Alerta Escuela — Peru's National ML-Based School Dropout Early-Warning System

At the end of 2020, Peru's Ministry of Education (Minedu) launched Alerta Escuela, a nationwide web platform that uses a machine-learning model, built with World Bank support, to estimate each student's risk of dropping out of school across Peru's initial, primary and secondary education levels.

Give school principals and tutors a dashboard flagging students at high, medium or low dropout risk — drawing on student characteristics, academic performance history, family context, school service characteristics and household socioeconomic information — to enable early intervention.

Minedu and the World Bank not only launched the tool but rigorously evaluated it: a 2021–2023 implementation evaluation examined platform access and use, and a companion study tested SMS reminders grounded in behavioral-economics principles to nudge principals toward using the platform.

The implementation evaluation found that only 7% of schools with high- or medium-risk students actually accessed the platform, and just 2% downloaded its management and pedagogical guidelines; many principals reported not understanding the model or the variables behind its risk scores, and found the tool too time-consuming because it only requested information rather than offering solutions. The SMS-nudge study found that reminders significantly increased access to and use of the early-warning system, but that increased usage did not translate into more preventive action by school staff, and ultimately produced no measurable reduction in school dropout. Connectivity was a substantive constraint: 36% of school facilities lacked reliable access.

This is a genuinely mixed, evidence-rich case: a technically capable national ML system for an important problem, evaluated honestly, whose central finding is that better data access alone does not change front-line behavior or outcomes without simpler interfaces, clearer guidance, and stronger connectivity. Peru continued to iterate on the platform after these findings; the core lesson — that early-warning systems must be judged on downstream action and outcomes, not platform usage — is a valuable evidence-based cautionary result.