Case study: Walk me through how would you predict who will renew their subscription next month? What data would you need to solve this? What analysis would you do? Would you build predictive models? If so, which algorithms?

Data Science Interview QuestionsCategory: Data ScienceCase study: Walk me through how would you predict who will renew their subscription next month? What data would you need to solve this? What analysis would you do? Would you build predictive models? If so, which algorithms?
MockInterview Staff asked 6 months ago
2 Answers
Paras answered 6 months ago

Tips:

  1. This is a churn prediction problem. There are lot of data that you can look and it will involve lot of feature engineering.
  2. You should mention that you worked with business leaders and analyst to get qualitative feedback and used it as a starting point to identify features
  3. Then you can say after you had identified key features, you used algorithms like Logistic regression, Random Forest, SVM, etc to build the model

Krishna answered 4 days ago

Determining who will renew subscription is a churn prediction problem,but also a customer engagement problem. Some of the data that is relevant to this predict are features related to engagement which will vary based on the type of subscriptions,

  1. For digital streaming/video subscriptions, engagement can be determined by looking at the time series of the hours for the previous months and determining patterns. Typically, the user will begin at some hours, and if the user is engaged the hours will increase and then become steady or even decrease slightly, like a normal distribution. Studying a large customer base and arriving at the patterns will help to predict which customers will renew their next month. The hours and engagement patterns vary depending the user’s occupation, if the user or user and his/her family also participate, genre that the user is partial to(specialized categories for example, in movies, foreign-language/indie are rarer or in Kids Coaching, subject like Robotics), length of membership.  All of this should be part of the data set for prediction.
  2. For fitness/diet program subscriptions: The data set would comprise of the hours of use, occupation, how many members from the same household
  3. For all others like retail/magazine subscriptions, it boils down to tracking user engagement.

Based on the domain, the exact data set would change. I would perform feature engineering to determine outliers, seasonality and trends. Finally use a classification algorithm like Logistic Regression or Random Forest to predict the probable users who will likely renew their subscription.

Your Answer