Gaming Analytics Summit 2015, San Francisco – Day 2 Highlights

Highlights from the presentations by Gaming Analytics leaders from Activision, Riot Games and Daybreak Game Company (formerly Sony Online Entertainment) on day 2 of Gaming Analytics Innovation Summit 2015 in San Francisco.

The gaming-analyticsGaming Analytics Innovation Summit was held by Innovation Enterprise in San Francisco on April 29 and 30, 2015.

Across many industries, large and small organizations are using analytics and data science to offer greater insight and customer service. The gaming industry is almost well placed in that - particularly with online and social gaming - the companies already keep a vast amount of data on gamers. The challenge remains to make use of this data in a way that offers true value for money whilst enhancing the experience of the customer.

Industry leading experts shared case studies and examples providing deep insight into how the gaming industry uses analytics and data science.

Highlights from Day 1

Here are the highlights from selected talks on Day 2:

ActivisionSpencer Stirling, Senior Data Scientist, Activision shared his experience and best practices for Game Analytics in his talk "Complex Event Processing & QoS". In-game analytics allow developers to create clear pictures of gamer behavior - this is particularly important in multiplayer online games to aid with matchmaking and detect cheating. Spencer explained how Activision is using data science to create better games.

He provided a quick overview of the Analytic Services team at Activision. The team's focus is on Complex Event Processing - advanced real-time analytics combining multiple streams of complex data. The Analytic Services team is working on a variety of projects including Quality of Service monitoring, identifying cheating (hacking detection, behavior monitoring), boosting detection, and recommender systems. The underlying data processing system is based on lambda architecture comprising of Kafka event queue, HDFS datalake, relational databases, batch and ad-hoc quering systems (greenplum), and real-time querying systems.

The real-time Analytics stack is based on Storm. Events from the Kafka queue are cached in Redis and processed by Storm. He noted that Storm is very chatty and to resolve that problem, they are:
  1. passing only Event IDs within the topology to storm and
  2. using only a single bolt.

Finally, the results are sent to something durable with fast writes, such as Cassandra or published to some other Kafka queue for further analysis.

The dashboards are created through a combination of Elastic-search, Logstash, and Kibana (ELK). Talking about quality control, he emphasized that the focus should not be on reducing the mean quality, but rather one should strive to reduce the variance.

riot-logo-rich-blacksRudi Bonaparte, Manager, Analytics, Riot Games delivered an insightful talk on team-work in his talk titled "Beyond the Math: Structuring Analytics Teams for Success ". Individual brilliance is great, but it’s only one part of the equation behind great analytics organizations. The best analysts work in teams that find ways to structure the way they work so as to increase exposure to the company while creating a safe space to explore data and hypotheses. Rudi shared the Analytics team structure at Riot, best practices and lessons learned.

Talking about team work, he explained a few attributes of the craft, relationships and experience. He emphasized the need to ensure personal and professional growth. For success, Analytics teams must have a healthy balance of business sense and analytic capabilities. He asserted the leadership for Analytics team must be divided across two tracks: Lead Analyst and Manager. The former should provide craft leadership and mentorship, and work on the toughest engagements. On the other hand, the latter should steward analyst career and work experience, while also working on building relationships and team brand.

The above team structure offers several benefits to the analysts as well as business partners by promoting collaboration, creativity, and quality. He mentioned that the continual advancement of analytics is a big challenge and requires structuring teams so that new techniques can be regularly developed, disseminated, and applied.

DaybreakBen Weber, Director, BI & Analytics, Daybreak Game Company (formerly Sony Online Entertainment) gave an interesting talk on "Holding Effective Data Meetings with Game Teams". One of the challenges faced by game analysts is holding effective meetings with game development teams that lead to actionable results. In addition to providing automated reporting, one of the key functions of the data team at Daybreak is to sit down with development teams each week to review the reports, provide context for the metrics, share new analysis, and make recommendations to improve player lifecycles. His talk was structured around 3 key questions: What data to share with game teams?, When and how to share data?, and How to operationalize your data?

Some of the most common data analytics challenges are -
  1. little communication between the data and game teams
  2. automated reports not providing actionable data
  3. data team is not working on actionable issues
  4. team structure separates game developers from analysts

At Daybreak, apart from a core Data Science team, there are data analysts embedded within game teams (such as H1Z1, PlanetSide 2, and DC Universe Online) to drive collaboration. The data pipeline (comprising of Vertica, Tableau server, and Excel) at Daybreak supports automated reports and a self-service portal.

In order to promote an interactive approach between game team and analytics team, he advocated that there should be a proper hand-off of the results, shared ownership of data model must be encouraged, and the ROI of results should be well communicated. In conclusion, he shared the following as the answers to the 3 key questions he had asked earlier:
  1. All the KPIs and more detailed analysis as needed should be shared with game teams
  2. Data must be shared regularly through automated reports, scrum meetings, and data insights meetings and
  3. In order to operationalize the data insights meet early with preliminary data, develop an action plan, and follow-up with more data.