Evan: Hello and welcome to the Mining Your Own Business podcast. I’m your host, Evan Wimpey, and today I am very excited to introduce Francisco Rius. Francisco is the Head of Data Science and Data Engineering at Minecraft. I think probably many of our listeners have heard of Minecraft. And it’s actually a funny story. We’ll get to Francisco, he’ll introduce himself in just a minute. But this was very interesting. I shared an episode on LinkedIn a few months ago and had a listener of the show, Marcus Mann—thanks Marcus. Here’s your shout-out. He mentioned, “Hey, you guys are called Mining Your Own Business. Why don’t you have a Minecraft episode?” So I actually reached out to Francisco. I actually heard him on another data podcast, the Data Storytellers podcast last year. Really good show. And Francisco and I chatted a little, and he agreed to come on the show today. So a meeting of the mining puns here. Francisco, very happy to have you on the show. How are you?
Francisco: Good, how are you? Thank you for having me. This has been a long time in the making as you point out and thank you for whoever tagged us in LinkedIn. It’s destiny. I truly agree. There’s a very good fit here.
Evan: Perfect. Thanks so much, Francisco. Hey, to get started, can you maybe just tell us a little bit about yourself and your personal background? What got you into the role that you’re in today at Minecraft?
Francisco: Yeah, absolutely. So I have been in analytics and data science for a better part of the past 15 years. It’s really, it is really amazing because it honestly feels like yesterday when I started in analytics, and the industry has really evolved. I started in marketing analytics, and the way that I jumped into analytics is I was actually doing marketing campaigns and I was mostly interested in the statistics of marketing and using predictive models, and using those predictive models to send out—back in the day we were used to actually send physical mailers to customers and I wasn’t in video games. I was in subscription television. So did a lot of that for some time and jumped into analytics because I really wanted to be the one building the models and I thought that was really interesting.
I have an academic background on business economics and marketing. But I also have been very technical for my entire life. I used to code and build software when I was about 13, 14 years old. I used to sell software in comp back in the day, and that was my mind to buy PC gadgets right back in the nineties. I’ve always been fairly technical in the analytics area and that world really intrigued me. So I made the jump after a few years working in that intersection between marketing and analytics, I got a call from Electronic Arts and they were looking for an analyst to do forecasting for one up and coming area of one of their video games for FIFA. FIFA is a massive video game. Lots of players all over the world play FIFA. But back around 2012, there was a new digital mini section of the video game called Ultimate Team. So this was the early stages of microtransactions and monetization in video games and they needed an analyst to come in and forecast. And that’s how I made the leap to to video games. After that I started building teams. All of a sudden, Ultimate Teams started catching fire. A few years fast forward, the stock price for EA is 10 times higher than where when I started. So I probably should have invested at EA more aggressively. But also the business ultimate team became really big in the world of digital monetization. I became a director over time at Electronic Arts. Became owned multiple areas, so all analysis and data science for FIFA, NHL, UOC and a few other titles.
And about three years ago I started talking to the Microsoft team here at Microsoft and made the leap. I joined the Minecraft team in 2020 as a Head of Data Science and Data Engineering. And it’s been awesome so far. I started right before the pandemic, so that was interesting. But lots of change over the past.
Evan: Very cool. Very cool background. And I know attribution modeling is very tough. Maybe you did some of that in your marketing days. I wonder how much of that 10x stock rise was due to your work in data analytics there. I’m just gonna say a lot of it.
Francisco, so you mentioned there’s been a lot of change in the last three years at Minecraft. Minecraft is I think a little over a decade old now. I’m curious, you know, given that 10-year history and even in your last three years has the emphasis on data and analytics changed? More at the forefront. There’s obviously a huge community, a lot of players, a lot of gameplay, so there’s a lot of data to be had there. So has data always been a focus or has it become more of a focus recently?
Francisco: Yeah, it’s really interesting. I think data has allowed the organization to understand our players better over the past number of years. Minecraft, as you pointed out, has been growing for the better part, I think we’re, you mentioned 10, we’re probably around 12, close to 13 years at this point, which is pretty fascinating. You think about the longevity of typical video games don’t last for longer than maybe a year. Right. And now we’re talking about video games that are engaging players for decades.
And I think at the, during the early stages of Minecraft there wasn’t a ton of data being captured. There was obviously a lot of insights being gathered. Discussions with the community, for example. And the developer community here at Minecraft is actually super active on social media. And they interact directly with players. But if you think about those interactions, those are personal relationships, right? And whether we like it or not, that creates bias, that creates a way of listening that only captures information from a subset of our own personal networks and the people that we can reach through our own language and through our own networks on social media or maybe through some focus groups here and there. So lots of primary research and form of gathering up information.
And I know that our data effort started actually before I joined the organization. So around 2017 or so, Minecraft became really serious about using data to make the player experience a little bit better. The Minecraft switched over to, or released a new engine for the game. It’s called Bedrock. You probably have heard of it. When Minecraft released the iOS game or the pocket edition, that’s when Bedrock came online. And that created a lot more capabilities from a data science standpoint. And that was one part of the vision, right, to actually use data insights to be able to inform what are players doing in the game. As you pointed out, being online for the past 12 years or so, that has created a pretty big community. We are upwards of 120 million active users per month. So that also creates a ton of diversity, right? And how do we understand, how do we, when we have segments that if you think about a behavior that is enabled in Minecraft, then there are hundreds of variations of that specific behavior. We like mining blocks or chasing mobs or whatever it is. There’s somebody around the world that’s doing something slightly different in some way. And as data scientists, we try to understand the differences in behaviors and what attracts players to the game and how to keep players entertained and engaged.
Evan: Awesome. Yeah. Very, very cool. And yeah, my eyes are wide open at the size and the scope and the velocity of data. If you’ve got a hundred million players per month playing, who knows how wide, I don’t know what the average time is, but that’s a lot of data collection.
I do want to touch on, you mentioned some of your personal connections and data that you collect. Is there other data from outside of the game that you’re using now? I know you mentioned bias, but I’m thinking about, you know, there’s Reddit communities, there’s Twitter accounts that are super active. Some of that content is probably text. Some of that is probably harder to get at if it’s video, if it’s recording. I don’t know if there’s TikTok data that, you know, you can get at understanding the voice of the customer through that venue as well.
Francisco: Yeah. So that’s what’s fascinating, right? When you have a customer base of a hundred million plus, right? Then listening to wherever they, our players are voicing their opinion is, becomes really important. We don’t necessarily have to sit back and wait for a help desk ticket or something along those lines to really understand that there’s something happening in the game when it comes down, because there’ll be Twitter chatter and there’ll be posts on Reddit complaining and things like that.
And our data scientists are actually done a really good job at scaling up our social listening capabilities. But more importantly, a contextualizing listening. And I think that’s really valuable. It’s a little bit of a shift and I think the entire industry is moving in this direction, or a good chunk of the industry is moving in this direction. What it means is that we try to capture data where our players are interacting. Text data as you pointed. But also try to use advanced natural language models that will actually contextualize. To what’s happening in the game at any given point in time. Minecraft is a little bit different than other games, so the language that players will use will be slightly different, right?
Whereas in Fortnite or Call of Duty, you’ll probably hear, will hear a lot about, you know, arms and chasing and things like that. In Minecraft, it’s more about creativity and building and then being with your friends. So we have to contextualize our social listening, and that’s where our data scientists play immensely critical role in reducing the bias and in increasing the interpretability of the data that we capture. And then one last thing I will point out about social listening is that it also allows us to listen to a very audience two different languages. So we’re also starting to experiment with translation in near real time. Japanese, and there’s a lot of places in Western Europe that are very active right now in German and French. And our data scientists, you know, some of them are really, really smart and speak multiple languages, but they don’t speak every language in the world. So I think that’s, that’s the next frontier for natural language is really being able to interpret that in language and also context.
Evan: Yeah. Very, very cool. And you guys are at a, such a very cool space to do that, that you’re generating, or not generating, but you have a very real need to sort of contextualize that data in many different languages, sort of map the English language across to others in the same setting. So very cool.
I’m curious how much of this is, you know, it almost sounds like a research institute or like a university, the type of work. Is this work that you guys, that your team is doing to do this? Or do you leverage, you know, we’ve seen a lot of large language models that come up in language generation. There’s also a lot in translation. So I’m curious how much of this is Minecraft-led versus just finding what’s published and what’s available that you guys could use?
Francisco: Yeah, that’s an amazing call out. So we do partner very closely with a few universities in particular. We have been working with a program at Purdue University for the past three years or so. It’s called the Purdue Data Mine, and it started out as a small program where a group of students would come in with us for a semester and work on a project. And the background of those students is very diverse. They’re in multiple disciplines, computer science, psychology, economics, but they all have an interest in data science. So we’ve assembled these many teams that are super effective and they pioneer a lot of the social listening approaches for us. Obviously taking from academia.
We also lean into our own research resources here at Microsoft. So MS Research pioneers a lot of the listening at scale. So part of our listening actually comes from cognitive services and if there’s any data centers that are interested in understanding how we use natural language, they can use the cognitive services APIs that are widely available and try their hand at it. But it’s really interesting to see how algorithms sometimes get things right, but also when they don’t get it right, that’s the more important nugget of information to try to understand why.
Evan: Absolutely. Yeah. Very cool. And it’s very cool the partnership with the, if I caught the name right, Purdue Data Mine. We’ve got another mine that in the name. Have to get ’em on the Mining You Own Business podcast. You gotta play out this mine.
Francisco: Absolutely. And they would love to join you because it’s a very bright group of students. It’s just amazing to see what they can do.
Evan: Awesome. Very cool. Okay, so to switch gears a little bit from the social listening. One of the things that we find challenging within data science, not necessarily on the technical forefront, but once you’ve got some insight, maybe you’ve done your social listening and, hey, here’s, you know, the data suggests this would be an interesting feature or an interesting change. But you’ve also got game designers, subject matter experts that maybe have their own opinions. So, and I’m curious what the relationship is like there? Does Minecraft tend to be data driven? Is data appreciated in helping to shape the way the game, any changes in the game?
Francisco: Yeah, absolutely. Great question and it’s interesting, I think this is true for the entire gaming industry. This is an industry that is in essence a technology industry, but at the same time, it’s so creative in nature, right? Very few industries are in this position where you really have to balance out this piece around art and science. And we approach this in multiple ways. I think we now have seen probably a little bit of an evolution between the—you’ve probably seen a couple of generations over the past 20 years of game designers and game industry experts that are injecting and using data insights in their decision making. So one of our strategies release around looking for folks around the organization that have experience using data insights and partnering with data scientists and working with them as ambassadors, right? Once you get your foot in the door as a small team and establish a relationship and also proving a use case that becomes really important. A lot of times the relationship aspect of working in big corporations is multiplied times 10 when you’re working in creative industries as well. So that relationship aspect is really important and, we are very careful when we consider folks for hiring and bringing into our team, that’s incredibly important. Their ability to make friends and into the emotional intelligence aspect of working shoulder-to-shoulder with folks that maybe see the world in a little bit of a different light. That tension is also very exciting for us, right? Because we hire a lot of folks that have academic backgrounds, but also a lot of backgrounds in exact sciences. And they, a lot of times there’s a little bit of inertia where we want to create analysis that will tell us a black-and-white story, and in the spectrum of player behavior, that’s just not true, right? This, it’s almost impossible to get to a black-and-white story. So we learn a lot from our creative counterparts as well. So the partnership is the most important factor here.
And, and a lot of times honestly, it’s around proving the use case and being very consistent around the message. So if we have an observation or an insight that we feel very strongly about, we might not necessarily be able to get a feature to be implemented or to make a change in the game, even though we believe that the signals are pointing us in a certain direction. But being consistent in that observation over time and asking the teams to help us prove it out. Hey, can we experiment? Can we think of a prototype? Can we actually have a focus group and get some primary research on this? There are different avenues for being able to pull folks that may see be seeing things from more of a creative standpoint into this world of using data insights for decision-making.
And the last piece that I’ll mention is it also makes a lot of times our lives more efficient to use data, right? Because as we were just talking about in the case of social listening, we can listen at scale. I don’t have to read a hundred posts in Reddit to really understand what players are experiencing.
I literally only have to look at a data point to know whether players are happy or not. So I think the world of data can make our lives a little easier too.
Evan: Awesome. Yeah, that’s a very good take and I wanna broadcast that message when there’s some of the friction between the folks trying to implement the change and the folks using data to suggest it. Yeah. Trying to make things easier.
Francisco, I’m curious for yourself as well as your team. I usually think about sort of the data scientist and then the subject matter expert, but I would, I would suspect in gaming, there’s a lot of folks that are drawn, even in the data science world, that have a background in gaming, or at least our end users, you know, participate, you know, engage in the game a lot. So do you find that most of your team, or a lot of your team is familiar, spends professional development time engaged in the game?
Francisco: Yeah, absolutely. A lot of our folks are definitely gamers or at least understand the industry really well. And it’s fascinating what that can do to, you know, breaking barriers and building relationships with, as you mentioned experts or subject matter experts.
But going back to the people piece, I think this is a really, really important element, is we also look for subject matter experts in multiple disciplines to build a data science team that’s very well-rounded, right? So it’s not only the creative aspects and the product aspects that we have to consider. There’s so much more to publishing a video game and making it successful. Right? So we talked a lot about the marketing piece on social media. That’s a good example. Then you also have your finance teams and your business teams and your publishing teams and your sales team all over the world. So for us to be able to scale as a data science team in all of these different verticals, we look for people that understand their worlds and our philosophy, and I hope that you’re seeing this more often in your podcast and this has been true over the past number of years—I feel like data science and analytics teams are, are maturing to be subject matter experts alongside decision-makers. And we’re seeing a little bit of a reduction in the number of layers of translation that happens between the data insight and the decision. And I hope that over time you will see folks with very strong data science backgrounds and analytics backgrounds actually making decisions. There’s no reason why somebody that’s really, really creative can’t also have a data science skillset set and be making decisions about a game. And it happens quite a bit in games where optimization is more of an outcome, or more of the North Star. Like free-to-play games in mobile. You think about Candy Crush and a few others, where there are creative elements for sure, but there’s also big optimization levers and you cannot see product managers with strong analytics and data science skillsets.
So over time I think this piece of that, yes, sometimes there are parallels where you have a data scientist and a subject matter expert moving in parallel directions. Our push is to intersect those as much as possible.
Evan: Wow, that’s great. Yeah, I think that would do wonders and I think the, the more this sort of skill set and industry grows, the more there’s an opportunity to make that intersection. So I do appreciate that.
Just roughly for scale how big’s your team ballpark of data science, data engineering folks?
Francisco: Total, we have about 28 people or so between data engineering, data science and analytics. We’re super diverse in the sense of multiple disciplines. Our data science team, we usually specialize in tracks, so we have product analysis, business analytics, so on and so forth. And then on the data engineering front—and we also have an analytics environment team that builds data structures for the team. So we’re an end-to-end shop. Our team does everything from coding the telemetry events that triggers in the data groups that go inside the game so that we can ingest that into a data environment, analyze that, create insights, and actually turn that back to the game teams.
And then in many cases these days, we’re actually using a lot of those data outputs as inputs into models for prediction and feeding those predictions back into the game for recommendations. And that’s very powerful.
Evan: Awesome. Very cool. Yes. Real end-to-end shop in real time.
You mentioned, you know, talking about the intersection of sort of the game expert, the subject matter expert and the data scientists. So, maybe this is a question for, within that intersection is about the pace of change. If there’s a data suggests some insight, this change would appreciated this change would improve gameplay, would improve the customer experience. But, there’s some friction to adding changes. You’ve got a hundred-plus million users in a month. You can’t make sweeping changes every month. So is there a trick to trying to balance, you know, making improvements without pushing too much change that your customers can’t absorb?
Francisco: Yeah, that’s a really good question and a great observation is you, we have to balance out the essence of Minecraft with what we may believe are either acquisition or retention opportunities or some other opportunity that’s identified through data insights. So I think the team does a really good job at playing this balancing act and making decisions for the long term. And one of the big challenges that we face in the video game industry, you’ll see that it is very, very common to make short-term decisions because a lot of video game companies are very pressured to monetize very quickly, right? And they just need to make their Wall Street numbers. And there are levers to do that. You can use very aggressive predictive models. There’s also monetization models that exist in the industry that have been very effective at monetizing the short term. But the strategy for Minecraft really is: Hey, we want, we want to create a relationship with our players that it’s gonna last for the next 10, 20, 50 years. Right. And as a data scientist, how do you measure that? Because also our data sets don’t go back for 10 or 20 or 50 years, right? So what signals do we look for to understand whether our player, our relationship with players is becoming better or becoming worse? A good part of it also comes out of game information, right? So socialism is very important, but we also have KPIs and, and measures that we pay attention to over the long run. Interestingly our definitions of retention are actually a little less strict than the definitions of mobile games. A lot of times mobile game a lot of times will say, Hey, if I didn’t see a player come back in day seven, I’m just gonna consider that player to have turned and I’ll never see that person again. And I’m gonna run an acquisition campaign to bring in another player. And for us it’s a little different because we’re looking at the long range. So we try to really understand what windows of opportunity we have to re-engage with players.
So seasonality of the times—when players are more likely to come back, you know, school breaks and things like that. We focus our attention on some incremental changes, right? And again, balancing this piece around not being too disruptive, not pushing the game out of balance, and trying to understand what’s the right mix of features and treatments when the game can be updated. And it’s really, really important that we also hit players or we get our releases out to players in the right time when they can actually consume.
Evan: Ah, sure. Yeah. Very good point. And you think about the challenges, it is very exciting to hear a gaming team think about 10, 20 decades longtime horizons. It just speaks to the power that Minecraft is at this point. And I hate that I’m drawing parallels to insurance, but I had, Rob Horrobin was on the show from Pacific Life and we talked about some of the similar challenges and, you know, how do you evaluate models that are looking out 20, 30, 40, 50 plus years? And it’s very tricky. It’s very fun challenge to work on.
Francisco I wanna ask you if, you know, you’ve got a big team there at Minecraft, you’ve got a lot of smart folks in, you know, helping build this game, building out the huge customer base. People that love this game, let’s just say everybody is aligned to the Francisco vision and y you are the stakeholder and you say, you know, you’ve got these team of engineers and scientists and you can point them at whatever effort that you. What would be a big lift or be interesting or be something very cool for them to work on. Everybody’s aligned. Where do you want to point your efforts?
Francisco: There’s a lot of things that I will probably try to tackle, but I’m a big fan of prototyping and failing fast. And I think that we’re getting really, really close to a major point of acceleration in the intersection of AI and video games. My team right now is just scratching the surface around this, the use of recommendation and machine learning. And we’ve used some really advanced algorithms, but at the end of the day, most of our efforts have been focused on taking recommendations and content to players, not necessarily, not yet at the point where I think gaming teams are gonna start pushing the boundaries of realism in, in video games. As you start seeing some of the advances in machine learning and AI that happened on the news over the past month or so, then you just can really tell that there’s, the industry is ripe for some major leaps in applications. Think about non-playable characters in any video game, all of a sudden being able to interact, and know who you are as a character, and actually respond to prompts and actually be able to have a conversation in the game with those characters. And that’s just, you know, the one example that comes to mind that you can actually do in text today. And it doesn’t really take much to take it to the next level.
So I think we’re gonna see applications on that front very, very quickly. But I think we are also seeing that games are gonna be adapting to player preferences a little bit more over time. And what that means is that our friends are gonna be easier to find or people that, you know, are easier to get along with in the game, for example, I might be able to find those experiences over time. And AI and ML are gonna be carrying a lot of those experiences. And that’s just on the player-facing, on the product side of things.
I think on the back end of a gaming operation, you also see a turn of opportunity. Security and safety, for example. Making sure that, you know, the internet can be a gnarly place, right? And there’s all sorts of bad stuff happening, but the safety systems can get a lot better through AI and ML. So I think that’s an area, really interesting area to continue to explore.
And even our support systems, right? If you think about how video game companies support their players, it’s very rare that a player will pick up the phone and try to dial technical support. They’re usually asking other players online what to try to, you know, diagnose their problem. Or they’re submitting a ticket or through email or through like a support system. There’s a ton of opportunity there to create agents that can actually help identify when there’s a big problem in the game and actually help us get to a resolution faster and point players in the right direction. So I’m super excited, to be honest, about this space in gaming. There isn’t a single thing, but I think there’s an area of opportunity. And around productizing AI and ML and games. I think that’s the next big horizon.
Evan: Very cool. Yeah, super excited to hear you talk about it. Yeah. Very big horizon, for sure.
Francisco, I do wanna close with one final question. Marcus Mann, the listener who brought us together, he posited a question. If you could be any Mob Francisco, which one would you be?
Francisco: That’s a great question. I love it. Okay, so I’m originally from Mexico, so I have to go with the Axolotl. I know that some of our designers actually put a ton of work behind that friendly Mob. And that’s definitely by far my favorite. How about you, Evan?
Evan: I hate to admit on this show, I’m too poorly informed to make a decision. I’m not a Minecraft player and I feel like I’m gonna alienate if I just pick at random here. I do like your answer.
Francisco’s been great to chat with you today. Thanks so much for coming on the podcast.
Francisco: Thank you for having me. It’s been great. And thank you again. Let’s stay in touch.
Evan: Yeah, absolutely. Please do. If you enjoy this, make sure to like and subscribe to get more content like this on the Mining Your Own Business podcast. Make sure to go play some Minecraft and look out for some of the changes that are, that are coming on the horizon that Francisco and his team are working on. Thanks so much.
Francisco: Thank you.