Who Directs Enrollment AI And Analytics When The Forecast Sets The Budget?

The Forecast Missed By 140 Students. Who Answers For That Now?

It is late October, deposits are in, and the class is 140 students short of the model's number. The budget was built on that number. Somebody has to stand in a cabinet meeting and say which assumption moved, whether the recruitment outreach that ran on a vendor's propensity scores reached the students it should have, and whether the aid packaging rules made the gap worse. That sentence is the job description.

What has changed is that the answer now has more surface. An enrollment office in 2026 typically runs a predictive model for inquiry and applicant propensity, an aid optimization model, a CRM sending sequenced outreach that increasingly drafts itself, and some form of assisted application reading or summarization sitting next to human readers. Those arrived from four different vendors on four different timelines, and in most offices no single person can describe how they interact.

So the role consolidates. Johns Hopkins University posted a "Director, Enrollment Analytics, Planning and AI" on the Chronicle of Higher Education job board, an enrollment-office leadership role naming AI in the title beside analytics and planning ¹. That title shape is the thing to notice: the institution did not create an AI office and hand it the models. It widened an existing enrollment leadership seat.

This category is still forming, and honest hiring says so. There is no settled title, no standard reporting line, and no consensus on where model governance ends and institutional research begins. Some institutions are visibly hiring it today; most are quietly stretching a director of enrollment research to cover it. Write the posting knowing the market has not agreed on a name yet, and search on the work rather than the words.

What Separates A Real Enrollment AI Director From A Dashboard Manager?

The tell is what a candidate does with a model that was wrong. Ask about a year the forecast missed. A real one narrates the decomposition: which segment moved, whether it was the model or the market, what they changed in the following cycle, and what they told the president while it was still uncertain. A dashboard manager narrates the dashboard, and the miss stays a weather event that happened to them.

Four traits separate the real version. First, accountability history: the person has owned a published number that money was spent against, not a report about someone else's number. Second, the ability to explain error to a non-technical audience without either mystifying it or flattening it, because deans and trustees will ask, and the answer sets how much trust the whole analytics function gets for the next three years. Third, a working grasp of student-record obligations and of what an admissions decision assisted by a model has to be able to show later, which is a question for general counsel at your institution rather than one a director resolves alone. Fourth, and least common, a stated list of decisions that stay human. An applicant denial and an aid appeal belong on that list.

The performed version has a signature too. Vendor names doing the work that a method description should do. Confidence about accuracy with no discussion of who the model is least accurate for, which in enrollment work is usually the small populations the institution most wants to grow. Enthusiasm about generative reading support with no account of what the human reader is still reading for. And a plan that begins with buying a platform rather than with auditing what is already running.

Ask one more thing, and listen carefully to the answer. What did the candidate stop doing after a model started doing it, and what went wrong the first time? The people who have really run this work have a scar. The people who have supervised it from a distance have a roadmap.

Which Backgrounds Produce This Person, And Where Do You Find Them?

The reliable pools are three. Directors of enrollment research or admissions operations at institutions a size or two larger than yours, who have run a yield model through several cycles. Financial aid leaders who have owned discount strategy, which is the oldest optimization problem in the building and produces people who already think in terms of a model's distributional effects. And institutional research directors who moved toward decision support rather than compliance reporting.

The unexpected backgrounds are worth real attention, because the obvious pools are thin and are being drawn on by every institution at once. Enrollment consultants and analysts from the firms that build these propensity and aid-optimization models have seen fifty offices instead of one, and some of them are ready to sit on the institution side. Actuarial and revenue-management people from insurance and hospitality translate more directly than the resumes suggest, since the work is forecasting a population under price sensitivity with a hard deadline, which is much the same problem an actuarial AI transformation lead handles in a different industry. Registrars know the student record and its constraints better than anyone else on campus. And a strong internal candidate is often a senior associate director who quietly rebuilt the office's reporting and is already the person everyone asks.

That last group is the one to check first, and there is a specific way to look. The person who got good at this used the tools on their own work before governing anyone else's: rebuilt a segmentation query with an assistant and then checked it against the warehouse, drafted a communication flow and then read what it would say to a first-generation applicant, tested a summarization tool against files they had already read by hand to see where it dropped the thing that mattered. That practice history is more predictive than any credential, and it is visible in about ten minutes of conversation.

For the search itself, the higher-education job boards where enrollment leadership actually posts are the right first stop, including the Chronicle of Higher Education board where the Hopkins role appeared ¹. Beyond that, the professional associations are where the candidates are already known to each other: AACRAO for the registrar and admissions side, NASFAA for aid, and the Association for Institutional Research for the analytics side. Conference presenters are a better list than applicants, because presenting on a yield model means somebody's institution let them, which is itself a reference. Some institutions instead grow the capability by hiring a corporate AI coach into the division first and promoting from the people who take to it.

Price This Role Against Your Enrollment Leadership Band, Not An AI Band

No published salary series covers this title, because the title is roughly a year old and no two institutions spell it the same way. Any point estimate you see quoted for an enrollment AI director is an average of a handful of postings for jobs that are not the same job. Refuse it, including from a candidate who brings one.

Band it against what your own institution already pays. The honest neighbors are your director or senior director of enrollment management band and, where the role carries the forecast and reports to the vice president, the associate vice president band. The higher-education administrator salary surveys your HR office already subscribes to carry both, segmented by institution type and size, and that segmentation matters more here than any national figure. Offers land toward the top of the range when the role owns the published forecast rather than advising on it, and toward the middle when it is a technical seat under a planning officer.

Expect to pay above the historical band for the same seat, and know why. Across roughly one billion job advertisements, PwC's 2026 AI Jobs Barometer put the average wage premium for AI skills at 62 percent ². That figure spans every industry rather than higher education specifically, and university pay bands move slowly, so treat it as a description of the pressure on your offer rather than as a target. The practical version: candidates who can do this work have private-sector options that your published range does not have to beat, but does have to acknowledge.

Budget two lines that get left out of the requisition. One is the analyst or two the director will need in year one, because a director alone with four vendor systems becomes a report writer within a semester. The other is the vendor audit itself, which is unglamorous, takes months, and is the work that makes every later number defensible.

Close The Enrollment AI Director, And Decide Where They Sit

Strong candidates close on authority rather than on title. Four things move them: a reporting line to the person who owns the class, a seat in the meeting where the forecast is set rather than a summary sent to it, standing to say no to a vendor, and an institution willing to write down which decisions a model does not make. That last one is free, and most postings skip it.

The offer killers are just as consistent. A brief that asks for AI adoption as the goal, which reads as a mandate to buy things. A forecast owned elsewhere with the models governed here, which is accountability without control and experienced people can smell it. No analyst headcount. And a president who wants the analytics to confirm a strategic enrollment plan already approved, which is an unwinnable assignment that candidates have watched a predecessor lose. Name the first year plainly instead: audit what is running, publish one forecast with its assumptions visible, and set the governance rule for assisted reading. A candidate worth hiring will argue with that order.

On location, the pattern in higher education is hybrid with a real on-campus obligation, and the obligation is not ceremonial. Reading season, cabinet meetings, admitted-student events and the awkward conversations with deans about a segment that missed all happen in a building. Institutions that have hired this role remotely tend to have done it for a proven internal person who already has the relationships. For an external hire, on-campus for the first cycle is worth insisting on, and generous flexibility afterward costs the institution very little.

One closing test, at the end of the finalist visit. Ask what the candidate would refuse to automate in your office specifically, having spent a day in it. A person who names something concrete, and can say who gets hurt if it is automated anyway, is the person you want holding this. A person who says nothing comes to mind has told you how the next three years go.

See the benchmarks >