Neither Reckless nor Frozen: Governing AI in Higher Education

When artificial intelligence (AI) reaches the board agenda, it tends to have one of two responses.

For some boards, the response is caution. AI shows up wearing a warning label, and the questions are about containment: What could go wrong? What is our exposure? What happens to the data, the reputation, the trust we have spent decades building?

For other boards, the response is excitement. People in the organization are already using large language models (LLMs) every day to draft notes, summarize documents, and answer questions. The tools are in the building, the conversation is lively, and it feels like the institution is thoroughly engaged with AI.

These look like opposite reactions, but they share a blind spot. The cautious board has not started. The excited board is deeply engaged with AI as a topic (the tools, the headlines, the possibilities) without having asked whether any of it has solved a real problem, built momentum, or driven growth. Enthusiasm and daily use feel like progress, but they are not the same as results. Neither posture has grappled with what transformational AI requires because both have mistaken a question about tools for a question about impact.

The question of how much AI risk are we willing to tolerate? or have our people started using these tools? are common, but perhaps a more useful one is: Are we building the conditions that let our institution stay relevant without becoming reckless?

A model already stretched thin

Using AI and transforming with AI are not the same thing. When a staff member asks a chatbot to polish an email, that is genuinely useful, but the work still happens exactly the way it always has. A familiar task has simply gotten faster, most commonly for individuals rather than embedded in an organizational workflow. The tool has been added on top of the same way of doing things.

A transformational approach to AI will see it embedded into how the work gets done, reshaping processes, informing decisions, and expanding what the institution is capable of. There is a reason this matters more in higher education than the general AI conversation would suggest; the business model underneath advancement was already under strain long before generative AI arrived.

For decades, fundraising scaled the only way it could: by volume—more appeals, more mail, more solicitations, more constituents contacted more often—and when a campaign was on the horizon, more people hired to complete more tasks. That model is now breaking on two fronts at once. The economics no longer work; institutions cannot keep buying growth with ever-larger volume and ever-larger teams, and the returns on that approach have been falling for years.

The problem with the old model is not just that it is expensive but that people no longer want it. This shift is not something donors merely tolerate; it is something they prefer. People do not want a deluge of undifferentiated information; they want to be engaged.

Meeting hundreds of thousands of people as individuals, understanding what each one cares about, and engaging them accordingly is not something a larger team can accomplish by working harder. This level of engagement requires transformation in how the work is done. The model that results is broader than any single capability, but two things define it: personalization at scale and a structure the institution can sustain over time. It is also something that constituents want. That is a rare alignment, and a very strong argument for change.

Why standing still is also a risk

The risks of real transformation deserve serious consideration, but less often examined is the risk of inaction. Asking what could go wrong if the institution changes is only half the question. The other half, what the institution gives up if it does not change, carries equal weight.

The second risk is harder to see because it builds gradually. Experimentation is essential, but it must lead somewhere. Each quarter in which it stays disconnected from the institution’s core work is a quarter of relationships left undeveloped, opportunities left unrealized, and expectations moving further beyond what the current model can deliver. From the boardroom, prudence can be difficult to tell apart from inertia, and activity from achievement. The difference is real, but it seldom shows up in the measures boards typically review.

Governance is what makes transformation possible

Innovation and caution are often framed as competing priorities, but in practice each depends on the other. Treating them as opposites tends to produce one of two outcomes: hesitation or momentum without structure. Strong governance lets an institution take meaningful risks with confidence. With the right guardrails in place, a board can commit fully to substantive change instead of approving it tentatively or deferring it.

Seen this way, the board’s role is to create the conditions for transformation to proceed responsibly. That means oversight close enough to manage risk, and commitment strong enough to carry the work past early experimentation to lasting results.

These conditions are easiest to establish where the stakes are manageable and the potential return is clear, which makes the advancement and foundation office a natural place to begin. The limits of the current model are especially visible there. The office is also where an important distinction comes into focus.

Not all AI is the same 

Much of the current conversation centers on LLMs and other generative tools, which produce text, images, and summaries. Predictive machine learning is a different discipline. It analyzes an institution’s own records to identify patterns and anticipate behavior, such as who is likely to give, when, and in response to what.

Institutions should hesitate to put sensitive constituent information into general-purpose generative tools. Private machine learning environments are built for this purpose, and the data stay secure. Used this way, machine learning has a well-established record of helping teams understand donor behavior, anticipate giving, and engage constituents at a scale no staff could manage alone.

The asset that cannot be created later

Waiting has one further cost. Most institutions hold substantial constituent data, including giving histories, relationship records, and years of transactional detail. Much of the data’s value goes unused because few institutions can process these data at the depth they allow. With the right tools, institutions can use existing records to derive a meaningful picture of past behavior, such as patterns in the timing, frequency, and trajectory of engagement. Some signals, though, exist only if they are captured as they happen: what constituents respond to, when their interest rises or fades, and what prompts them to act. These cannot be recovered later. An institution that starts now makes better use of the data it already holds and begins building a record that deepens every year. The one that waits leaves both untapped.

Rachel Crosbie is vice president of operations and strategy at Fundmetric.