Powering a PlayerFirst Culture with Massive Gameplay Data

Powering a Player-First Culture with Massive Gameplay Data A Sneak Peek into Data and Electronic Arts Navid Aghdaie, Ph. D Sr. Director of Data Science & Engineering Sep 2015

About Me UCLA Ask. com Electronic Arts Computer Science Ph. D Distributed/Fault. Tolerant Systems Comparison Shopping Startup Search Engine Core Web/News Search Components VP Data Systems Digital Platform, Data Science & Engineering New Large Scale Data Platform Unlock Value of EA’s Rich Gameplay Data

Outline • EA and Games Why Data Matters • Large Scale Data Platform Design and Architecture for Gamer & Game. Play Data • Data in Action Examples of Data Usage 3

EA Overview • Rich history of games, founded 1982 Current Strategic Goals: • Digital Transformation • Player First Culture • Dozens of games, multiple platforms: console, pc, mobile • • • Sports: FIFA, Madden, NHL, NBA DICE: Battlefield, Star. Wars Battlefront Bioware: Dragon Age, Mass Effect Maxis: The Sims Franchise (Sims 4), Sim. City Need for Speed, Bejeweled, Plants vs. Zombies, Simpsons Tapped Out, etc… • 10 s M players/day, across the world 4

Data Usage at EA (Gameplay Data) Game Design and Development • Game updates, new features, new games Live Services • • • Game operations Gameplay optimization Fraud Marketing • • • Player acquisition, re-engagement Cross Promotions Advertisement Customer Service • Player Facing Issues with Game Executive Decisions 5

Example Player Journey through EA Ecosystem Acquisition Email Push Note In-game Banner Advert Personalized features CE Customer Experience

Digital Platform: Data Science & Engineering 7

Core Tech Principles Leverage Open-Source • Join the community and ride its progress – requires investment in talent Embrace the Benefits of the Cloud • • Downward price trend Lowers risk of volume/game success mispredictions Build and spend only as needed Avoid vendor lock-in Build with Scalability, Extensibility, Reliability from the Start • • One platform for all EA games Standards with flexibility to support variations of use Invest in “Crown Jewel IP” Data Components • • Data Science, Algorithms, Data Layer Tools Smart build vs buy decisions 8

Data Platform Architecture External Sources Marketing, Ads, … Platform Services Game Servers And More… River (Capture layer) Tide (Batch Ingestion) Lightning (Streaming Ingestion & Processing) Black Pearl (RDBMS) Capture & Ingestion Shark (Processing) Ocean (Hadoop storage) Surf (Data Science) Data Sources Storage & Processing Pearl (RDBMS) Pond (Hive) Access Layer Reporting & BI Tools Applications Player 360 Bug Sentry Game Analytics Segmentation Manager Live Viewer Subscription API Engagement Manager Access API Experimentation Access & Applications

Data Capture & Ingestion Data Sources • Client Telemetry (mobile, console, pc) • Server Telemetry • EA Internal Services • • e. g. online e-comemerce, micro txn, virtual goods purchase/trade, etc 1 st Party (e. g. sales data from xbox, playstation, android, ios) 3 rd Party (e. g. acquisition marketing, ads) EA web sites traffic Challenges: • Definition and Enforcement of taxonomy standards • Silos and Duplication 10

Streaming and Lambda Architecture Tech Stack • Kafka • distributed pub/sub messaging • Storm • stream event processing 11

Storage & Processing Engine Storage: multi-tier approach • HDFS • Cloud Storage • Archive/Backup Tradeoff: cost vs performance Processing Engine • Apache Hadoop Stack: Hive, Oozie 12

Data Access & Applications • Reporting & Dashboards • Adhoc Analytics • Hive (HQL) • RDBMS (SQL) • APIs, Data Subscription • Closed-Loop Data Driven Online Applications • Personalization/Targeting Systems • Recommendation Engines 13

Data in Action: Examples 14

Dynamic Player Experience Real-time recommendation engine • Modify game configuration to optimize for targeted metrics • Example: Maximize retention by manipulating game difficulty according to user state 15

Initial Configurations Dramatically Affect Win-Rates Level: Deep Sea Creature • Initial seed affects the starting board configuration • # of orange, green, and purple pegs • Potential locations of the pegs • Win ratio ranges from 10 -50% depending on the seed • Effective knob for us to create a better experience 16

How Dynamic Experience Works Targeting Recent Gameplay Predicted Churn Risk (0% – 100%) Recommendation Churn Risk Mapping to Chosen Difficulty Game Client Historical Profile Recommended Levers to Pull 17

Managing Player Relationships Who to target? How to reach them? Segmentation Engagement Data Science Segmentation A self-serve tool which enables granular targeting of EA players. Engagement EA Games Provide the right value A set of tools to curate the player journey through differentiating and improving the player engagement Manage and deliver targeted messages to players in-game, out of game, across the EA network What to show them? Optimization Identify the best placement to engage, track, and test messages to our players Data Science Optimize the Player First experience using Data Science 18

Player Relationship Management – Application Components • Player Profile • Segmentation via Indexing of key attributes, leverage Lucene • Examples: demographics, game ownership, play time, etc • within seconds • Run-time Decisioning Engine • Communication Channels • Email, Push. Note, in-game msg • Campaign management • Recommendations, optimizations 19

Anomaly Detection and Reacting to Issues 20

We’re Hiring! Thank You! Data Scientists & Engineers Contact me! Navid Aghdaie naghdaie@ea. com 21