OECD Digital Education Outlook 2026: Exploring Effective Uses of Generative AI in Education

In the last 10 years, the field of learning analytics has increasingly expanded from student-facing dashboards and systems to analytics that improve programme evaluation, curriculum design, and course delivery in higher education (Greer et al., 2016). Although a recent review of the literature concluded that there is a lack of curriculum analytics studies investigating how these AI systems influence higher education stakeholders (De Silva et al., 2024), we summarise case studies that offer clear perspectives into how curriculum can be designed using machine learning models and AI trained on enrolment, course, and other institutional data. As generative AI becomes increasingly capable in explaining learning analytics to stakeholders (Yan et al., 2025), innovation informed by AI will increasingly shape institutional workflow practices.

For instance, recent work has demonstrated how curriculum analytics can be enhanced with statistical and psychometric techniques to identify inequities in course difficulty and monitor changes over time. Baucks et al. (2024) introduced Differential Course Functioning (DCF), an Item Response Theory (IRT)-based method that controls for overall student performance while detecting systematic differences in course-specific success rates between student groups. Applied to data from over 20 000 undergraduates, the Differential Course Functioning method revealed patterns linked to disciplinary alignment and preparedness, guiding targeted interventions for students taking courses outside their major and transfer students. In a complementary study, Baucks et al. (2024) applied IRT to quantify temporal shifts in course difficulty, finding a marked downward trend during the COVID-19 pandemic and proposing IRT- adjusted pass rates to mitigate the confounding effects of fluctuating cohort performance. Both approaches provide actionable evidence for policymakers, accreditation bodies, and student advisors aiming to improve fairness and consistency in academic programmes.

Analytics have also been used to address mismatches between credit hours and actual student workload, offering an actionable basis for curriculum analytics. Credit hours, while central to degree requirements and course planning, explained only 6% of the variance in how students perceive their course workload in Pardos et al. (2023), whereas learning management system (LMS) features based on forum, assignment, and submission activity explained six times more variance (36%) in measures of time load, mental effort, and psychological stress. LMS indicators such as number of assignments and late-semester course drop ratios as well as historical course GPAs provided a more accurate reflection of the student experience, giving institutions greater confidence in these measures as a basis for action. Building on this, Borchers and Pardos (2023) developed Course Load Analytics (CLA), a predictive model that integrates LMS and enrolment features to estimate perceived workload at the course and semester level. Applied across an entire university catalogue across a full undergraduate degree duration, CLA revealed that first-semester students-particularly in STEM fields-often carry some of the heaviest predicted workloads despite low credit-hour counts (Figure 11.3), a hidden load linked to higher attrition. Such meso-level insights position CLA as a practical tool within curriculum analytics, enabling institutions to redesign programme structures, adjust course sequencing, and align workload expectations with student capacity. Because CLA’s modelling approach generalises to new courses and contexts, institutions can deploy it broadly to monitor and balance workloads, improving retention and the overall first-year experience.

Looking ahead, curriculum analytics research continues to span a wide range of curriculum-related areas, including programme structures, course sequencing, competency attainment, workload measurement, and curriculum- employability alignment. Studies examine curriculum components from multiple angles, such as mapping prerequisite networks, identifying instructional bottlenecks, tracking competency coverage, modelling student progression pathways (De Silva et al., 2024), and analysing elective course selection strategies (Srivastava et al., 2024).

Recent work has also expanded curriculum analytics beyond student modelling and workload analysis to include the automation of academic record processing through direct application of generative AI in institutional workflows. For instance, Bhaskaran and Pardos (2025) conducted a comparative study of Optical Character Recognition (OCR) and vision-language model pipelines for transcript evaluation, a critical task in credit transfer and course articulation. They showed that combining OCR with semantic reasoning through multimodal models such as GPT-4o1 and Claude 3.7 achieved extraction accuracies above 90 percent, reducing the manual effort needed to align courses and grades across institutions, and thus facilitate transfer and degree/credit recognition across institutions domestically (and potentially internationally).

By converting unstructured transcript data into structured formats suitable for downstream analysis, these methods help connect administrative processes with analytical insights in curriculum design and enable scalable pipelines for workload modelling and course equivalency mapping. In the future, Curriculum Analytics research is likely to close the gap between research applications and practical impact gained through sustainable deployment of tools, as identified by De Silva et al. (2024), by improving efficiency in syllabus, learning objective, and content generation using large language models (Sridhar et al., 2023) and by supporting the wider deployment of university-level initiatives such as course load analytics.

While currently being developed in higher education, where course diversity is much greater than at the school level, those techniques could also be used for middle and upper secondary education in education systems where students can choose different tracks, majors or options to get their high school diploma. This may either highlight the relative difficulty and workload of choosing different study paths or change their perception or design and provide more equal opportunities to all students to study in higher education, regardless of their subject preferences.

AI systems deployed in institutional workflows aim to advance a range of administrative and student success outcomes. While specific goals may vary by domain (e.g. advising, transfer, content infrastructure), a shared emphasis lies in improving efficiency, personalisation, and educational equity. Below are key outcome categories and associated metrics that institutions may target:

Student progress and completion Time-to-degree: Reduction in semesters or credits attempted to graduate. Curriculum analytics can help identify student mismanagement during academic planning (e.g. workload overload) and help learners arrive at more balanced course sets (Pardos, Borchers and Yu, 2023).

Credit applicability: Increased proportion of transferred or enrolled credits that count toward degree requirements. AI can help match more course equivalencies that help students graduate (Pardos, Chau and Zhao, 2019).

GPA and academic performance: Stability or improvement in course and cumulative GPA under AI-supported planning systems. Curriculum analytics can help identify which parts of a course are challenging or lack instructional effectiveness, guiding more effective re-design.

Retention and persistence: Improved term-to-term and year-to-year retention, particularly among at-risk or transfer students. Curriculum analytics can identify which courses are particularly challenging for transfer students (Baucks et al., 2024) and help allocate institutional resources.

Transfer and articulation outcomes Articulation coverage: Expansion in the number and accuracy of course equivalencies across institutions (Goulart and Pardos, 2025).

Credit mobility: Decrease in articulation loss, especially for community college students, through better support of equivalency officers through AI-based articulation recommendations (Xu, Pardos and Pai, 2023).

Degree completion for transfer students: Increase in successful 4-year degree attainment among students transferring from 2-year institutions.

Time and administrative effort saved: Reduction in manual reviews required by articulation officers and advisors or total review time (Xu, Pardos and Pai, 2023).

Advising and personalisation Workload fit: Better alignment between student capacity and course intensity, potentially reducing dropout or course failure, which has been shown to be associated with student workload (Borchers, Xu and Pardos, 2025).

Well-being indicators: Improved student reports of stress, burnout, or overload through course workload analytics when supported by personalised advising tools (Pardos, Borchers and Yu, 2023; Borchers, Xu and Pardos, 2025).

Advisor efficiency: Enhanced capacity of advisors to manage caseloads through intelligent recommendations (e.g. regarding workload or course sequences: Khan and Polyzou, 2024) and predictive alerts.

Curriculum and learning infrastructure OER discoverability and reuse: Increased alignment between institutional curricula and open educational resources (Li, Pardos and Ren, 2024). Improved alignment models could significantly help reduce human authoring and search time.

Content production efficiency: Reduction in faculty time spent authoring assessments or tagging materials, through generative tools.

Curriculum analytics: Enhanced ability to detect gaps, redundancies, or misalignments in curriculum via structured metadata and content classification.