Predictive Modeling for Optimizing Teacher Placement in Rwanda

Teacher turnover is a major challenge in Rwanda, where approximately 20 percent of primary teachers leave their positions each year. Current centralized placement practices assign teachers randomly to schools within districts, often misaligning teachers' preferences with outcomes and leading to staffing instability, higher administrative costs, and reduced educational quality.

This project evaluates two AI-enabled interventions through a two-tiered randomized controlled trial embedded in Rwanda's national teacher placement system. The first intervention is an interactive chatbot that supports teachers during the application process by providing personalized placement probability estimates, explaining trade-offs, and helping teachers make better-informed choices. The second is a machine learning-based matching algorithm that replaces random within-district assignment with data-driven matching optimized for retention and teacher effectiveness.

Using administrative data from the Teacher Management Information System (TMIS) and Comprehensive Assessment Management Information System (CAMIS), the study measures impacts on placement satisfaction, teacher retention, value-added, and student learning outcomes. The chatbot experiment randomizes at the individual level among approximately 80,000 applicants; the matching experiment randomizes 30 districts into three arms (status-quo, preference-based, and ML-optimized matching).

Implemented in partnership with Rwanda's Ministry of Education and IPA, the interventions are integrated into existing government infrastructure and designed for immediate nationwide scale-up if effective.