Predicting Conversion of Free Trial Users to Paying Customers to Increase Sales by Developing an Effective Free Trial Program

Project Details


Fall 2019


Chawisa Mahajindaplan, Thuc Han Bui, Jariya Tienmongkol, Ryjill Roa





Migo is a digital company that utilizes satellite technology to serve digital contents to consumers who live in urban areas that have limited data connection due to high internet prices and find it too expensive to subscribe to the existing streaming services in the market.
As a recently launched services application, we are looking to find a way to increase sales by recruiting new customers. But engaging new customers is very challenging, especially for a new brand where people have no previous brand awareness. Free trial promotions are a good way to engage new customers but its effectiveness should be evaluated to ensure that return on investments and efforts are maximized.
For starters, we are looking for an increase in the conversion rate of free trial customers to a paying customer within one month from their first subscription to a free trial by 10%. To support the business goal, we will focus on predicting whether free trial customers will purchase a subscription for their next transaction within one month after their first free trial ends.
We used four tables as inputs to the prediction: (1) transactions, (2) download, (3) engagement and (4) title. The tables were cleaned, filtered then merged to reflect one record (user) per row while the columns from the four tables were used as columns in the merged table. A new column was created called ‘Customer Spending’ with a ‘Yes’ when the user converts to a paid user for their next transaction within one month from their first free trial while ‘No’ if otherwise. Three predictive algorithms were applied to the merged data - Naive Bayes, Random Forest and Lasso Regression - and the results were compared using a lift and gains chart.
Comparing the accuracy and lift and gains charts from the three algorithms, we found that the Random Forest showed the best performance with a top decile of 4.089 which implies that the Random Forest model can perform approximately 4 times better than random selection in the highest 10%. Based on the variables that had the most predictive power, the unique and absolute engagement of the users, followed by the two most effective free-trial promotions such as the 2-Day Pass and PisoMigo_trial, and download space had the most significant impact on the conversion of users to a paying one. Based on this, the Marketing team can consider providing short-periods of free trial promotions and creating a rewards program to encourage users to be more engaged with the application which will therefore, increase the conversion rate of free-trial users to paying customers.

Application Area: