Calculating and Predicting Customer Lifetime Value (CLV) for Smile.io A Comparative Study between Statistical and Machine Learning Models by Milad Beigi Harchegani – in Collaboration with Smile.io

Abstract In the fast-changing world of e-commerce, understanding how much value a customer can bring over time is very important for business decisions. Customer Lifetime Value (CLV) is a key metric that helps companies estimate the total revenue they can expect from a customer during their relationship. This study, developed in collaboration with Smile.io, a loyalty platform for Shopify merchants, compares different approaches for calculating and predicting CLV. The focus is on both statistical models and machine learning models. The project begins by reviewing common methods for calculating CLV, including both historical and predictive techniques. Then, it explores and implements a selection of models. These include probabilistic models such as BG/NBD and Gamma-Gamma, as well as machine learning models like Linear Regression, Random Forest, and XGBoost. All models were trained and evaluated using real transactional data from Smile.io merchants. The features used included standard RFM metrics (recency, frequency, and monetary value) and customer loyalty indicators. To compare the models, several evaluation metrics were used, such as Mean Absolute Error and Root Mean Squared Error. The study considered both predictive accuracy and how practical the models are for real-world use. Probabilistic models were more interpretable and easier to apply with limited data, while machine learning models offered better performance overall. The results showed that customer frequency and average spending were the most important predictors of CLV, and Loyalty-related features had a limited effect on improving predictions. This research contributes in two ways. First, it offers a comparison of CLV models in the specific context of a loyalty program platform. Second, it provides a practical and repeatable framework that Smile can use to estimate customer value for merchants. Although the study has some limitations, such as the lack of data on customer satisfaction or marketing interactions, it offers useful insights for future development of CLV tools at Smile.io and similar companies.