AI-Supported Lecturer Decision-Making in Higher Education: A Socio-Technical Perspective

Synthesis: A PRISMA-guided systematic review of 27 empirical studies published between 2016 and 2025 develops an empirically grounded taxonomy of AI-supported lecturer decision-making, organized across four interdependent dimensions: decision types, AI system types, student data and learning outcomes. Eight decision types are identified, but AI support clusters tightly in three of them — instructional, Feedback and Assessment decisions — while emotional, ethical, administrative, curriculum and learning-environment decisions remain underrepresented. Learning analytics dashboards were the most frequently reported system type, and they overwhelmingly made textual and log data visible. Read socio-technically, the reviewed systems rendered behavioral student data actionable and thereby channelled lecturers' attention toward behavioral outcomes, leaving motivation, Metacognition, emotion and learning-environment concerns comparatively unsupported.

Key Findings

1. Eight lecturer decision types were identified inductively: instructional, curriculum, assessment, feedback, learning-environment management, emotional support, administrative and ethical decisions. Decision timing was also coded as pre-interaction, real-time or post-interaction.
2. Support was unevenly distributed. AI support concentrated on instructional, feedback and assessment decisions; emotional, ethical, administrative, curriculum and learning-environment decisions were underrepresented in the reviewed corpus.
3. Learning Analytics Dashboards (LADs) were the most frequently reported system type, ahead of Intelligent Tutoring Systems and agentic AI systems. Interfaces included dashboards, chatbots and hybrid dashboard-chatbot environments.
4. The dominant data flow was narrow: text and log data processed through dashboards into behavioral outcomes and thence into instructional, assessment and feedback decisions. Pathways involving Multimodal AI and interaction-based data, agentic AI systems, and cognitive, metacognitive, motivational and affective outcomes were comparatively thin.
5. AI methods reported across the studies included supervised and unsupervised learning, natural language processing, computer vision, explainable AI and symbolic AI, with functions coded as monitoring, feedback, assessment, personalization and coordination.
6. The authors frame the finding as an attention problem rather than a capability problem: what AI systems make visible shapes which decisions lecturers consider, so the decision space is being narrowed by data availability rather than by Pedagogies and Teaching Strategies.

The socio-technical reading

The review deliberately separates what the studies found from what the authors read into the distribution. Its interpretation is that making behavioral data visible and actionable is not neutral: it privileges decisions that behavioral indicators can inform. Motivation, metacognition, emotion and learning-environment design are not unsupported because lecturers consider them unimportant, but because the systems in the sample rarely rendered them visible. This leads the authors to pose the guiding question of the review — how alternative forms of student data visibility and attentional guidance could broaden the decision types AI supports — and to note the design implication that systems should surface interpretive context rather than more volume.

What this means for practice

• Instructors. Inventory the decision types your current tools actually inform, then deliberately reserve attention for the emotional, ethical and curriculum decisions the review found AI ignores. Behavioral dashboards nudge you toward what they display, not toward what students need.
• Administrators. Ask vendors which of the eight decision types a system supports before purchase, and treat the unsupported categories — emotional, ethical, administrative, curriculum and learning-environment — as gaps to cover through staffing rather than features to assume.
• Designers. Surface interpretive context rather than more data volume, and add indicators for Motivation, Metacognition and emotion so that non-behavioral student states become visible enough to act on.
• Researchers. Adopt the eight decision types and the timing codes (pre-interaction, real-time, post-interaction) as a coding scheme, and target the thin pathways the review maps: Multimodal AI and interaction-based data, agentic AI, and affective outcomes.

Limitations

• The evidence base is small and uneven: several categories rest on very few studies (agentic AI systems, n = 4; image and biological data, n = 1 each), so the reported frequencies are indicative rather than representative.
• Publication bias applies, since studies reporting successful implementations are more likely to be published, and the exclusion of preprints and non-English publications further limits coverage.
• AI in education is developing rapidly and the search was completed in October 2025, so the distribution reported is a snapshot of the literature at that point rather than a stable characterization of the field.
• The review synthesizes how systems were designed and described by their authors, not how lecturers used them; its socio-technical reading therefore concerns the informational conditions these systems create, not their effects on decision-making. The taxonomy describes how the literature is distributed, not causal relationships established by the included studies.

Connected Concepts

• Learning Analytics — dashboards as the dominant system type
• Teaching — decision-making as the core of lecturer practice
• Assessment — one of the three well-supported decision types
• Feedback — likewise concentrated
• Learning Design — curriculum and learning-environment decisions as gaps
• Higher Education — the review's institutional context

Connected Articles

• From Learning Analytics to Educational Interventions: Enhancing Decision-Making and Learning Design — From learning analytics to educational interventions
• Understanding the Role of Learning Analytics and Generative Artificial Intelligence on Decision-Making and Learning Design Practice in Higher Education — Educator reasoning with learning analytics and GenAI
• Architecting an AI-Driven Decision Support System for Enhanced Online Learning and Assessment — AI decision support in online learning assessment
• A scoping review of generative AI-powered agentic AI in education: Research landscape, agentic capabilities — Agentic AI in education scoping review
• Federated and Explainable Learning Analytics for Privacy-Preserving Academic Risk Modeling Across Heterogeneous Educational Institutions — Federated explainable learning analytics

Citation

Köroğlu, M. N., Mayrhofer, J., Steinmaurer, A., Wintersberger, P., Dennerlein, S. M., & Weinhandl, R. (2026). AI-Supported lecturer decision-making in higher education: a socio-technical perspective. International Journal of Educational Technology in Higher Education, 23(1), 49.