Higher Education's Next AI Challenge: Turning Access Into Institutional Value

For the past few years, much of the conversation about AI in higher education has centered on access. Should students be allowed to use it? Which tools should faculty and staff have? How should academic integrity policies change? What information is safe to enter? Should the institution provide an approved tool rather than allow people to use whatever is publicly available?

Those were - and still are - important questions. Colleges and universities had to respond quickly to technology that faculty, staff, and students were already using. CIOs and other institutional leaders have done a tremendous amount of work to evaluate tools, manage risk, develop policies, and help their communities understand what responsible use should look like. That work created an important foundation.

Now, many institutions are moving into a second phase of AI adoption. Not every institution is in the same place, and even departments within the same institution may be at very different stages. But the conversation is beginning to shift. The question is no longer just, “How do we give people access to AI?” The next questions are harder:

• Which use cases are worth continued investment?
• What value are they actually creating?
• How should different uses be governed?
• Is the institution’s data ready to support them?
• Who will own and maintain these solutions over time?
• How do leaders decide what to scale and what to stop?

At the same time, colleges and universities are being asked to demonstrate increasingly complex outcomes. They need to understand and communicate performance across enrollment, retention, completion, research, financial sustainability, workforce alignment, and community impact.

AI may help institutions address some of these challenges. But access to AI will not create those outcomes on its own. The real opportunity is not to find as many places as possible to use AI. It is to identify where AI adds meaningful value and where the institution may need to strengthen its data, processes, or analytics first.

Not Every Problem Is an AI Problem

There is no shortage of ideas for using AI in higher education. It could help employees draft and summarize content. It could make institutional policies easier to find. It might support research, personalize student communications, identify enrollment patterns, or help advisors recognize students who may need additional support.

Many of those ideas are worth exploring. But a possible use is not automatically a good investment. Sometimes, what looks like an AI opportunity is actually a data problem.

For example, if an institution cannot produce a trusted view of student retention, putting an AI tool on top of that data will not solve the underlying issue. The institution may first need consistent definitions, better-connected systems, clear data ownership, and reliable reporting.

The same is true for enrollment, research performance, workforce outcomes, and financial planning. If leaders do not trust the information they already have, an AI-generated answer will not suddenly make it trustworthy.

In fact, AI can sometimes make the problem harder to see because it can present an incomplete answer in a very convincing way. Before moving forward with an AI use case, institutions should ask:

• What problem are we trying to solve?
• Is AI actually necessary to solve it?
• Could better data, reporting, automation, or process improvement address the need?
• What additional value would AI create?
• What new risks or responsibilities would it introduce?
• How will we know whether the AI-enabled approach works better than what we are doing today?

This is not about slowing down innovation. It is about making sure the institution is solving the right problem with the right approach.

Different Uses of AI Need Different Guardrails

One of the challenges with talking about “AI” is that the term covers a wide range of uses. Using an approved AI assistant to draft an email is very different from using a model to identify students who may be at risk of stopping out. Summarizing an institutional policy is not the same as recommending an academic intervention. Supporting a researcher’s work is different from influencing a decision about admissions, financial aid, or employment. These uses should not go through the same process or be governed in the same way.

A low-risk productivity tool may need basic usage guidelines, privacy protections, and employee training. A tool that uses student data or influences an important decision needs much greater scrutiny. Institutions may find it helpful to group AI uses into categories such as:

• General productivity and content support
• Institutional search and knowledge management
• Data analysis and decision support
• Predictive models and recommendations
• Student-facing or faculty-facing services
• Research applications
• Uses that influence consequential decisions

As the potential impact increases, so should the expectations around data quality, testing, documentation, accessibility, transparency, and human review.

For higher risk uses, saying that a person remains “in the loop” is not enough. Institutions need to be clear about who reviews the output, what information that person receives, how someone can challenge a recommendation, how errors are reported, and who is ultimately accountable for the decision.

A Good Idea Still Requires Capacity

Higher education leaders will likely identify more worthwhile AI opportunities than their institutions can realistically pursue. That is especially true when technology and data teams are already managing cybersecurity, system modernization, integrations, reporting requests, technical debt, regulatory requirements, and day-to-day support.

An AI pilot can seem small at first. But turning a successful pilot into a dependable institutional capability takes ongoing work. Someone has to:

• Prepare and connect the data
• Review security and privacy
• Manage identity and access
• Evaluate vendors and contracts
• Redesign the affected process
• Train and support users
• Monitor results and accuracy
• Respond when something goes wrong
• Maintain the solution as systems and models change

That means the question cannot simply be, “Could this create value?” It also has to be, “Does the potential value justify the time, cost, and capacity required to implement and sustain it?” A practical way to evaluate proposed use cases is to consider:

• Alignment: Does this support an important institutional priority?
• Value: What should improve if it works?
• Alternatives: Could we solve the problem through better data, analytics, automation, or process improvement?
• Feasibility: Do we have the necessary data, technology, skills, and ownership?
• Risk: What happens if the output is incomplete, biased, exposed, or wrong?
• Adoption: Will the people affected understand, trust, and use it appropriately?
• Sustainability: Who will own, fund, and maintain it after the pilot?
• Measurement: What will tell us whether to expand, change, or stop it?

This gives leaders a way to prioritize opportunities instead of allowing a collection of disconnected pilots to grow across the institution.

Governance Should Help People Move Forward Responsibly

AI governance is often treated as a policy exercise. Institutions establish rules about approved tools, sensitive information, academic integrity, and acceptable use. Those guardrails matter, but good governance should do more than tell people what they cannot do. It should give them a clear way to move a good idea forward.

Faculty, researchers, and staff need to know how to propose and test a use case. Technology and data leaders need clear standards for evaluating security, integration, and data requirements. Academic, legal, accessibility, procurement, research, and risk leaders need to understand when they should be involved. Executive leadership also needs visibility into what is being tested, how much it costs, and whether it is producing value.

Higher education adds another layer of complexity because institutions are rarely fully centralized. Faculty autonomy, shared governance, research independence, and individual schools or departments all shape how decisions are made. A workable governance model should balance consistency with appropriate flexibility. That may include:

• Institution-wide minimum standards
• Clear accountability for enterprise risk
• Faculty, academic, research, accessibility, and administrative input
• Department-level flexibility within defined guardrails
• Additional review for higher-risk uses
• A clear path from an idea to a pilot to broader adoption
• A process for stopping an initiative that is not working

The goal is not to eliminate every possible risk. It is to make thoughtful decisions, apply the right level of review, and make responsible experimentation possible.

AI Is Already Entering the Institution

Institutions are not starting with a clean slate. Faculty, staff, and students may already be using public AI tools. Departments may be buying specialized products. Existing software vendors are adding AI features to platforms the institution already owns. This creates a practical challenge. A policy alone will not stop people from using tools that make their work easier, especially if approved alternatives are unclear or difficult to access.

Institutions need to understand where AI is already being used, educate people about what information is safe to share, and provide practical guidance. They also need to look carefully at the AI capabilities entering through existing vendors. Important questions include:

• Will institutional data be used to train the vendor’s models?
• How long is the data retained?
• Which other companies or subprocessors can access it?
• How are outputs generated and monitored?
• Does the contract provide adequate protection?
• Are we purchasing functionality we already own somewhere else?
• How might usage-based costs change over time?
• Can we move our data and configurations if we change vendors?
• What happens if the vendor changes its model or terms?

This is not only about evaluating new AI tools. It is also about understanding how much AI is already inside the institution’s technology environment.

Data Readiness Is a Major Part of AI Readiness

AI readiness involves more than data. It also depends on security, technology architecture, integration, procurement, legal review, capacity, process ownership, and change management. But data readiness is one of the most important, and often underestimated, pieces.

AI applications connected to enrollment, advising, student success, research administration, workforce reporting, financial planning, or institutional decision-making all depend on reliable data. Before moving forward, institutions need to understand:

• Where the relevant data lives
• Who is responsible for it
• Whether important terms are defined consistently
• Whether the data is complete, accurate, timely, and accessible
• How information moves between systems
• Which privacy and security requirements apply
• Whether the data can appropriately be used for the proposed purpose
• How the resulting output can be reviewed and validated

This does not mean an institution has to fix every data problem before it can use AI. It does mean that readiness should be evaluated one use case at a time. An institution may be ready to create an internal assistant that searches a controlled set of approved policies. That same institution may not be ready to build a predictive model using student information from several disconnected systems.

“Are we AI ready?” may simply be too broad of a question. A better question is, “Are we ready to support this use case responsibly?”

Research Needs Its Own Consideration

Research creates opportunities and risks that do not always fit neatly within administrative AI governance. AI may help researchers navigate literature, analyze data, write code, recognize patterns, or move more quickly through parts of the research process. 

At the same time, researchers may be working with unpublished findings, intellectual property, controlled information, grant requirements, human-subject data, export restrictions, or proprietary partner information.

Institutions need to protect research independence while also providing clear guidance about what information can be shared with external tools and how AI-supported work should be documented. Research leaders should have a meaningful role in institutional AI governance, but research use cases may also require their own standards and review processes.

Value Does Not Always Mean Financial Return

Higher education outcomes are not always easy to measure. Improvements in student success, faculty experience, research capacity, accessibility, or community impact may take time to appear. They may also be influenced by several factors, which makes it difficult to tie an outcome to one tool or intervention.

That does not mean every AI initiative needs to produce a simple financial return. But the institution should define what it expects to gain before a pilot begins. Value could mean:

• Giving faculty or staff time back
• Reducing administrative burden
• Lowering risk
• Improving service
• Expanding access or accessibility
• Helping leaders make faster or more informed decisions
• Improving the student or faculty experience
• Strengthening educational or research outcomes
• Better demonstrating workforce or community impact

Some pilots may be intended primarily to help the institution learn. That is a valid outcome too as long as leaders are clear about what they are trying to learn and what decision that learning will inform. There may not always be a perfect measure. But there should be enough evidence to decide whether the institution should continue investing.

Equity and Accessibility Cannot Be an Afterthought

AI has the potential to expand access and provide new kinds of support. It also has the potential to deepen existing differences. Not every student, employee, department, or institution has the same access to tools, training, data, or technical support. If adoption is left entirely to individual departments, the areas with the greatest resources may move ahead while others fall further behind. Institutions should consider:

• Whether approved tools are available equitably
• Whether the tools and their outputs are accessible
• Whether the underlying data represents different populations fairly
• Whether automated recommendations could create uneven outcomes
• Whether users understand the tool’s limitations
• How someone can question an AI-influenced result
• Whether adoption could widen differences across departments or student groups

These are not separate issues to address after implementation. They are part of deciding whether a use case is responsible and valuable in the first place.

The Next Phase Will Require Choices

The institutions that get the most value from AI may not be the ones with the most tools or the largest number of pilots. They may be the ones that become disciplined about deciding:

• Which problems are most important
• Which problems actually require AI
• Which data is ready to support the work
• Which risks are reasonable
• Which uses need additional scrutiny
• Which initiatives justify limited capacity
• Which results would demonstrate value
• Which experiments should not continue

Providing access was an important first step. It created safer options, helped institutions learn, and gave faculty, staff, and students room to understand what AI might make possible. The next phase is harder because it requires institutions to make choices.

The goal should not be to use AI everywhere. It should be to use it where it can create meaningful value, and to recognize when better data, clearer processes, stronger analytics, or a different solution should come first.