The Challenge
This state university system with 4 campuses and 45,000 students faced declining retention rates, particularly among first-generation and underrepresented students. Student data was siloed across the Student Information System (Ellucian Banner), LMS (Canvas), advising platform (EAB Navigate), and financial aid systems. Academic advisors had no way to identify at-risk students before they dropped courses or left the university. The provost wanted data-driven insights to guide intervention strategies, but existing institutional research reports were static, quarterly, and arrived too late to act on.
Our Solution
Built a student data hub in Microsoft Fabric that integrates data from Ellucian Banner (enrollment, grades, demographics), Canvas LMS (login frequency, assignment completion, participation), EAB Navigate (advising interactions), and financial aid systems.
Developed a machine learning model in Fabric Data Science that predicts student dropout risk based on 35+ features including GPA trajectory, LMS engagement patterns, financial hold status, credit completion ratio, and advising appointment attendance. The model scores students weekly.
Created role-based Power BI dashboards: provost-level retention KPIs, dean-level college/department comparisons, department chair enrollment trends, academic advisor caseload views with at-risk student flags, and financial aid yield analysis.
Implemented automated alert system that notifies academic advisors when students cross risk thresholds, triggering outreach campaigns through EAB Navigate. Alert rules are configurable by college and student population.
Built faculty self-service analytics allowing professors to see course-level engagement metrics from Canvas and grade distributions compared to department averages, enabling early identification of struggling students.
Results
First-year retention increased from 72% to 83% through data-driven early intervention programs.
Advisors see weekly-updated risk scores instead of waiting for quarterly institutional research reports.
Proactive outreach to at-risk students prevented 850+ dropouts in the first academic year.
All students across 4 campuses included in the predictive analytics model.
Faculty self-service dashboards adopted by professors across all colleges.
Dropout prediction model achieves 87% accuracy with weekly scoring updates.
Implementation Methodology
Phase 1 (Weeks 1-3): Data integration. Connected Banner, Canvas, EAB Navigate, and financial aid systems into Fabric data hub with student identity matching.
Phase 2 (Weeks 4-6): Predictive model development. Feature engineering, model training on 3 years of historical data, validation with holdout semester, and scoring pipeline deployment.
Phase 3 (Weeks 7-9): Dashboard development. Built role-based dashboards for provost, deans, advisors, and faculty with automated alert configuration.
Phase 4 (Week 10): Training and launch. Trained advisors and faculty, configured alert thresholds, and activated the early intervention workflow.