Author: Ivaylo Bahtchevanov

Why Online Gaming Needs Personalization

The online gaming industry is one of the largest and most lucrative businesses in the world, projected to reach $321 billion in revenue by 2026. The largest PC gaming distribution service Steam released 300 games in 2012 and over 12,000 in 2022. With so much competition, game developers are pushing to capture players’ attention and keep them engaged while at the same time trying to monetize their user base to build a successful business. The reality is - most players will churn shortly after trying the game, with retention decaying and only 10% of players actually finishing a game.

The game needs to interact with its players in order to guide them through an optimal gaming experience, while at the same time showing them relevant ads or providing in-app purchases to enhance the gameplay. By personalizing the outreach and in-game experience, games can ensure players don’t leave while increasing their overall spend. This blog will discuss why some of the latest advancements in machine learning can drastically improve user engagement and ARPU.

The Challenges of Personalized Recommendations and Notifications

As a game expands its user base, it introduces a level of scale that makes it very difficult to uniquely engage with and curate experiences for each of its users.

Personalizing gameplay at scale is challenging, and poor execution can do more harm than good. Irrelevant recommendations erode trust, while excessive or ill-timed notifications irritate even the most engaged players. Personalized additions and outreach should seamlessly integrate into the game experience. Because games lose so many players early on, some developers try to overdo personalized notifications to retain players without considering long-term engagement. A study of mobile games found that the average game sent a push notification every day for the first two weeks, but more notifications than gameplay sessions actually decreases engagement in the long-run.

Common methods to identify players’ preferences – like supervised or reinforcement learning – also face a cold-start data problem with new players. These techniques are not able to effectively personalize messaging or experiences for new players without available data on their preferences, losing out on the players that need to be retained the most.

These shortcomings with existing methods are why graph learning has become a popular choice to drive gaming personalization and recommendations.

Graph Representation Learning for Personalization

Graph Neural Networks (GNNs) are the leading machine learning approach for processing graph data. They enable representational learning without feature engineering, capturing complex relationships and dependencies between entities. GNNs leverage the entire graph to make predictions, utilizing similarities from other users as indicators of potential behavior. For a brand new user - there’s plenty of signal that can be leveraged across the entire graph to make predictions without historical precedent.

Compared to traditional methods, GNNs scale effectively and process large amounts of heterogeneous data, including as player interactions and contextual information. They also overcome the data cold-start problem for new players by accurately predicting future behavior based on rich context and graph features, even when there is limited data for these players. These advantages result in more accurate predictions for maximizing engagement and ARPU across players.

You can read more about the technical benefits of GNNs in our deep dive blog.

Some of the largest game studios use graph learning to personalize marketing and outreach. Activision uses graph-based machine learning to recommend game modes and prioritize in-game notifications for Call of Duty, a game with over 70 million daily active users.

Similarly, NetEase implemented Graph Neural Networks to drive in-game purchases, increasing gross merchandise value (GMV) by 15%. By leveraging connectivity between players and entities in the game, the team was able to drive highly effective recommendations for bundled activities and products.

The Kumo Approach

Kumo’s approach enables you to go from raw data to large scale GNN-powered personalization in minimal time-to-value. The user connects their data tables directly, and Kumo will build your enterprise graph from the raw data, learning the relationships and interactions across all entities in your data tables.

Kumo provides a flexible no-code interface that makes it easy to generate any number of predictions for many use cases. Imagine being able to write SQL queries to predict which games or items a player will buy while simultaneously identifying how that will affect your ARPU. You can see an example of how a game developer can build a graph and provide personalized recommendations in the personalization walkthrough.

Beyond personalization, Kumo provides a flexible no-code interface for quickly generating concurrent predictions for any number of use cases. This powerful abstraction, called the Predictive Query, allows users to perform a comprehensive user analysis to accurately forecast different entities based on each user's unique attributes.

If you are interested in hearing more about Kumo, reach out to us directly!

Cookies settings