Scholars at University of Phoenix have published a conceptual framework that maps how generative AI and predictive analytics can operate as a single, closed-loop support system for online higher education. The 16-stage model, introduced in the International Journal for Educational Media and Technology, addresses a practical gap: institutions often deploy these technologies separately, missing the chance to connect data-driven early warnings with AI-generated, faculty-reviewed interventions.
The paper, authored by Pamayla E. Darbyshire and Carl Beitsayadeh, both research fellows in the university's Center for Educational and Instructional Technology Research, draws on systems theory and the learning analytics cycle. It organizes the process around four interconnected phases-data and modeling, risk-aligned interventions, monitoring and feedback, and institutional refinement.
A 16-stage closed-loop model
The framework begins with institutional data systems such as student information systems and learning management platforms. Predictive models analyze these data streams for signals like disengagement, late submissions, or declining performance, translating them into risk tiers. From there, generative AI can produce tailored interventions: personalized messages, formative quizzes, study plans, or resource recommendations.
What distinguishes the model is its insistence on human judgment as a core interpretive layer. Faculty review and contextualize every AI-generated output, ensuring that support reflects both data-informed insight and a human understanding of the learner's situation. "The goal is not to replace the human relationship in learning, but to help educators respond with greater context, clarity and care," Darbyshire said.
The cycle then feeds back into institutional refinement. Audit trails, governance policies, and ongoing evaluation of accuracy and equity allow the system to improve over time. The authors argue this continuous feedback loop is what transforms a collection of tools into an adaptive socio-technical ecosystem.
Bridging two technologies that are often siloed
Much of the current conversation around AI in education treats predictive analytics and generative AI as separate domains. Predictive models flag who might need help. Generative tools create content. The paper's central argument is that these functions become more powerful when they are designed to inform one another within a single, governed system.
"This framework brings them together within a single adaptive system, where data-informed insights, AI-enabled support, faculty judgment, and institutional oversight operate as interconnected parts of a continuous improvement cycle," Beitsayadeh said.
The model also surfaces practical considerations that institutions must address before implementation. These include building secure and interoperable data infrastructure, preparing faculty to interpret predictive outputs and AI-generated recommendations, and establishing clear policies for data access and model evaluation.
Ethical safeguards and institutional readiness
The paper does not treat ethics as a separate module. Instead, it weaves safeguards related to transparency, fairness, bias monitoring, student trust, and human discretion directly into the operational stages of the framework. The authors note that responsible AI implementation requires more than adopting new technology. It demands governance structures that evaluate accuracy, equity, and alignment with institutional goals from the start.
The publication also identifies the need for feedback loops that allow institutions to refine interventions over time, rather than deploying a static, set-and-forget system. This includes monitoring for unintended consequences and designing AI tools that support, rather than replace, instructor judgment.
Why this matters for educators and instructional designers
For professionals working in online higher education, the framework offers a specific, actionable blueprint. It moves beyond broad calls for AI adoption and shows how to sequence data ingestion, predictive modeling, generative feedback, and faculty review into a single, repeatable cycle. The emphasis on educator judgment as a required checkpoint-not an afterthought-provides a model that preserves the human relationship at the center of learning while enabling earlier, more personalized support for students.
Your membership also unlocks:
