Wednesday, July 22, 2015

Zynga builds big data innovation culture by making analytics open to all developers

The next BriefingsDirect analytics innovation case study interview explores how Zynga in San Francisco exploits big-data analytics to improve its business via a culture of pervasive, rapid analytics and experimentation.

Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy.

To learn more about how big data impacts Zynga in the fast-changing and highly competitive mobile gaming industry, BriefingsDirect sat down with Joanne Ho, Senior Engineering Manager at Zynga, and Yuko Yamazaki, Head of Analytics at Zynga. The discussion is moderated by me, Dana Gardner, Principal Analyst at Interarbor Solutions.

Here are some excerpts:

Gardner: How important is big data analytics to you as an organization?

Ho
Ho: To Zynga, big data is very important. It's a main piece of the company and as a part of the analytics department, big data is serving the entire company as a source of understanding our users' behavior, our players, what they like, and what they don’t like about games. We are using this data to analyze the user’s behavior and we also will personalize a lot of different game models that fit the user’s player pattern.

Gardner: What’s interesting to me about games is the people will not only download them but that they're upgradable, changeable. People can easily move. So the feedback loop between the inferences, information, and analysis you gain by your users' actions is rather compressed, compared to many other industries.

What is it that you're able to do in this rapid-fire development-and-release process? How is that responsiveness important to you?

Real-time analysis

Ho: Real-time analysis, of course, is critical, and we have our streaming system that can do it. We have our monitoring and alerting system that can alert us whenever we see any drops in user’s install rating, or any daily active users (DAU). The game studio will be alerted and they will take appropriate action on that.

Gardner: Yuko, what sort of datasets we are talking about? If we're going to the social realm, we can get some very large datasets. What's the volume and scale we're talking about here?
Become a member of myVertica today
Register now
Gain access to the free HP Vertica Community Edition
Yamazaki: We get data of everything that happens in our games. Almost every single play gets tracked into our system. We're talking about 40 billion to 60 billion rows a day, and that's the data that our game product managers and development engineers decide what they want to analyze later. So it’s already being structured and compressed as it comes into our database.
Gardner: That’s very impressive scale. It’s one thing to have a lot of data, but it’s another to be able to make that actionable. What do you do once that data is assembled?

Yamazaki: The biggest success story that I will normally tell about Zynga is that we make data available to all employees. From day one, as soon as you join Zynga, you get to see all the data through our visualization to whatever we have. Even if you're FarmVille product manager, you get to see what Poker is doing, making it more transparent. There is an account report that you can just click and see how many people have done this particular game action, for example. That’s how we were able to create this data-driven culture for Zynga.

Yamazaki
Gardner: And Zynga is not all that old. Is this data capability something that you’ve had right from the start, or did you come into it over time? 

Yamazaki: Since we began Poker and Words With Friends, our cluster scaled 70 times.

Ho: It started off with three nodes, and we've grown to 230 node clusters.

Gardner: So you're performing the gathering of the data and analysis in your own data centers?

Yamazaki: Yes.

Gardner: When you realized the scale and the nature of your task, what were some of the top requirements you had for your cluster, your database, and your analytics engine? How did you make some technology choices?

Biggest points

Yamazaki: When Zynga was growing, our main focus was to build something that was going to be able to scale and provide the data as fast as possible. Those were the two biggest points that we had in mind when we decided to create our analytics infrastructure.

Gardner: And any other more detailed requirements in terms of the type of database or the type of analytics engine?
Become a member of myVertica today
Register now
Gain access to the free HP Vertica Community Edition
Yamazaki: Those are two big ones. As I mentioned, we wanted to have everyone be able to access the data. So SQL would have been a great technology to have. It’s easy to train PMs instead of engineering sites, for example, MapReduce for Hadoop. Those were the three key points as we selected our database.

Gardner: What are the future directions and requirements that you have? Are there things that you’d like to see from HP, for example, in order to continue to be able do what you do at increasing scale?

Ho: We're interested in real-time analytics. There's a function aggregated projection that we're interested in trying. Also Flex Tables [in HP Vertica] sounds like a very interesting feature that we also will attempt to try. And cloud analytics is the third one that we're also interested in. We hope HP will get it matured, so that we can also test it out in the future.
We we have 2,000 employees, and  at least 1,000 are using our visualization tool on a daily basis.

Gardner: While your analytics has been with you right from the start, you were early in using Vertica?

Ho: Yes.

Gardner: So now we've determined how important it is, do you have any metrics of what this is able to do for you? Other organizations might be saying they we don't have as much of a data-driven culture as Zynga, but would like to and they realize that the technology can now ramp-up to such incredible volume and velocity, What do you get back? How do you measure the success when you do big-data analytics correctly?

Yamazaki: Internally, we look at adoption of systems. We we have 2,000 employees, and  at least 1,000 are using our visualization tool on a daily basis. This is the way to measure adoption of our systems internally.

Externally, the biggest metric is retention. Are players coming back and, if so, was that through the data that we collect? Were we able to do personalization so that they're coming back because of the experience they've had?

Gardner: These are very important to your business, obviously, and it’s curious about that buy-in. As the saying goes, you can lead a horse to water, but you can't make him drink. You can provide data analysis and visualization to the employees, but if they don’t find it useful and impactful, they won’t use it. So that’s interesting with that as a key performance indicator for you.

Any words of advice for other organizations who are trying to become more data-driven, to use analytics more strategically? Is this about people, process, culture, technology, all the above? What advice might you have for those seeking to better avail themselves of big data analytics?

Visualization

Yamazaki: A couple of things. One is to provide end-to-end. So not just data storage, but also visualization. We also have an experimentation system, where I think we have about 400-600 experiments running as we speak. We have a report that shows you run this experiment, all these metrics have been moved because of your experiment, and A is better than B.

We run this other experiment, and there's a visualization you can use to see that data. So providing that end-to-end data and analytics to all employees is one of the biggest pieces of advice I would provide to any companies.

One more thing is try to get one good win. If you focus too much on technology or scalability, you might be building a battleship, when you actually don’t need it yet. It's incremental. Improvement is probably going to take you to a place that you need to get to. Just try to get a good big win of increasing installs or active users in one particular game or product and see where it goes.

Gardner: And just to revisit the idea that you've got so many employees and so many innovations going on, how do you encourage your employees to interact with the data? Do you give them total flexibility in terms of experiments? How do they start the process of some of those proof-of-concept type of activities?
Become a member of myVertica today
Register now
Gain access to the free HP Vertica Community Edition
Yamazaki: It's all freestyle. They can log whatever they want. They can see whatever they want, except revenue type of data, and they can create any experiments they want. Her team owns this part, but we also make the data available. Some of the games can hit real time. We can do that real-time personalization using that data that you logged. It’s almost 360-degree of the data availability to our product teams.
If you focus too much on technology or scalability, you might be building a battleship, when you actually don’t need it yet.

Gardner: It’s really impressive that there's so much of this data mentality ingrained in the company, from the start and also across all the employees, so that’s very interesting. How do you see that in terms of your competitive edge? Do you think the other gaming companies are doing the same thing? Do you have an advantage that you've created a data culture?

Yamazaki: Definitely, in online gaming you have to have big data to succeed. A lot of companies, though, are just getting whatever they can, then structure it, and make it analyzable. One of the things that we've done that do well was to make a structure to start with. So the data is already structured.

Product managers are already thinking about what they want to analyze before hand. It's not like they just get everything in and then see what happens. They think right away about, "Is this analyzable? is this something we want to store?" We're a lot smarter about what we want to store. Cost-wise, it's a lot more optimized.

Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy. Sponsor: HP.

You may also be interested in:

Monday, July 20, 2015

How big data powers GameStop to gain retail advantage and deep insights into its markets

The next BriefingsDirect analytics innovation case study interview highlights how GameStop, based in Grapevine, Texas, uses big data to improve how it conducts its business and better serve its customers.

By accessing data sources that were unattainable before and pulling that data out into reports in just a few minutes across nationally distributed retail outlets, GameStop more deeply examines how its campaigns and products are performing.

Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy.

To learn more about how they deploy big data and use the resulting analytics, BriefingsDirect sat down with John Crossen, Data Warehouse Lead at GameStop. The discussion is moderated by me, Dana Gardner, Principal Analyst at Interarbor Solutions.

Here are some excerpts:

Gardner: Tell us a little bit about GameStop. Most people are probably familiar with the retail outlets that they see, where you can buy, rent, trade games, and learn more about games. Why is big data important to your organization?

Crossen: We wanted to get a better idea of who our customers are, how we can better serve our customers and what types of needs they may have. With prior reporting, we would get good overall views of here’s how the company is doing or here’s how a particular game series is selling, but we weren’t able to tie that to activities of individual customers and possible future activity of future customers, using more of a traditional SQL-based platform that would just deliver flat reports.

Crossen
So, our goal was to get s more 360-degree view of our customer and we realized pretty quickly that, using our existing toolsets and methodologies, that wasn’t going to be possible. That’s where Vertica ended up coming into play to drive us in that direction.

Gardner: Just so we have a sense of this scale here, how many retail outlets does GameStop support and where are you located?

Crossen:  We're international. There are approximately 4,200 stores in the US and another 2,200 international.

Gardner: And in terms of the type of data that you are acquiring, is this all internal data or do you go to external data sources and how do you to bring that together?

Internal data

Crossen: It's primarily internal data. We get data from our website. We have the PowerUp Rewards program that customers can choose to join, and we have data from individual cash registers and all those stores.

Gardner: I know from experience in my own family that gaming is a very fast-moving industry. We’ve quickly gone from different platforms to different game types and different technologies when we're interacting with the games.

It's a very dynamic changeable landscape for the users, as well as, of course, the providers of games. You are sort of in the middle. You're right between the users and the vendors. You must be very important to the whole ecosystem.

Crossen: Most definitely, and there aren’t really many game retailers left anymore. GameStop is certainly the preeminent one. So a lot of customers come not just to purchase a game, but get information from store associates. We have Game Informer Magazine that people like to read and we have content on the website as well.

Gardner: Now that you know where to get the data and you have the data, how big is it? How difficult is it to manage? Are you looking for real-time or batch? How do you then move forward from that data to some business outcome?

Crossen: It’s primarily batch at this point. The registers close at night, and we get data from registers and loads that into HP Vertica. When we started approximately two years ago, we didn't have a single byte in Vertica. Now, we have pretty close to 24 terabytes of data. It's primarily customer data on individual customers, as well Weblogs or mobile application data.
Become a member of myVertica today
Register now
Access the FREE HP Vertica Community Edition
Gardner: I should think that when you analyze which games are being bought, which ones are being traded, which ones are price-sensitive and move at a certain price or not, you're really at the vanguard of knowing the trends in the gaming industry -- even perhaps before anyone else. How has that worked for you, and what are you finding?

Crossen: A lot of it is just based on determining who is likely to buy which series of games. So you won't market the next Call of Duty 3 or something like that to somebody who's buying your children's games. We are not going to ask people buy Call of Duty 3, rather than My Little Pony 6.

The interesting thing, at least with games and video game systems, is that when we sell them new, there's no price movement. Every game is the same price in any store. So we have to rely on other things like customer service and getting information to the customer to drive game sales. Used games are a bit of a different story.

Gardner: Now back to Vertica. Given that you've been using this for a few years and you have such a substantial data lake, what is it about Vertica that works for you? What are learning here at the conference that intrigues you about the future?

Quick reports

Crossen: The initial push with HP Vertica was just to get reports fast. We had processes that literally took a day to run to accumulate data. Now, in Vertica, we can pull that same data out in five minutes. I think that if we spend a little bit more time, we could probably get it faster than half of that.

The first big push was just speed. The second wave after that was bringing in data sources that were unattainable before, like web-click data, a tremendous amount of data, loading that into SQL, and then being able to query it out of SQL. This wasn't doable before, and it’s made it do that. At first, it was faster data, then acquiring new data and finding different ways to tie different data elements together that we haven’t done before.

Gardner: How about visualization of these reports? How do you serve up those reports and do you make your inference and analytics outputs available to all your employees? How do you distribute it? Is there sort of an innovation curve that you're following in terms of what they do with that data?
We had processes that literally took a day to run to accumulate data. Now, in Vertica, we can pull that same data out in five minutes.

Crossen: As far as a platform, we use Tableau as our visualization tool. We’ve used a kind of an ad-hoc environment to write direct SQL queries to pull data out, but Tableau serves the primary tool.

Gardner: In that data input area, what integration technologies are you interested in? What would you like to see HP do differently? Are you happy with the way SQL, Vertica, Hadoop, and other technologies are coming together? Where would you like to see that go?

Crossen: A lot of our source systems are either SQL-server based or just flat files. For flat files, we use the Copy Command to bring data, and that’s very fast. With Vertica 7, they released the Microsoft SQL Connector.

So we're able to use our existing SQL Server Integration Services (SSIS) data flows and change the output from another SQL table to direct me into Vertica. It uses the Copy Command under the covers and that’s been a major improvement. Before that, we had to stage the data somewhere else and then use the Copy Command to bring it in or try to use Open Database Connectivity (ODBC) to bring it in, which wasn’t very efficient.

20/20 hindsight

Gardner: How about words of wisdom from your 20/20 hindsight? Others are also thinking about moving from a standard relational database environment towards big data stores for analytics and speed and velocity of their reports. Any advice you might offer organizations as they're making that transition, now that you’ve done it?

Crossen: Just to better understand how a column-store database works, and how that's different from a traditional row-based database. It's a different mindset, everything from how you are going to lay out data modeling.
Become a member of myVertica today
Register now
Access the FREE HP Vertica Community Edition
For example, in a row database you would tend to freak out if you had a 700-column table. In the column stores, that doesn’t really matter. So just to get in the right mindset of here’s how a column-store database works, and not try to duplicate row-based system in the column-store system.

Listen to the podcast. Find it on iTunes. Get the mobile app for iOS or Android. Read a full transcript or download a copy. Sponsor: HP.

You may also be interested in: