Predicting Trial Class Attendance in EdTech: A Machine Learning Approach to Funnel Optimization

Customer acquisition in the Educational Technology (EdTech) sector relies heavily on digital marketing and is characterized by high lead volumes, significant acquisition costs, and a distinctive conversion event: the free trial class. While bookings are abundant, they are a weak signal of purchase intent, whereas trial-class attendance strongly predicts downstream conversion. This asymmetry, combined with the heterogeneity of leads and the limited capacity of tutors and sales teams, creates a need for predictive approaches to lead qualification. This thesis designs and evaluates a machine learning–based lead scoring model for predicting trial-class attendance in a real EdTech acquisition funnel. The work is grounded in an applied research project conducted during an internship in the analytics department of Kodland PTE. LTD, a company offering online lessons for children and teenagers. The study formulates the prediction task using only features available at the moment of scoring, with particular attention to the prevention of data leakage, and investigates behavioural features that differentiate users by motivation.