In this presentation, we will discuss a few examples of how analytics and technology teams at Ubisoft combine machine learning and statistics to create value and improve player experience. Among others, recommendation algorithms help us serve personalized content in the game to our players (customization items they might like, specific in-game content they might enjoy, etc.).
We make extensive use of statistics in our modeling process. Each model is first inspected using custom metrics and refined estimators of performance. Then a live experiment with a carefully crafted sample of users is run. Interestingly, performance of models between prototype and live phase can differ.
We also try to popularize causal inference methods to support decision making. We now routinely run A/B tests to test how new ideas fare once they’re put to the test in front of actual players. We also look for quasi-experiments or create estimates that are tailored for each application in order to provide decision-makers the best possible information.
These applications of statistic supported by a strong engineering stack created by our cross-disciplinary teams.