The Challenge

Our client, a pioneer in online learning, wanted to improve their existing models used to predict student lifetime value (LTV). Lifetime value is a prediction of the future dollar value that can be attributed to the future relationship with a customer. 

The Solution

The client had developed models to provide LTV predictions for specific dates for new and existing students. However, predictions between those dates needed to be interpolated from general student data. Elder Research was contracted to create a new student model framework to provide:

  • Daily predictions for 3+ years into the future
  • Improved granularity (predictions for each cohort of incoming students and for each existing student)
  • Improved accuracy

To predict student lifetime value the team first captured an “experience footprint” for all students and then quantified the impact of those experiences to predict the likelihood a student will still be enrolled on future dates. The new student model framework employs Kaplan-Meier survival curves to learn survival patterns of students to a daily granularity and provides daily probability of survival for the groups of new and incoming students. For the existing student model, a Cox Proportional-Hazards framework was leveraged. 


The improved prediction accuracy was widespread across different grades. The two new models outperformed the incumbent models in 22 of 26 grade-by-grade comparisons. Accuracy improved 5.53% for incoming students and 13.25% for existing students.