Customer churn is a very addressable problem for machine learning. This use case we found to discuss focuses on mobile apps user engagement. With this use case as the basis, this is the first in a series of posts we will share that walk through the concepts business people will want to understand when considering machine learning as a tool for reducing churn.
How to define and predict churn for machine learning?
Defining what is churn is always specific to an organization and a given service. In this case, the mobile app provider defines a churned customer as one who has been inactive for 30 consecutive days.
This definition was chosen because a lost user of a free mobile app is not as easily defined as with many other situations. For example, a cloud service business can easily define a churned customer as when their client notifies them to terminate a subscription. However, even in an easy case such as paid subscriptions, sometimes it is a good idea to carefully consider additional questions to predict churn based on behavior of the client. For example, should a customer who hasn’t used the subscription for an extended period of time be considered churned or at risk of churn even though the subscription is not terminated?
Clearly there are multiple factors one must consider to define and or predict future churn in order to choose a wise churn definition, and it probably requires some type of analysis with traditional business intelligence software. For instance, an analysis can be conducted to find out how many subscription users end up as churn when activity has dropped or stopped.
If such a useful signal can be defined, then it would be used for the machine learning training rather than business metric definition of churn. This is because you want to be able to identify at-risk users or clients while there is still time to take action on them. Once these predictive metrics are decided upon, the dataset must be updated with a new field and values such as “churned” or “not-churned” and or “at risk of churn”.
To read the case study itself see https://insidebigdata.com/2017/04/14/predicting-mobile-app-user-churn-training-scaling-machine-learning-model/
To see examples of visual analyses you can create with our software, please see our gallery at https://www.inetsoft.com/evaluate/bi_visualization_gallery/
The next post in this series is at: https://www.inetsoft.com/blog/data-for-machine-learning-churn-detection/
