This paper presents the impact evaluation results of the Colombian program Todos a Aprender (Everyone Learning Program, ELP), a multi-level intervention targeting low-performing schools. The main objective of the program was to increase math and language test scores of these schools through on-site teacher training, principal training and textbooks for students. Using census data from public schools containing detailed longitudinal information since 2010, the starting year of the program, and taking advantage of targeting rules based on dropout and grade repetition rates we fit a fuzzy regression discontinuiTY design to estimate program impacts. We also fit a difference-in-difference matching model as well as blocking with regressions to estimate the ATT impact of the program, based on observed characteristics used in the targeting process. Overall results indicate no significant impact of the program on test scores, grade repetition nor dropout rates. Additional analyses from a representative sample of 400 schools collected in the field suggest that deficiencies in the program's design and implementation could explain the lack of significant program impacts.