In 2018, less than 20% of students in low and middle-income countries met minimum proficiency levels in mathematics (World Bank, 2017). The COVID-19 pandemic has further deepened this learning crisis, with schools worldwide closed for an average of seven months between February 2020 and February 2022, leading to significant learning losses (World Bank, 2022).

Increased access to AI-powered education tools presents a promising solution to this crisis. Among these tools, computer-adaptive learning (CAL) softwares have emerged as a potential solution to the “2-sigma problem” - the absence of instructional methods that are as effective as one-to-one tutoring (Bloom, 1984). In classrooms where students are not equally prepared, and teachers may face limitations in catering to this variation, CAL softwares use students’ responses to predict their individual learning levels and tailor learning activities to match their specific needs, all at a fraction of the cost associated with private one-on-one tutoring.

This study evaluates the impact of combining an AI-powered CAL software (ALEKS) with group tutoring through a multi-stage randomized controlled trial (RCT) in the Dominican Republic. By integrating CAL with targeted group tutoring for students most in need, the intervention aims to accelerate learning recovery and reduce learning disparities. The study design and survey instruments will allow us to assess whether the use of adaptive learning software, by itself or when combined with group tutoring, can improve teaching efficiency and student outcomes. Preliminary results suggest that the CAL component led to large and positive effects on student performance, with improvements ranging from 0.2 to 0.3 standard deviations.

Keywords: Technology, education, development, RCT.