For too long, risk predictive modeling has represented the core of machine learning analytics, leaving real-world evidence (RWE) of treatment effectiveness untouched. As many have found out, predictions alone do not lead to student success outcomes, often being used to discourage students. Further, their opaque and nonlinear nature can lead to human suspicions and more often an exercise of explaining scores instead of taking actions. Randomized controlled trials (RCTs) are slow, expensive, and sometimes unethical. Furthermore, population heterogeneities can make such RCT results difficult to replicate. That is, treatments need to be personalized to various student segments and measuring a single treatment across all student groups can dull the true effects of interventions designed for heterogeneous student populations, including students who are likely to experience equity gaps.