interested in the subject, obviously, I got in touch with a couple of people who are using data science more, as I said, on the production side. And this is also, by the way, where we have really the big data. Because you can imagine if you have online games which have several hundred thousand concurrent users and everybody leaves a trace in the game some how.
It’s like how much time did he play, the lengths of the session, the starting time of the sessions, what did he do, did he finish any levels, and so forth. They really have a lot of data, and they start to really work with it. So, if we really have a data science specialist right now within the company, they are definitely more on the studio side. Whereas in the distribution side, I would always label what we have “smart data” because it’s just simply not as big. They have the big data, and to come up with some concrete examples that I stumbled across and I talked about with people, I think the most interesting one I found so far was actually an Android game, so a mobile game, and they used the past data to train a neural network to basically see the propensity to buy of a customer. Because it’s usually what they call in the industry a “freemium model”. So, basically it’s free but if you want to advance quicker in the game play you need to pay in-game items, or buy in-game items. That’s the way how you monetise these sorts of games.
Then it’s obviously very interesting to find out what are the patterns of typical customers which later on bought something and can I identify this pattern with current users, and in this case I can approach them with an offer they cannot refuse, sort of like “Get 100% more in-game gems today for the same price!” And that’s exactly what they did. It did work well. Actually, they started out a little late, so unfortunately the adoption curve of these mobile games is very steep. That means you have a lot of users at the start, and once you fall out of the top 10 on your local Android store, or the iTunes store, or wherever your game is offered, usually the install rates just fall down very steeply. So, they could only apply this now on lower numbers. So I would take it with a grain of salt. They obviously said like “Yeah, okay. We pushed the adoption rate or the buying decision by plus 75%,” which is impressive, but if the total number of cases is in the hundreds, and not in the thousands, I always go like “Hmm… Yeah, but is it statistically really relevant?” But now they have the positive part. I mean, it might sound too negative but the positive thing is now they have the know-how to integrate it at an earlier stage and then we get valid data.
Kirill: Yes. So that was like a training exercise and now for the next game they can start earlier. And that’s pretty cool. That’s like Big Brother but in the gaming world, right? So you’re playing a game but you’re actually being watched by a machine learning algorithm that is then averaging these out and producing some insights for the developer. That’s really cool.
Ulf: Yeah, even though Big Brother sounds very negative. You know, it’s more like a servant in the background who tries to make sense of what you’re doing, and then predictively offer you with something that really suits your needs right now. So that may be a more positive way to put it. Actually, another example where they use machine learning which I found out is really within the game design. When they have battle phases, they take real players in, but it’s not really commercialised yet. They really measure where people fail on maps to improve map design. So, that’s really a game lap where they collect the data, where they invite players, they use the data from the battle phases and everything, and they map how people move through levels and where they fail and where they get frustrated. Then it’s really a decision of, “Okay, do I make it easier for people, or do I keep the level up?” Because you also need—I mean, that’s obviously one of the basics of game design. You always have a cycle of challenge, of learning, of catching up and then going to the next threshold of the next challenge you’re facing. You need to keep it demanding on the one hand side, but it should never become really frustrating. And for this, they also use data science methods combined with data visualisation because you also—they have sort of heat maps where they show where people run into trouble during game sessions.
Kirill: That’s really cool. I wish they had these methods back in the day when I was into computer games. I remember there was this game—I don’t know if you remember. You probably do. It was called Duke Nukem and in that very first version of the game there was like this part where you had to like jump on this box and then find like this bazooka to open a door with it, and it took me ages. I remember days and days and days and days I tried to get past that part and I’m sure if they had these machine learning algorithms back then, they would have found that a lot of players were getting stuck at the very beginning and they would have improved that part. Yeah, so I can totally appreciate the value of data science in computer game design now.
Ulf: And it’s not only in that. I’ll be giving you the second example just to make sure because I know that’s a critique that is very often voiced towards games companies. We’re just normal companies. We need to make money, period. It’s a fact. But it’s not all geared towards only monetisation, you know, like pressing the maximum amount of money out of people. But in the end developers are very aware that we only have the product which sells when it’s fun. So I would say the main purpose is really to—it’s really in the mission statement of the company, by the way, to present gamers with memorable experiences. So that’s the first degree, and when the odd finance guys like me come in, then it’s suddenly also about monetisation because it needs to be.
Kirill: Yeah, exactly, and it’s fair enough. Like, you can’t expect a company the size of Ubisoft, or for that matter of fact any company, to run without any funding because you develop a game, and if you don’t monetise it, where are the funds to develop the next game and to create those new memorable experiences?
Ulf: That’s exactly the point.
Kirill: I totally understand.
Ulf: Yes, but it always sounds different when you go to the community. When you look at community forums it’s like ‘You’re supposed to do it for the glory.’ Yeah, well, parts of.
Kirill: But we need to eat as well!
Ulf: That’s the point.
Kirill: Yes, exactly. All right, Ulf, now moving back a little bit to your exploration of data science, can you share a bit what kind of tools do you use when you’re learning data science, and when you’re slowly starting to adopt it and use it more and more in your role and day-to-day life?
Ulf: In day to day life, unfortunately it’s still 95% Excel in my usual daytime work which is really a drag. Why? I mean, Excel was a major advance in the way how to treat data, yes, but it’s conceptually flawed at a very basic level, which is it mixes up data and logic, and that works out nicely if you have small models. But as soon as you scale up and you just do the slightest mistake, it becomes a nightmare. So Excel—really, one of the interests for me of learning data science is finding ways how to liberate myself, I must really admit.
And the major tools I’m using are—it was first R, the first language I learned, and then for some reason—I know there’s a lot of discussion going on—I decided to drop R, and now I’m working with Python and all the connected libraries, especially pandas, of course. Because in the end it just looks more versatile to me and this style of thinking seemed to be more appropriate to my style of thinking. That was what made the decision up.
On the professional level, we used to work with Business Objects as a BI tool, and this has switched to MicroStrategy. And MicroStrategy, we’re still in the early phase. I got introduced to it just a couple of months ago, but what I value a lot already is, for example, the data visualisation. Data visualisation possibilities are, I think, sometimes underestimated. Because when you’re talking data science, science sounds so like, “Yeah, it’s formulas, it’s complicated and everything.” But in the end, you need to communicate the results. And if you talk to people who are not familiar with data science, a good visualisation can go a long way. And now MicroStrategy offers some new possibilities that we are already using which we did not have before.
Kirill: Very interesting. A couple of questions on that. So, I really like how you put it very succinctly that Excel’s main pullback is that it actually mixes data and logic. So in Excel, you can put in some information, like data, or you can put in a formula. And I totally agree with that, that a real, a proper data science tool shouldn’t allow you to do that. You have to have your data separate and your logic separate. But on the Python and R—so that’s the first question, probably, on this part – Python versus R – what exactly made you make the switch from R to Python? You said the mentality of Python fits your way of thinking. Can you elaborate on that a little bit, please?
Ulf: Yes. I somehow fell in love with this thing, like methods. Like you have an object. You do just one dot–you have the object and then you apply a message or something. It’s a very natural way somehow how I think about manipulating data. You have an object and you apply something to the object, and that’s expressed in the syntax of the language. I mean, it becomes more natural. It’s more naturally understandable for me reading the code that I see written out in Python than in R. It’s as simple as that. It might be totally different for somebody else. You know, it’s like somebody might find it easy to learn French and another guy would say, “French is impossible but I like Spanish.” I mean, both languages are very close but somehow you feel more inclined towards the one and not the other.
Kirill: Funny that you mentioned that. I know exactly a person like that who cannot learn French, and she’s been trying for years and she picked up Spanish in a matter of few months. Great analogy! And MicroStrategy—so, do you use Tableau and if you do, how do you compare—how would you say MicroStrategy compares against Tableau?
Ulf: Tableau is even more of a visualisation tool. I mean, Tableau—by the way, it’s through one of those contacts that I got in touch with through our internal social network. He gave us the opportunity to get some test licenses of Tableau, because usually standard edition, we as a distribution sub wouldn’t have a Tableau license. But if you ask for it and you say like, “Okay, we want to test something.” So we got a license three months ago. We looked into it. As I said, first we had business [indecipherable 36:47] definitely have brought an advantage for us.
I think Tableau is really great from what I’ve seen so far on the visualisation level. The only point is that MicroStrategy now brings so many possibilities with its standard edition that exporting something to Tableau creates yet another interface to run, export from MicroStrategy to Tableau, and then you need to make the trade-off. It’s like, “Do I get so much more visualisation possibilities with Tableau that it’s really worth going this additional step and having again to run some exports and get the data into Tableau from MicroStrategy?” And we currently do not see the value added anymore for using Tableau as a separate visualisation tool on top of what MicroStrategy offers as standard for us. But that might change when we come up with some new smart ideas. If we find a way how we can visualise something of importance and we can only do it with Tableau, then we will use Tableau. But right now, we can realise our ideas, albeit with less panache, so it looks more basic in MicroStrategy, but it will still do the job.
Kirill: Okay. So basically MicroStrategy offers you more than just visualisation and that’s why you’ve decided to go ahead with it?
Ulf: Yes.
Kirill: Okay. Sounds good. And next question is, what is the biggest challenge you’ve ever had as a data scientist or applying data science in your role, or maybe in your past experiences?
Ulf: As I said, in essence it’s new and I’m still trying to connect all the dots. There’s not an example that I can give from my current job because before I brag about it, I want to have real valid results.
Kirill: Fair enough.
Ulf: But I do have several experiences and I want to tell about one from a time which was data science before data science. Because now, knowing data science, I realised actually I was already doing it a while ago, it’s just that it didn’t bear that name yet. That is a very interesting story. Why? I was working for a very big German industrial engineering company at the time, and they were supposed to buy another company. The target was—or the objective of this planned merger was to diversify from production into services.
So that was a service company and a very big and very well-known consulting company – I don’t want to mention any names — they had written an expertise that this is the way to go and there’s a huge strategic value attached to buying this company. So the result of the due diligence that I got involved in was basically forgone. So it was a forgone conclusion we are going to buy the company or as Trump would have said, it was rigged. Yeah, it was rigged due diligence already. The point was that since it was rigged from the start, the due diligence team that was assigned to do the technical due diligence on the spot, was completely reduced from planned of like 10-15 persons to just 2 persons: my boss at the time–because it was one of my first working experiences–and me. Plus, our time window was totally reduced from six months to three months. So, we were just two people with 3 months to do the due diligence on the company. To make matters worse, the management of the company that was the takeover or the merger target, was also briefed, so it was like, “Yeah, okay.” You know, it was sort of like, “We need to have these two guys sitting there but don’t feed them too much information because we already know the outcome.” That’s exactly what they did. They literally—and it’s not a joke–we asked for data at the time, and they came up with huge piles of paper, printed out from the EIP system. So it’s like, “Here’s the data you asked for. Go analyse it.” We thought, “What the heck is this?” And within the given timeframe, it would never have been possible just to enter the data into any IT system and analyse it.
Now, this has nothing to do with the techniques of data science nowadays, but it shows how important it is to come up with clever ideas to get to the right data. Because that company at the time, because it was still the time of mainframes and outsourcing your EIP software to data centres, they had an external service company that they had outsourced their IT to. And one part of due diligence is also doing the contractual due diligence, so check the contracts with external service providers. So I innocently asked if we could have a quick chat with the guys they had outsourced their EIP system to, and they didn’t have any objections. And I had rightfully guessed that this company, since it was just a service company, was not briefed in the same way to not provide us with any information. And then I pulled the leg on the rank and file IT guy that was assigned to me, because we set up an appointment with the guys and they said, “Okay, it’s totally unimportant.” So they sent this rank and file IT guy. I connected with him on the social level, and then it was really a sort of social engineering, you might say, which was more important than the technical part, to really pull his leg and say, “Hey, come on. Can’t you do anything better than provide us with stacks of paper?” And he said, “Yes, of course we can!”
He provided me at this time with two or three floppy disks with the aggregated data from the last 2 ½ years of the company’s performance. And lo and behold, suddenly we had the data we were looking for and we were able to establish an analysis which in the end, to make a long story short, reduced the acquisition price of this service company from an initial something like €42 million, based on the expertise of this huge consulting company, to a mere €2 million. Quite a decrease! And we really deflated the strategic value that the consulting company had interpreted into the service company based on pure data.
Kirill: Wow! That is so cool. Like, you cut off €40 million from the price. That is incredible.
Ulf: It was really one of the major achievements in my professional life, I must admit.
Kirill: Wow! That’s so cool.
Ulf: It’s 40 million. You don’t save that every day.
Kirill: Yeah, for sure. And I’m certain a lot of people were upset about it as well, especially like those that were in the company that was the target of the merger. But I guess it’s—you know, you’ve got to do what you’ve got to do. You’ve got to do the job diligently and that’s a very good example of data science in the corporate wars, I’d say. That’s awesome.
Ulf: One or two other examples from that—for example, I also worked for a telco company. In a telco company, the most important metrics at the time—nowadays it probably has changed to MB of data, but it was minutes. Like, how many minutes did you do per day, how many breakout minutes, input minutes, whatever, you know, because that’s what you’re billing to the customers in the end – minutes of voice traffic. There, I also witnessed complete meetings, for example, between the finance and the technical guys going completely astray. Why? Because people—everybody meant to have the right data and they couldn’t agree on what the right data was. It’s like, “Yeah, but we had X minutes”; “No, we have a completely different figure.”
After a while, I got so upset about this because we never talked about the subject matter because people were already not agreeing on the data they were using. And then I tried to move my way up from both sides. I asked the technical guys, “Where do you get your information from?” I asked the finance guys that I was a part of, “Where do you get the information from?” And in the end, I always ended up with just a couple of real IT guys who were managing the database on the switches.
So the switches are the heart of switch network in the telecom company, and these guys were completely overworked. When I approached them, they were ready to shoot me! “Not another guy asking me for information about how many breakout minutes do we have to the U.S. in the last blah-blah-blah!” We had a tool, what they call—I don’t know if you’re familiar with this—an OLAP cube, an online analytical processing software that we used in finance for controlling tasks. But we had the license, we could use it for other purposes and nobody had thought about his. I asked the guys, “Hey, come on. You’re totally overworked.” And then—nowadays you would call it the idea of self-service analytics. I told them, “Let’s systematically break down what’s happening in our network. Every day you do the SQL requests to your switch database. Once a day we feed it into the OLAP cube and the OLAP cube becomes our single source of truth.” And they were going like, “It doesn’t sound as too stupid as an idea.” And we gave it a test and it was not a huge project. It was really under the radar screen of the upper management of the company. And it worked! And the funny thing is, after a while everybody used this data because it was the only reliable data that everyone could agree upon.
Kirill: Fantastic! That is so cool. That’s another great example of data science and being innovative about data science. I’m sure those engineers were so happy that you saved them from all these incoming requests of—every day they would get all these requests and now people just use the OLAP cube.
Ulf: And this is what I refer to and this is where I see the beauty of data science. Nowadays, it’s a very good example of what I call mechanising data. The job got mechanised.
Kirill: Yeah, exactly. And you basically answered my next question. I was going to ask what is your most favourite thing about being a data scientist or actually using data science and analytics, and from this I can only assume that it’s your mechanising data. Is there anything else you can add to that? What is—what are your most favourite things about using data science? What excites you the most?
Ulf: Yeah, mechanising is the one thing. The other thing is—it’s more the metaphor that you can use with visualisation, it adds colour. Analysing data doesn’t add more data. It just changes the way how you can look at it. For me, the metaphor is really like going through the transition from black and white movies to full-colour movies. You still have actors, you still have a screenplay and everything, and yet everybody would agree there’s nothing really fundamentally different, but everything has changed. That’s a good metaphor.
And the third part where I’m really intrigued about the possibilities of modern data science is to change the perspective from looking at data in a retrospective way into a forward-looking way. I mean, the metaphor you always find once you get started with this subject is always rear-view mirror or windscreen view. That’s one of the things where I still don’t understand what the finance guys are doing. As I said previously, controlling, accounting and so forth, they are still so much in love with their rear-view mirror, it’s incredible. What happened in the company, and from there we try to extrapolate. No, you can do prediction even now! And you can do it with more precision nowadays than it was possible 10 years ago. But they just don’t have the toolset.
Kirill: Love that analogy about the rear-view mirror and the windscreen. Also, it’s like reactive versus being proactive, right? So you wait until something happens and you react to it, or use data to understand what might happen, what’s more likely to happen, and you take proactive measures to that and therefore you save yourself money, time and efforts. I definitely agree with that sentiment.
Ulf: And then, as I said, suddenly, once you have an idea about the tool sets and what you can do with the data, there are so many creative ideas that suddenly spring up, that come to life for your daily operations. Coming back to my previous example, saying we’re still lacking the information, or qualitative and quantitative information, about our customers in this Ubisoft distribution case. For example, if you have this information, you could do some predictive analysis on, for example, where does it make sense to invest money in marketing measures, in trade marketing measures. Sort of like, “Trade marketing works with this store but it doesn’t work with this store. Why don’t we give him more money than to the other guy?” Nowadays it’s just based on gut feeling.
Kirill: Yeah, yeah. Data beats gut feeling 10 days out of 10. And you definitely have this vast experience in data even though you’re not in a data role per se; but you have all this previous and even current experience with data. From what you’ve seen and what you’ve learned about data science, where do you think this field is going in the future, in the coming years?
Ulf: Tools will get easier to handle, so there will be a sort of democratisation of data science, certainly. So there’s a huge chance also for the vendor companies, like you mentioned Tableau. And what I personally got very endeared with digging into data science was, of course, open-source software because both R and Python are open-source. And I just find it incredible, the quality of support that you’re getting when you have questions, for example. It’s just amazing! So I think data science will also advance the course of open-source software to a certain extent.
But to answer your question then: Where do I think would we head with data science in the near and medium future? In the narrow professional sense, or really concentrating just on finance, I would say it will really help to, what we said previously, permanently separate data from the data processing logic to get rid of this mess that Excel is making when you take it to a bigger scale. It will help to combine more systematically the use of data analysis techniques and data visualisation for really the discovery of patterns and rules in your data, which is like my previous example of what I said, do some prediction, sort of like, “Where do I invest my money?” “Is this customer better than this customer?” And I have more than just the gut feeling to base my decision on.
Ultimately, for me, because I think it’s really the most important part of the traditional finance role in an operative company, not on the corporate level, though there too, but it’s still different, but on an operative level, the planning process is really important because you always—I mean, you need to know where you want to be tomorrow. And integrating predictive methods into concrete planning of costs, and especially revenues, will become paramount. If you really focus on the financial function, I think this is where data science can really improve things and will play a major role.
But if you take it to a larger sense, data science, the most intriguing field right now, and I don’t think you can make a huge mistake learning about it, is unsupervised learning. Because again, I only have the theory really, but the difference is, of course, with supervised learning, you need to know the outcome. You need to have a training dataset. And with unsupervised learning, you just learn structures from data by itself. So it’s sui generis, as they would say in Latin. So from your own original data, structure emerges from nothing. And the very simple fact is that as soon as you leave the field of structured data, EIP systems, and you enter the field of big data from resources as disparate as, let’s say, Facebook entries, Tweets, feedbacks from website commentaries and so forth, you just end up with so much more unstructured data and any tool that will allow you to have structure emerge from initially unstructured data, will be enormously important and helpful.
Kirill: Wonderful! I love that analogy about structured versus unstructured, and also the importance of learning about unsupervised methods. I also think that’s a field where you cannot make a mistake. If you start learning about unsupervised learning and mastering that field, then that is where the world is going with AI and with—that’s how we as humans operate, not always do we have a test dataset or some test information or training information that we can learn from and then apply that. That definitely is part of life but otherwise we often come across situations where there’s something absolutely new where we have to make decisions on the spot. And machines are also going to be like that, so unsupervised learning, even though I find it’s more of a complex and challenging field, it’s definitely something that aspiring data scientists should consider looking into.
Ulf: Yes, and I can only advise it because it’s just a hint, because just like you said, I mean, it’s really challenging. Even if you’re into mathematics and statistics and you brought yourself up to speed, the conceptual basis is still difficult to grasp for me. But my hint would be to go to YouTube and look at some of the presentations of Geoffrey Hinton. It’s just pure bliss! Because this guy, he’s obviously one of the driving forces in the development of all this on the neural network level. I don’t fully grasp all the theory yet, but I’m starting to connect the dots. But just listening to this guy in his posh English accent, you suddenly realise you should make it mandatory that all scientific lecturers should have this accent. 
Kirill: Yeah. Definitely, that’s interesting. I haven’t seen Geoffrey’s videos myself but I will definitely include a link to his YouTube channel or videos in the show notes. And speaking of aspirations and kind of wrapping up our episode today, what other aspirations do you have that push you to learn more about data science, and maybe that you can share, and some of our listeners can take on as well?
Ulf: My main aspirations are actually satisfying my intellectual curiosity and understanding what’s going on, what kind of new mechanics influence my daily life. For my personal career, I have to admit that I’m so much into finance that I don’t really see my career as a future data scientist somewhere because I’m too “settled” for that. But I think if I can still contribute to make my company truly data-driven, and really contribute to the fact that Ubisoft puts these new tools to the best use for itself and for its customers, then I’m perfectly satisfied. On the level of personal learning, as I said, I totally recommend using Udemy on the one hand side, and then I might consider doing Coursera course within the fixed timeframe as a sort of cross-validation, sort of, “Did I really understand what I’ve been learning on Udemy?” by checking if I can very quickly pass the Coursera course. So that’s about my aspirations.
Kirill: That’s been very interesting learning about your past experiences and the work you’re doing now. I’m sure that you will find ways to apply these data science methods in Ubisoft as more time passes, and as you slowly build out this data landscape and find ways to leverage upon that. If our listeners would like to contact you or even just follow your career and find out more about you, how can they do this, where can they find you?
Ulf: In this podcast, you’ve seen my name spelled out. It’s pretty unique. So if you just enter my name, you’ll immediately find my LinkedIn profile for, let’s say, more the Anglo-Saxon part of the world; Crossing or Xing for the German ones; Viadeo for the French ones. I do have those profiles and that’s enough to get in touch with me, I would say.
Kirill: Okay. Fantastic! We’ll definitely include all those links in the show notes. And one final question for you for today: What is your one favourite book that can help our listeners become better data scientists and analysts?
Ulf: As I mentioned before, it’s “Data Science for Business” by Fawcett and Provost. Definitely an eye opener!
Kirill: Wonderful, thank you. We’ll also include that in the show notes, which are available at SuperDataScience.com. And thank you very much, Ulf, for your time today, I really appreciate it. And best of luck with your new game releases for these coming holidays.
Ulf: Yeah, we’re looking forward to that as well. Thank you. All right, and have a great day in Brisbane then!
(background music plays)
Kirill: Thank you very much. Have a great day in Dusseldorf. So there you have it. That was the Finance Director of the German branch of Ubisoft, Ulf Morys. And I cannot thank Ulf enough for joining me on this session of the SuperDataScience podcast. As you can see, it was fully packed with value. It would be especially valuable for those of you who are just starting out into the field of data science, who don’t have a background in data science. Even if you don’t do data science in your role and you don’t classify yourself as a data scientist, you can still see from this podcast how you can leverage the tools of data science and find ways to improve the work you do by using analytics and by introducing analytics. There’s always a way, there’s always a place for analytics in whatever role you perform, and whatever organisation you’re in because data is all around us, in our day to day lives, and in our professional careers.
So, for me personally, the most valuable parts of this podcast were, of course, how Ulf saved €40 million just by applying data science to a very complex challenge where he was faced with lots of data, and he had to do some quick analytics, and without data science he wouldn’t have been able to save that huge amount of money for the company on that merger. Also, it was very inspiring to get a bit of an insight into how computer game designers use data science in the different areas of the business, whether it’s in-game analytics or it’s monetisation or, as Ulf has shown us now, in the finance world, where he’s slowly introducing a data landscape and how they’re going to be analysing different stores to maximise their relationships with their partners.
And finally, it was just very inspirational and very exciting to see how Ulf has slowly learned data science, has found this pathway for himself, has created this pathway to learn data science and enhance his career. I think it’s a great thing that we have online education platforms where we can pick up these skills, whether it’s Udemy, whether it’s Coursera, the SuperDataScience platform where you can come and just learn these skills for yourself at your own pace and then improve your knowledge and get better and better.
So, there you go. That was episode number 8 of the SuperDataScience podcast. You can find the show notes with all the links and materials at www.www.superdatascience.com/8, so just the number 8. And if you’re listening to this podcast on iTunes, make sure to like and rate this podcast and leave us a review. That would be fantastic if we could get some of your feedback to get this podcast into the rankings of iTunes. Thank you very much for your attention, and I can’t wait to see you next time. Until then, happy analysing!
Show All

Share on

Related Podcasts