← All domains
Domain 03 · Domains 1–4
Institutional Algorithmic Decision-Making & Student Services
AI and algorithmic systems wherever they make or inform consequential decisions about people — students, applicants, donors, and employees.
Key AI risks
▹Biased consequential decisions on admission, aid, and retention
▹Opaque automated denials with no human recourse
▹FERPA violations in AI-assisted student analytics
▹Hallucinating public-facing chatbots giving policy guidance
Governed through the RMF Playbook
Each function pairs the outcome to reach for this domain (the Playbook's About) with the Playbook's suggested actions.
Govern
Establish that people — not systems — remain accountable for consequential decisions, and that affected individuals keep clear rights and recourse.
Playbook · suggested actions
Adopt human-alternative, transparency, and appeals policies for automated decisions.
Map
Recognize which decisions AI makes or informs, who is affected, and the legal and privacy obligations that engagement triggers.
Playbook · suggested actions
Inventory the consequential decisions AI touches, the groups affected, and the legal scope.
Measure
Test whether these systems are fair, accurate, and explainable for every population they affect.
Playbook · suggested actions
Run bias, accuracy, and explainability testing before any deployment.
Manage
Keep a human in the loop, provide notification, appeals, and redress, and monitor consequential systems over time.
Playbook · suggested actions
Keep human review, notify affected individuals, and audit systems on a cycle.
Worked case studies · the RMF Playbook applied
10 scenarios
Each step is a Playbook suggested action for that function; the evidence line is the transparency & documentation you keep.
A retention early-alert model
Student Success · predictive analytics
The student-success office wants predictive analytics to flag students at risk of dropping out.
Govern
Apply the right-to-human-alternative and transparency policies — an advisor, not the model, owns every intervention.
Map
Classify the system as consequential, document FERPA scope and affected groups, and file an impact assessment.
Measure
Run disparate-impact testing across race, Pell status, and first-gen, and require explainability of top features.
Manage
Advisors review every flag, students can appeal, and the model is audited quarterly with clear notification.
Outcome — The model informs — never decides — outreach; a bias audit catches over-flagging of Pell students and the feature is removed.
Evidence: Impact assessment + bias-audit report + appeals log
AI-assisted admissions review
Admissions · application reading
The office wants AI to score applications to manage a surge in volume.
Govern
Human-alternative and transparency policies mean readers decide, not the model.
Map
Classify as consequential and document affected applicants and Title VI scope.
Measure
Bias-test scores across protected groups and require explainability.
Manage
A human reads every admit and deny, applicants can appeal, and the model is audited.
Outcome — AI triages workload while humans own decisions, and the bias check clears the model for use.
Evidence: Impact assessment + bias-audit report
Financial-aid packaging optimization
Financial Aid · enrollment analytics
An algorithm optimizes aid offers to hit enrollment targets across the incoming class.
Govern
Transparency and appeals policies apply to aid decisions.
Map
Classify as consequential and identify affected students and equity implications.
Measure
Test for disparate impact on low-income and first-generation students.
Manage
Human review of edge cases, clear notification, and an appeals path.
Outcome — Packaging is efficient without disadvantaging the most vulnerable students.
Evidence: Disparate-impact test + appeals workflow
A policy chatbot advising students
Registrar · student-facing chatbot
A chatbot answers academic-standing and probation questions for students.
Govern
Disclose that AI is involved and require a human-escalation path.
Map
Classify as consequential where it advises on policy outcomes.
Measure
Test accuracy and hallucination rate and verify escalation works.
Manage
Escalate to staff for real decisions and monitor answers over time.
Outcome — Students get fast information with a clear human path for any actual decision.
Evidence: Accuracy test + escalation log
Donor wealth-screening in advancement
Advancement · prospect research
Advancement uses AI to score donor capacity and giving propensity.
Govern
Donor-privacy and prospect-research ethics guidance governs the use.
Map
Identify the individuals affected and the privacy obligations that apply.
Measure
Review the model for biased or intrusive inferences.
Manage
Gift officers decide, and data sources are documented.
Outcome — Prospecting improves within clear donor-privacy limits.
Evidence: Donor-privacy review + data-source log
Enrollment yield prediction
Enrollment Marketing · recruitment
Marketing targets outreach spend using a yield-prediction model.
Govern
Transparency and fairness standards apply to recruitment analytics.
Map
Identify affected prospective students and the equity risk in targeting.
Measure
Check that targeting does not systematically disadvantage groups.
Manage
Keep human oversight of targeting and monitor equity outcomes.
Outcome — Outreach is efficient without inequitable targeting of applicants.
Evidence: Equity review + targeting audit
Remote proctoring flags
Testing Center · online exams
A remote-proctoring AI flags students for suspected cheating during exams.
Govern
Human-alternative and appeals policies bar automated penalties.
Map
Classify as consequential and identify false-positive and accessibility risks.
Measure
Audit false-positive rates across demographic groups and disabilities.
Manage
Humans review every flag, students can appeal, and an alternative assessment exists.
Outcome — Flags inform but never decide, and false accusations are avoided.
Evidence: False-positive audit + appeals record
A housing assignment algorithm
Residence Life · housing
Residence Life uses an algorithm to assign students to housing.
Govern
Transparency and appeals apply to assignment decisions.
Map
Identify affected students and the fairness considerations at stake.
Measure
Test the algorithm for biased or exclusionary patterns.
Manage
Provide human review of exceptions and an appeals path.
Outcome — Assignments are efficient and remain appealable.
Evidence: Fairness test + appeals process
Conduct case triage
Dean of Students · student conduct
AI helps triage and prioritize incoming student-conduct reports.
Govern
Human decision-makers retain authority and students keep due-process rights.
Map
Classify as consequential and identify due-process and bias risks.
Measure
Review triage output for bias and accuracy.
Manage
Staff make all determinations, document them, and allow appeal.
Outcome — Triage speeds workflow without affecting due process.
Evidence: Triage-review notes + due-process record
AI-generated recruitment content
Marketing & Communications · public content
Marketing generates recruitment imagery and copy with generative AI.
Govern
Disclosure standards apply and fabricated campus representations are banned.
Map
Identify the public-facing AI content and its accuracy risks.
Measure
Human review before publication verifies no fabricated diversity or facilities.
Manage
Approve with disclosure and monitor public-facing materials.
Outcome — Recruitment content is authentic and its AI origin is disclosed.
Evidence: Content review + disclosure log
Risk-tier examples
Assistive
Chatbots providing general campus information with no decision authority.
Operational
Donor prospect-research tools and human-reviewed recruitment content.
Consequential
Predictive analytics determining financial-aid eligibility or retention interventions.
Tools & artifacts
Algorithmic impact assessment template
Bias / fairness audit report
Model card
FERPA data-use agreement
Key controls & instruments
Right to human alternative policy
Bias review for admissions / aid / retention
FERPA compliance standards
Notification & appeals for AI-influenced decisions