Data Readiness Is the AI Work: A Case for Starting with Advancement

University leaders weighing where to invest in transformational artificial intelligence (AI) are facing a question of sequencing as much as strategy. The pressure to “do something with AI” is real, and the temptation is to invest where the headlines are loudest. But sequencing matters. The first place an institution applies transformational AI sets the tone for everything that follows: the data infrastructure it builds, the governance norms it establishes, the trust it earns or loses with its community, and the institutional confidence it develops to take on harder problems later.

Institutional AI adoption in higher education jumped from 49 percent in 2024 to 66 percent in 2025, a 17-point move in a single year, with 43 percent of institutions placing AI directly in their strategic plans and nearly two-thirds of executive leaders reporting allocated funding (Ellucian, 3rd Annual Higher Education AI Survey). The pivot from posture to placement is already happening.

A Metric Vanishes While the Pipeline Collapses

Among 627 institutions that reported alumni donor counts in both 2019 and 2023, 83 percent saw declines, with the average institution losing 20 percent of its alumni donor pipeline over four years. Alumni contributions fell 13 percent on an inflation-adjusted basis in FY 2023 alone. Princeton University, an institution that has historically defined what alumni participation looks like, posted an undergraduate participation rate of 43.9 percent in 2024–2025, the lowest figure in nearly 80 years (CASE Insights on Voluntary Support of Education 2025).

Then, in 2025, U.S. News and World Report removed alumni participation rate from its rankings methodology entirely. That is quite a big development. The participation rate was the metric that propped up annual giving programs for two decades. It was arguably the external reason boards cared and the internal reason institutions invested in the function.

The advancement model is going to evolve whether the institution is ready or not. There is an important question of whether it evolves deliberately, with capability the institution has built, or reactively.

The AI Bottleneck Is Data Readiness

The single most important thing for boards to understand about transformational AI is that the constraint has moved. A model is, increasingly, a commodity. The constraint now, almost entirely, is whether the underlying data can support what the model is being asked to do. Before an institution can predict, segment, score, personalize, or surface useful insights, it has to assemble the data. Across most advancement operations, that assembly work is the project. It takes longer than the model. It is harder than the model. And it is the part nobody wants to fund.

That distinction matters because it changes what “transformational AI” means. It does not mean licensing a smart-looking tool, buying a chatbot, or rolling out a single application. Transformational AI is what becomes possible when the data foundation underneath it is in place. Without that foundation, every AI initiative is a one-off integration project—organizing information, reconciling definitions, rebuilding pipelines, and starting from zero each time. The cost is linear. Every new use case is its own engagement.

With a data foundation in place, the math inverts. Every new model, dashboard, or AI tool leverages what has already been built, and the compounding starts. The same data work that was holding back the first use case is now accelerating the tenth, and it is what separates a feature from a strategic capability. AI itself is not the work.

The question of how an institution gets there—whether it builds the foundation in-house, licenses a platform that helps build it, or has a partner deliver it as a managed service—is downstream of the question of whether it gets there at all. The right delivery model depends on what the institution already has, what it can staff, and what its operating reality looks like. What does not change is the underlying point. The foundation must exist, and the institution has to be able to keep building on what was built.

Why Advancement First

If data readiness is the work, advancement is unusually well suited to be where it happens.

For decades, universities have been collecting structured data on donors and prospects—gift histories, engagement records, event attendance, wealth indicators, and relationship maps. The data are imperfect, often siloed, and unevenly maintained, but they are real, longitudinal, and tied to outcomes that are measurable. Few other domains in the university offer such clean feedback loops, which allow a model—and the institution’s understanding of it—to improve over time rather than ossify around initial assumptions.

There is also a second-order benefit that often goes unnamed. Every successful AI initiative in advancement builds institutional capability. The team, the governance frameworks, the practices built around data definitions and feedback loops—all of these are reusable. An institution that learns how to deploy AI responsibly in advancement is materially more prepared to deploy it responsibly in admissions, in student success, in research administration, and beyond.

Remember what happened with email. Higher education was conspicuously late to recognize that alumni email addresses would become a critical institutional asset. Many institutions did not begin systematically collecting them until well after the rest of the economy had moved on, and the gap between the institutions that did and the institutions that did not is still being measured today, in engagement, in giving, in the simple ability to reach a graduate. AI capability, and more importantly the data foundation underneath it, can be compared to the email address of this decade.

A Philanthropic Dollar Has Flexibility; a Business Dollar Does Not

If advancement is the right place to build the foundation, the next question is how the institution pays for it. And here advancement has a structural advantage.

University budgets are not monolithic. Tuition revenue is tied to instructional capacity. State appropriations come with restrictions. Federal research dollars are scoped to specific projects. Auxiliary revenue is encumbered by the operations it supports. Each of these is a “business dollar”—a dollar with a job to do, a line item to occupy, and a fiscal-year cycle to live within. None of these dollars is well suited to underwriting transformational technology investment, particularly one whose payoff curve is uncertain and whose first attempts may not succeed.

A philanthropic dollar is a different instrument entirely. A gift made to support innovation, transformation, or strategic capability-building can move faster, tolerate more risk, and absorb more learning than a budgeted operational dollar ever could. Philanthropy can underwrite a pilot. Philanthropy can fund the hire that the operating budget cannot yet justify. Philanthropy can pay for the failure that produces the insight that makes the next attempt work.

What makes advancement uniquely positioned is that this flexible capital is already in its orbit. Advancement cultivates philanthropic relationships, which means advancement can credibly raise philanthropic support for its own transformation. There is a natural symmetry to it: a donor or philanthropic partner funds the AI capability that helps the institution build deeper, more thoughtful, more personalized relationships with donors like themselves. The investment funds the very work that makes the case for future investment. Few other units on campus can tell that story to a funder and have it land the same way.

The scale matters too. American higher education raised $61.5 billion in voluntary support in fiscal year 2024, up 3 percent on an inflation-adjusted basis over the prior year. Roughly 21 percent came from alumni, 15 percent from non-alumni individuals, and about 33 percent from foundations (CASE Insights on Voluntary Support of Education 2024). That is the size of the operating layer being asked to evolve. A philanthropic dollar deployed against the data foundation that protects and grows that base is arguably among the highest-leverage uses of philanthropy on campus.

Asymmetry of Risk

Predictive scoring of donors raises legitimate questions about privacy, about consent, and about the optics of wealth screening. Sophisticated donors and most board members understand that they are already being scored; that has been true for a long time and across many sectors well beyond higher education. But that fact is not a substitute for governance. Institutions that take this on responsibly will need clear consent practices and transparent data policies. Data privacy remains the top reported barrier to AI adoption in higher education for a reason, and any institution starting in advancement has to contend with it directly. These issues need to be acknowledged, built into the operating model from the beginning, and revisited as the work matures.

With the identification of these concerns, the comparison still holds. The errors a model makes in advancement are reversible in ways the errors a model makes in student success are not. A donor misclassified as a major gift prospect can be reclassified, and the relationship continues. A student misclassified as “at risk” of not completing their degree can have that classification follow them, signal something to advisers and faculty that subtly shapes how they are treated, and, in the worst case, become a self-fulfilling prophecy. Predictive models trained on historical retention data inherit the patterns of historical inequity; a model that learns who has graduated in the past will, without careful intervention, learn to predict that students who look like the students who did not graduate in the past will not graduate in the future. That is not a technical bug. That is the model doing exactly what it was trained to do, and it is the kind of outcome that should give a board pause.

Advancement risks are real and need to be managed. Student-facing risks are real, harder to manage, and harder to recover from when they go wrong. That is the honest comparison, and it still favors starting where the recovery path is shorter and the people most directly affected are the ones the institution has already invited into a deliberate, two-way relationship.

The Fiduciary Case

Boards are often told they need to “embrace AI” without being given a framework for where embracing it makes sense and where it does not. The framework worth holding on to is straightforward. Start where the data are mature. Where the feedback loops are tight. Where the return is measurable. Where the cost of error falls on the institution rather than on the people the institution exists to serve. And where the funding model is flexible enough to support real experimentation. Advancement meets all five tests.

An institution that tries to start with the harder problems and stumbles publicly may find that its license to use AI anywhere on campus has been quietly revoked. And an institution that does not start at all may find that the base it was counting on has moved on without it.

For boards asking the question, “Where do we begin?,” the answer may be the place where the asymmetry of risk most strongly favors trying and where the institution’s work compounds into the capability it owns for everything that comes next.

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