AI-Based Analytics as Feedback to Teachers: Bridging Classroom Data to Pedagogical Action

Abstract

Although multimodal classroom data and AI-driven learning analytics continue to expand rapidly, most systems are still off the periphery of daily teacher practice by providing retrospective metrics and student-level risk indicators that often do not directly lead to actual pedagogical intervention. This design-based research project, implemented in three iterative cycles involving 18 secondary mathematics and English teachers and their 420 students, bridged this "knowing-doing" gap by creating and testing Pedagogy Mirror: a teacher-facing AI feedback system that turns audio, video, LMS, and interaction data into weekly Pedagogical Insight Briefs with task-level diagnostics, annotated teaching moments, explanatory why statements, and classroom-ready what-to-try-tomorrow moves. Results indicated that teachers greatly preferred moment- and task-level information to individual labelling, nearly all of them (81-86) accepted questioning and pacing suggestions based on the eight principles of human-centered design of the study, and the effect of the study significantly changed practice: data-informed lesson planning increased by almost a quarter (12 to 68), formative assessment and differentiation strategies doubled, and teachers felt much more confident working with heterogeneous classrooms (d = 1.12). The interaction between students had increased but by a low value (d = 0.38) even though it was not significant in short term achievement effects as was anticipated in the duration of 10 weeks. The first cases of resistance and algorithmic bias were replaced by a developing teacher-AI collaboration based on co-design and joint calibration. The study concludes that when analytics are deliberately engineered as reflective, agency-preserving feedback partners rather than surveillance tools, they can meaningfully bridge classroom data to pedagogical action, offering a scalable model for human-centered learning analytics that places teachers’ professional judgment at its core.

Keywords: Learning Analytics, Teacher-Centered AI, Pedagogical Feedback, Design-Based Research, Multimodal Data, Reflective Practice.