The Future of Gaming is Big Data | Nate WIGER

Good afternoon, my name is Nate Wiger. I'm from AWS, responsible for heading up our gaming solutions at AWS. So I'm here to talk to you about big data and trends we see as they affect gaming worldwide. Really things that are kind of happening now that are interesting. To start with, just want to do a thought exercise if you bear with me. I want you to think about a game that you've developed recently, and I want you to think about that game and just mentally answer: in that game, what was the favorite thing for your players? What level, what item, what power-up, etc.? What really made players excited about your game? And then I want you to think about the reverse: what really didn't turn your players on? What turned them off? What made them leave the game? What made them not want to keep playing? And I'm not talking about DAU or MAU or any of the common metrics. I'm talking about pieces of your game, parts about the experience you're crafting for players. And if you're like a lot of developers, you probably can't answer that question definitively. Don't feel bad. I was in the same boat once myself as a game developer. But that's kind of at the core of crafting an important experience and an enriching experience. I'm sure that's obvious to many of you. Analytics are not a new thing, and the concept of analytics is not new. But what I want to talk about is analytics from a different tilt. And I'm not going to talk at all about monetization today. The reason why is I think monetization is just a subset of analytics, and more importantly, I don't think it's the most important thing or the most interesting thing about analytics. And we're seeing more and more trends in big data, and I'll explain a little bit of what that means, that really unlock a lot of interesting potentials for your game.

So I don't want to talk about monetization. I'm going to start by talking about maps. So this is actually a heat map of cyclists. This part of the European Cycling Challenge was something that the fitness app Endomondo put on their European base. And what they actually did was kind of fun: they made a game out of it. They pitted a dozen different European cities against one another, and they basically made a challenge. They said, over two weeks, whoever rides the furthest wins. And it's a really interesting real-world social experiment if you think about it, because you're motivating people to get out in the real world and actually do something, interact with one another. And so as a result of this, they gathered a whole bunch of different data from their app. They used, as I mentioned, they have a fitness app, so both through their app and through their sensors, they're able to capture a whole bunch of interesting data. One aspect of which is just where most people rode. Red sections are obvious layers where people rode the most, green less often, etc. But they also captured a whole bunch of other interesting time series data. Namely, they found that 50% of rides happen during the week, and moreover, they were five kilometers or less. That's got some interesting implications that you might not expect. That basically means, hey, people are willing to ride a couple kilometers just around town as part of their daily activities, getting to and from work, etc. Now that might not be that revolutionary in parts of Asia, but it is interesting because it lets you go back and actually make changes potentially to government structures and say, hey, we actually have data now that shows we can change the way people actually commute to and from work. But heat maps are not necessarily in and of themselves anything new. This is an example from Assassin's Creed 3. To be honest, I picked it just because I thought it was really cool-looking, looks really good on this slide. But again, there's a lot of really useful information, and this shows you what parts of the map are played most, what parts the least. Well, this is really just scratching the surface, right? This is only one portion of the data. This shows you where people are at a given snapshot in time, but it doesn't necessarily show you how they're progressing through the world. It doesn't show you, okay, they started here, they went there, they died there, etc. So when we talk about big data, it's about getting data like this, but then enabling yourself to answer deeper questions, to gather more data, and iterating and gathering more and more data. So I'll give you one more example about this which I found really fascinating. So this is from a company called SportVU. And SportVU is actually a company that outfits sporting activities with cameras and nationally monitors players as they're going around playing basketball or football, soccer, what have you. And they can actually plot out in real time what players are doing on the court. And so this is again one example: they could build a heat map. This is from a real game, the Celtics versus the Mavericks, actually showing where the players were spending most of their time. If you see on the left side, you can see that the Celtics are spending most of their time kind of deep in the key. And if you look at the Mavericks, yeah, they've got some people in the key, but they've also got people kind of around here, kind of at the top around the three-point line. So if you've ever watched sports, I'm sure you all have, the old method for doing this, the old coach's method, was to actually sit there and go back to the tape. And before playing an opponent, they would go and watch hours and hours of tape in order to glean these insights. Now they can do it all based on data. And so this is another great example of big data: multiple different streams coming in. And again, this is just one representation. They also can spit out a whole bunch of other interesting things about the players. They can spit out where they're taking most of their possessions, how long they're running before they make a shot, how long they're going before making a pass, the mean number of times between making passes until they make a shot. It starts getting your wheels turning. You think about what the implications would be for your game. It really fundamentally changes things when you can actually get access to this data. So hopefully that gives you some food for thought.

So how do you actually translate that into practical steps? We've talked a little bit about things that are interesting happening, but I'm big into practical applications. So all three of these examples really boil down to the same thing: they're just a series of events. And if you've done anything with analytics, this is not news. Basically, every time you have an event—somebody's running down court, somebody's taking a shot—these are all just a series of events. You're just capturing the four main things: who, what, where, and when. What did the player do? Who were they? Where were they in the game? And then a timestamp. And game companies that are very successful in this space, the Supercells and the Rovios of the world, they're capturing these kinds of events all the time. So the first question I asked to kick off the session, saying, okay, well, tell me about your players, their experiences, what were the favorite things? If you weren't able to answer that question, don't feel bad. But realize that the Supercells and Rovios of the world can absolutely answer that question in minute detail. And they can give you a stack ranking. They can say, here's what our favorite players' items are when they're playing in these modes, here's what they do. And it's all through just capturing events constantly throughout your game. So the first major point I want to make is: don't worry about what you're going to do with all this event data. Worry about what you're going to do without it. And just capture as much data as you can from the start of your game. And what does that look like? Actually, in-game code is quite straightforward. It's just a series of events that you're constantly writing out as part of your game. And you can either do this natively or you can use any number of the SDKs that are available out there. There's some great solutions out there. We also have an analytics SDK. But the tools are out there now. So the point is that the bar has become so simple to gather this, there's really no reason not to just gather as much data about your game as possible from the very first day. And the reason why I say that over and over again is the fact that as your game becomes more popular, then you'll say, oh, well, hey, we're gathering steam, what can we do to actually make this a big hit? And if you start gathering data then, it's too late. Because by the time you start gathering data, then it takes a while to get the data in, you start running your analysis, and now potentially your game has started going back down the other way down the slope. So this is kind of the idea of analytics in a nutshell. But what does big data have to do with it? Well, big data just has to do with the volume of data that you're pulling. And that's one aspect of it. You're getting this from multiple different sources. You're getting this not just from your mobile client, but you can be getting this from other server applications you wrote, you can be getting this from GPS data on the phone, you can be getting this from third-party feeds that you're pulling in as well, and combining this all to give you a very holistic view of what players are doing in your game.

So how does that translate into what people are actually doing? We wanted to give an example of real-world usage here. So this is the actual analytics pipeline for Supercell for Clash of Clans, which runs entirely on AWS. And the Supercell guys are great. They're very open and sharing. Outside of being very successful, they're also very open in terms of sharing their architectures and talking about their developers. Just an overall great team. So how did they actually implement their game analytics pipeline on AWS? Very recently we introduced a new service called Kinesis. And I don't want to turn this into an AWS sales pitch, but I do want you guys to have a real-world example of how Supercell is actually doing this. So Kinesis—the best way to think about it is just this big data fabric. And essentially it's just a very low latency, high speed, near real-time, multiple input, multiple output message bus. You can configure this as a service. Just put as much data in as you want, and then on the other end you can have one or more consumers that can do a whole variety of different things with this data. So gaming is a prime use case for this, and there's also many other interesting use cases outside of gaming. But gaming is a huge driver. So what Clash of Clans did with this is—and you can't really see this, but you get the idea—they actually instrumented their mobile game client so that it's just streaming up that same kind of event data that I outlined over and over again. And then once it gets to Kinesis, one of the ways that they actually consume that is they have a custom set of applications that they wrote on EC2, our virtualization platform, that is just pulling all of that data off and doing a number of different things with it. They were doing custom different processing. They're actually writing out different logs. One of the things they're spitting out is an in-game trends dashboard, real-time in-game trends. So they can, through the combination of Kinesis and their custom EC2 code, actually look in real time and see in extreme detail what their players are actually doing in-game. And this is a great example of the workflow of Kinesis because of the fact that you can flow data in one way and then pull it out not only one way but another way as well: S3, which is our storage service. Now, one thing I meant to mention earlier but forgot to mention when I was talking about collecting data: before the cloud, big data was very complicated to do. And one of the main reasons why it was complicated was the fact that it was extremely expensive. It required very large investments in server hardware, large investments in database software. So we've obviously spoken about our cloud offering, but across the industry there's a number of great cloud offerings now. And it is so cheap to store the data in the cloud, it really fundamentally changes the things you can do. You can store terabytes of data in something like S3 for literally tens of dollars a month. And that's data that's backed up and replicated in multiple locations, etc. So when I was talking about earlier, hey, just don't worry about what to do with all your data, just get it into the cloud and worry about it later, this is a prime example of that. It's cheap. There's no reason to get hung up on all the possible use cases. Just shove it in the cloud from day one and then come back to it later. So what Supercell is doing is they're forking their Kinesis stream in two different directions. They're looking at real-time data on their EC2 dashboards, and then they're funneling all of their aggregate statistics into S3. And the last major piece of their architecture is Redshift, which is our data warehouse as a service. Again, all of these services are pay-as-you-go, which is another key factor in building these analytics pipelines. Previously, you would have to potentially spend tens of thousands, if not millions of dollars. And the only people that could really build analytics pipelines were the very big game companies. And that's not true anymore. You're talking about something that could cost you literally a few hundred dollars to get up and running. And after you test it out and you run your analysis, you can turn it back off and you've paid a few hundred dollars. It really fundamentally changes the way you're approaching analytics. And so what Clash of Clans is doing with the Redshift data then is they're running their weekly reports, their overall aggregates, their DAU, MAU, and more in-depth analysis of what their players are doing as well. Well, this is a very straightforward architecture. And the last piece I want to mention about this is the fact that it's all locked. One of the big things that we aim for at AWS is not having you bound to the entire platform. If you want to run game servers somewhere else, if you want to run telco, there's a whole bunch of different things in the Asian market especially that are fundamentally challenging, especially when you get into areas that have latency between islands, etc. So you can still run local game servers in all those regions and still push all of your data up through Kinesis. All data going into AWS is free. I forgot to mention that. A key part of this architecture is you can shove as much data across and you're not paying any bandwidth fees. So you're not paying any bandwidth fees. We're not charging you for any bandwidth that you're consuming by pushing data into us. So again, get your data into the cloud. You'll pay a nominal fee to store it in S3 until you have the time and the resources to come back and do some more interesting analysis later.

So the last part I want to talk about is the evolving field of real-time analytics and some things to think about in terms of the possibilities. One of the things that Clash of Clans is able to do by using Kinesis is actually do real-time analytics. What does that mean? Well, this is a great example from Cookie Run. I'm sure you're familiar with this game. Devsisters, the developer for this, they actually did this for our Korea AWS office. They basically got hit in a good way by the App Store running a promotion overnight without telling them. And they were running on AWS, so their service could scale, etc. But think about what you could do if you could actually react to something like this in real time and run some kind of promotion. Think about if you could actually get that data stream and actually change what you were selling or how to target users. Think about if you could more finely tune your monetization strategies. So while monetization is a piece of it, you're fundamentally building an infrastructure that allows you to do all kinds of more interesting things. So maybe it's not targeting with offers. Maybe it's saying, hey, grouping people together, creating challenges on the fly. You think about MMOs. Think about if you can reallocate people between worlds if there's a big influx of people, as there often is. And to round out this discussion, I want to highlight one final part here. And that is, I still do think we're kind of at monetization 1.0. So everybody's played Candy Crush. I admit I have as well. I'm sure we all have at this point. I'll admit I've spent money on it. I'm not too proud to admit that. But if you look at them, they're extremely popular. I picked this game intentionally because nobody can fault Candy Crush. King Games, I mean, they're obviously taking over the world. They're doing a fantastic job. But if you look at what they're doing, it's a very simple strategy at this point. So everybody basically gets the same pop-up message. It could say, hey, here's your generic pop-up message: you can buy this candy and it's $1.99. And if you're midway through and you're almost out of moves, here, buy some extra moves for $0.99. What if you more finely tune that message? What if you say, hey, based off all the different people playing and based off the historical information we have about you, actually if you choose this candy, you're actually going to have a 30% chance of finishing the level? Think of how that would change the motivation for somebody to actually click on that, rather than it being everybody getting the same message. And then again, same kind of thing over here. You could say, hey, you're only two moves away. You have an 80% chance of finishing it if you just buy five more moves. So these are just very simple examples, and the sky's the limit. You can do all kinds of other interesting things in terms of motivating different players. In terms of saying, okay, people are in this kind of world, how about you go over and join them right now? How about if you move between these different games? Try this other game mode. You really like this game mode? Try this other game mode. Ultimately, everybody in your game is connected in some way. And the real goal is to figure out how the games are not unlike a giant video recommendation engine. If you're trying to say, hey, based off these users and their past behaviors, then we can map it to these other users and we can find some interesting content these people would really like. So these are the kinds of trends that I personally think are really interesting. And they're really unlocked by the fact that you can do this relatively inexpensively now. You can just store your data up on the cloud, worry about how to analyze it later. When it comes time to analyze it, you can pay for a few hours of compute, you can run it through some analytics, and then you can throw that all away and store your results and stop paying for it. So I wanted to leave you with a couple last things to think about. This is a kind of personal challenge. Even if you are doing analytics, or especially if you're not, I just want you to think about—hopefully this has gotten some ideas running about things that you could measure. Additional things, maybe where people are in levels, simple, or maybe the types of items they're using in order to complete levels, frequency that they're actually using those items, which items did they only buy once? I mean, there's a whole bunch of analysis where you probably already have part of the data. Just go and just start measuring that. If you already have an analytics SDK set up, which is fantastic, then just add that custom metric and just start measuring it and start brainstorming about some of the more interesting things you can do with it. And the last thing I wanted to mention is just kind of bringing it all back together. As I said, don't worry about what you're going to do with the data. Worry about what you're going to do without it. Gather the data, come back and analyze it later. If you found this interesting, the last thing I'll mention is Drew Harpeia, who's a local AWS solution architect with a background in gaming as well. He's actually giving a talk on Thursday where he's going to go into deep nuts and bolts about this. So if you're all interested in how this could actually work under the covers, I'd encourage you to go see his talk. So that's all I have. Thank you very much for your time.