Abstract: Learners in various contemporary settings (e.g., K-12 classrooms, online courses, professional/vocational training) find themselves in situations in which they have access to multiple technologybased learning platforms and often one or more non-technological resources (e.g., human instructors or on-demand human tutors). Instructors, similarly, find themselves in situations in which they can provide learners with a variety of options for instruction, practice, homework, and other activities. We seek data-driven guidance to help facilitate intelligent instructional “hand offs” between learning resources. To begin this work, we focus on an important element of self-regulated learning, namely help seeking. We build classifier models based on proxies for learner prior knowledge and data-driven inferences about learners’ disengaged behavior (e.g., gaming the system) and affective states (e.g., confusion) to determine the extent to which (and when) learners tended to seek out help via human tutoring while using an intelligent tutoring system for mathematics. Insights into cognitive, behavioral, and affective factors associated with help seeking outside of a system will drive future work into providing automated, intelligent guidance to both learners and instructors. We close with discussion of the limitations of the present analysis and avenues for future work on intelligently guiding instructional hand offs