|
Abstract:
|
A common customer acquisition strategy in the software industry is to offer consumers a limited period to try the product for free. However, no research has empirically examined the issue of optimizing the length of these free trials. Using data from a large-scale field experiment, we study the relative effectiveness of three trial lengths: 7, 14, and 30 days. We find that the shortest trial length of 7 days is the best option at the aggregate level. Next, we use the causal forest methodology to uncover the rich heterogeneity in treatment effects and quantify effects at the individual level. We discuss and empirically show that the causal forest algorithm overfits to the training data, especially when the signal to noise ratio is low. Alternatively, we suggest using the profit evaluation criterion for hyper-parameter optimization to resolve the overfitting issue. Using the modified causal forest, we design a personalized free trial assignment policy and evaluate its effectiveness. Finally, we correlate users' optimal free trial length with their pre-treatment and post-treatment variables to offer suggestive evidence on why different trial-length works better for different groups.
|
