Freemium-to-paid conversion; SaaS analytics; machine learning; class imbalance; explainable AI.

Abstract

This study reframes freemium-to-paid conversion in Software-as-a-Service (SaaS) products, moving away from treating it as a single moment of decision and toward understanding it as a cumulative behavioural pathway of value discovery and deepening engagement. Drawing on real behavioural usage data from a SaaS project-management platform covering 17,024 users over a twelve-month period, and following the CRISP-DM methodology, three supervised machine-learning models—Logistic Regression, Random Forest, and XGBoost—were trained after addressing severe class imbalance (a 6.96% conversion rate) through SMOTE, stratified sampling, and threshold optimisation. Models were evaluated using precision, recall, F1-score, and the area under the ROC curve, and the strongest model was interpreted using SHAP values.

The results show that converting users carry a distinct behavioural signature: they log in more often, spend longer per session, and explore a wider range of features, with premium-feature interaction, engagement consistency, and engagement growth emerging as the strongest determinants. The trajectory of engagement over time proved decisive—users on an increasing path converted at 19.4% against only 1.8% for those in decline. XGBoost achieved the best performance (F1 = 0.526, ROC-AUC = 0.804) by a statistically significant margin. The study thus offers an interpretable, actionable framework enabling product and marketing teams to target high-potential users early and on the basis of evidence rather than guesswork.