Customer churn can have various definitions but we here focus on churn in free-to-play (F2P) apps and games. In F2P, users do not perform a particular action that indicates their leave explicitly. For example, a cancellation notice in a monthly subscription business. Instead, users churn quietly and one has to observe their behavior carefully in order to recognize their churn. One has to find the last session of users before they will never return again on their own. However, how can we do so? Can we be certain that a user does not return after 14 days of inactivity? Or after 21 days? This questions cannot be answered in general but has to be decided for each app individually. Fortunately, this is also a question that we can answer in a data-driven way.
Predicting churn allows you to proactively take measures to prevent churn before it actually happens. There are multiple ways to predict customer churn based on different types of data. At goedle.io, we focus on predicting churn based on telemetric user behavior.
Consider a mobile game for example. After installing the app, the players will be highly engaged and our AI will receive various kinds of events. Over time, our AI learns a model that predicts how likely it is that a player will continue to be engaged based on these events. These prediction can then be used to personalize the communication, tailor personalized offers, or adjust in game content. More examples can be found below.