Learning Analytics (LA) has been identified as one of the top technology trends in higher education today (Johnson et al., 2013). LA is based on the idea that datasets generated through normal administrative, teaching, or learning activities--such as registrar data or interactions with learning management systems--can be analyzed to enhance student learning, academic progress, and teaching practice. Examples of LA projects in colleges and universities include Purdue University's "Course Signals" system, an early-alert notification for struggling students, and Austin Peay State University's "Degree Compass," a course recommender program based on predictive analytics. This article describes one large-scale LA initiative at the University of Michigan (U-M) to improve performance for thousands of students in gateway physics courses. The goal is not only to describe the development and implementation of this unique initiative in STEM education but also to discuss how the approach can help meet some of the challenges to more widespread LA adoption.