Keeping an eye on students: evidence from a low-cost early alert system

We evaluate an early alert system intervention at a German university, where at-risk students received warning emails based on dropout probability. While the average effect was insignificant, estimates point to heterogeneous responses. Previously active students, particularly in non-STEM fields, reduced exam registrations by up to two courses and modestly improved grades, indicating strategic workload adjustment. Inactive students showed no response, and dropout probability increased slightly in structured programs. Causal forests reveal further treatment heterogeneity. Our findings point towards the importance of pre-treatment academic activity and intervention timing, indicating that warnings must align with student engagement and program structure.