The Effect of AI-Enhanced Instructional Coaching on Student Outcomes

This year-five LEVI RCT uses a high-contrast design and grade-level random assignment to compare high-dosage tutoring combined with an innovative AI-supported coaching model (Saga+HAT) to a business as usual (BAU) control in the state of New Mexico. This study would build upon a sequence of prior highly successful studies demonstrating the efficacy of high dosage tutoring (Bhatt et al., 2024; Ed Labs, 2025; Guryan et al., 2023; Guryan et. al, 2021) to answer the question of whether technology-assisted implementation can achieve similar results at scale. We aim to recruit approximately 50 middle schools, enrolling up to 3,000 students total, with 750 to 1,000 students receiving tutoring. We will evaluate outcomes using end of year test scores for all students.

This project is supported through the Learning Engineering Virtual Institute (LEVI), in collaboration with J-PAL North America.