Abstract

Universities hold extensive academic and operational data, yet fragmentation and weak governance limit their use in decisions. This design science study developed an artificial intelligence (AI)-driven educational management information system (EMIS) architecture and evaluated its content appropriateness before implementation. A bounded thematic synthesis of 25 verified sources informed seven design requirements, operationalised in five layers, two cross-cutting planes, and a human feedback loop. Seven experts in architecture, AI and analytics, and university information-system administration completed a two-round modified Delphi evaluation of 14 items. All held doctorates in ICT for education and came from one national context. After Round 2, item-level content validity indices ranged from .86 to 1.00, the scale-level index reached .98, the overall mean was 4.65 (SD = 0.52), and item endorsement was at least 85.7%. Gwet’s AC1 was .96; Fleiss’ kappa and the intraclass correlation did not meet the pre-specified criteria under pronounced ceiling effects. Explainability and institutional feasibility remained the lowest-rated areas, with residual reservations from institutional administrators. Six experts accepted the final design without further revision. These findings support content appropriateness for this expert profile and context and conditional progression to a prototype; they establish neither operational feasibility nor effectiveness in other settings. The contribution is an evidence-traceable integration of established architectural patterns with university decision workflows, governance responsibilities, and staged implementation. Incomplete screening records and possible conformity between Delphi rounds limit the inference.

This work is licensed under a Creative Commons Attribution 4.0 License.