Making AI Work in Higher Education

Start with why AI stalls

Higher education is spending more on AI than ever, and most of it is stalling. The recurring reason is not the model. It is a fragmented data foundation and the distance between what a campus knows about a student and what it actually does about it. We laid out the evidence for that in a companion piece, Why Higher Ed's AI Keeps Stalling.

This piece is about the other half of the problem: what it actually takes to make AI do real work on a campus, and how the pieces fit together into a single operating loop.

What the exceptions actually do

Across the research, organizations that get AI to work converge on a recognizable set of behaviors. The evidence is not perfectly higher-ed-specific. Scale AI and Reuters Insights, for example, surveyed nearly 500 senior enterprise AI decision-makers across industries. But the pieces we can check against higher education data, especially the centrality of the data foundation and executive sponsorship, line up closely enough to treat the broader pattern as a serious operating model.1

• They consolidate instead of accumulating. The strongest programs avoid a drawer full of single-purpose apps and combine internal capability with integrated external platforms.1
• They treat the data layer as the first decision, not the last. High-quality proprietary data was the single most-cited factor in Scale AI's research on successful enterprise deployments, and the higher-ed numbers show the same pressure in budget form.1
• They buy a partner, not a product. Successful organizations want deep co-development or managed outcomes, not a vendor who hands over a login and calls it transformation.1
• They frontload the unglamorous work. Executive sponsorship, workflow redesign, change management, and outcome measurement have to be decided before go-live, not after the pilot starts to drift.23

What it actually takes to put AI to work

It helps to stop treating AI for student success as a single thing you buy. Putting it to work is really several decisions that have to hold together, and they start below the model, with the ground it runs on.

Begin with one connected foundation. The reasoning can run on models any institution can license. The advantage comes from what the reasoning runs on. High-quality proprietary data was the single most-cited factor in Scale AI's research on successful enterprise deployments, and the higher-ed budget data points the same way. This is the layer Edvise builds first: a connected student record that brings admissions, advising, LMS, engagement, and financial context into one current view of each student, so an agent reasons over the whole person instead of a fragment.14

It is also why the buying decision matters as much as the build. The institutions that succeed look for a partner who will co-develop and own outcomes, not a vendor who hands over a login and calls it transformation.1

A connected record is necessary but not sufficient. A model that can reason is not the same as an agent that can act correctly inside your institution, and the gap between the two is knowledge. Every agent Edvise deploys into an office is given two things before it does anything: your definitions and your domain knowledge.

The first is a set of semantic definitions: what your campus means by retention, full-time, at-risk, or satisfactory academic progress, encoded so the agent uses your meaning rather than a generic one. Definitions are where most analytics quietly go wrong, and where institutional trust is won or lost.

The second is a working corpus of what actually moves the outcome. Every Edvise agent draws on Atlas, a living knowledge base of student-success practice: the interventions, playbooks, and evidence that have helped students persist across the sector. An agent without that corpus is articulate but uninformed. An agent with it behaves like a colleague who knows the field, recommending what has been shown to work instead of improvising.

What the orchestration layer actually does

With that foundation in place, the orchestration layer does a few distinct kinds of work on top of it, and because they all run on the same connected record, each one makes the others sharper.

It finishes routine work instead of just flagging it. Where most campus AI stops at answering a question or raising an alert, the Edvise agent carries the task forward: it reaches out to a student, gathers the records a case needs, prepares the outreach or the form, and moves a routine process along until the only thing left is the part that genuinely needs a person. This is the engine behind Agentic Retention and Agentic Enrollment.

It organizes the human side of the work. The people who hold the judgment, advisors and student-success staff, get a short, prioritized view of who actually needs them, with the routine already handled and the context already gathered. The goal is to give time back, not to hand someone another screen to watch.

It learns your campus. A model trained on your own students, rather than a national benchmark, surfaces who is quietly drifting toward not returning and, just as usefully, why: the particular mix of course trouble, fading engagement, missed advising, or a financial hold that is pulling a given student off track. That is what points everything else at the right students first.

None of these is worth much on its own. The value is the loop they form on the shared record. A student interaction creates a signal, the orchestration layer turns scattered signals into a decision, the action flows back out through the right person or channel, and the result becomes new signal that sharpens the next decision. Run them as three disconnected tools and you get three more logins and three more versions of the truth. Run them on one record and they compound. Remove the agent and the model goes blind and mute. Remove the model and the agent becomes a faster help desk. Campus Technology described the same gap from the analytics side: institutions can generate insight, but the work stalls between insight, ownership, and action. AI does not close that gap on its own. It has to be wired into how the institution runs.3

Built so you can actually turn it on

None of this is deployable unless it is governed, and on a campus that is not a footnote. Student records are regulated, and the people accountable for them, the registrar, the CIO, the general counsel, need to see the controls before an agent touches anything real.

Edvise is built so the institution keeps control of its own data, access is scoped to each role, every action an agent takes is logged, and the decisions that carry weight wait for an accountable person to approve them. Policies, escalation thresholds, and the outcomes that matter are defined once and applied consistently across every office, instead of being renegotiated tool by tool.

That governance also protects the part of student success that should never be automated. The relationship between an advisor and a student is what actually moves persistence, and it stays human. The agent's job is to clear everything around that relationship, the scheduling, the paperwork, the follow-through, the first draft, so the human hours go where they are irreplaceable.

What this means for decision-makers

If you are a provost, CIO, enrollment leader, or vice president for student success weighing where to place an AI bet, the research points to a short and unsentimental checklist.

• Anchor a named executive sponsor before funding anything. Pilots run from a departmental line, without leadership attention beyond procurement, are the ones most likely to stall.1
• Treat the student-data layer as the first architectural decision. Resolve ownership, access, permissions, and integrations before choosing a model or locking the use case.61
• Resist the point-tool reflex. Another single-purpose app is more sprawl, more silos, and another source of truth to reconcile.1
• Redesign the advising or support workflow before go-live. The integration failures that kill pilots are usually designed in at this stage.23
• Measure persistence, retention, and conversion, not adoption. Logins and messages sent are activity metrics. They are not outcomes.
• Keep humans in the loop where judgment belongs. AI can classify, summarize, draft, nudge, and route. High-impact student decisions still need accountable human review.

The window is still open

None of these moves is about new technology. Each one takes executive sponsorship and the discipline to follow through, which is why so few institutions make them. The campuses that come through the next several years in the best shape will not be the ones chasing the newest model. They will be the ones that did the foundational work on data, integration, sponsorship, workflow, and measurement, then built AI into how the institution actually runs.

The fastest, most measurable returns sit on the surface, which makes student communication and support automation a practical place to prove value quickly. The durable value sits underneath, in the connected student record and the orchestration layer that can convert signal into action. Institutions that win will build both.

The window to do that work is open. Given the demographics bearing down on the sector, it will not stay open indefinitely.