Scaling AI-Powered Math Tutoring Across Diverse Educational Contexts: A Full-Scale Randomized Controlled Trial in Ghana 

This study proposes a large-scale RCT to evaluate the effectiveness and scalability of Rori, an AI-powered math tutor, when implemented across diverse educational contexts in Ghana. The intervention provides students with personalized, adaptive math instruction for one hour weekly, addressing the acute shortage of quality tutoring in Sub-Saharan Africa, where less than 15% of students achieve minimum math proficiency. Building on a pilot that demonstrated feasibility and provided crucial insights into contextual factors affecting intervention effectiveness, this research addresses the critical question: "Can an AI-powered tutoring system deliver consistent learning gains when scaled across different school types while maintaining implementation fidelity?"
We hypothesize that students receiving access to Rori will demonstrate significantly greater learning gains compared to a control group, while implementation effectiveness will vary by school type and contextual factors. We will recruit 45 schools in Ghana, with an anticipated sample of 8,100 students, using classroom-level randomization to maximize statistical power, while incorporating lessons learned about contamination control from our pilot. This research directly serves low-income students in LMICs who cannot afford private tutoring and have low-quality academic infrastructure, providing evidence for scaling educational technology interventions across similar contexts.

This project is supported through the Learning Engineering Virtual Institute (LEVI), in collaboration with J-PAL North America, which builds off of a previous pilot project.