Last year Teege Mettille and I wrote a book, The Signal Solution: How Smart Colleges Stop Chasing Applications and Start Converting Students. Its core thesis can be simplified into a single question: if we all agree the enrollment environment has permanently changed, why would we expect playbooks written for a different era to still work the same way? And what should enrollment leaders do about it now?
No one yet has disagreed with me on the first part of that premise, but the second part generally creates some friction. And one of the playbooks that’s changed the least is how we get from data to a decision: on a strategy, an investment, an individual student.
We talk to a lot of enrollment teams every day here. Almost none of them tell us they need more analytics.
What they tell us, in one way or another, is that they’re out of room to think. They’ve got dashboards, scores, weekly funnel reports, the whole thing.
But our analytics, even the most advanced, still only answer a portion of the question. For the rest, we rely on the interpretation, understanding, and pattern recognition of the team. And that’s where we hit the cognitive ceiling. When we studied how admissions counselors actually spend their time, we found they were spending 30 to 50% of it just wrangling data and lists, trying to figure out who to focus on, why, and what to do next.
Think about what we’ve unlocked when we can look at 300 to 400 individual behavioral data elements, not serially, but concurrently, and turn those into individual meaning for each student. The reality is that none of us, present company included, could possibly process that level of data.
We’ve tried to simplify it into predictive scores. But in doing so, we eliminated one problem (trying to assess what all of the data means together) and didn’t solve another: what does this truly mean for each student?
That’s the gap I want to talk about. It’s the difference between enrollment analytics and what I’d call AI thinking support for teams. They sound similar. They’re not, and I think the difference matters more right now than it ever has.
Where the work stops
Analytics stops at a fact. Thinking support stops at a point of view.
Analytics tells you what happened, what’s happening, maybe what’s likely. Conversion by stage. Yield by segment. A propensity score. Deposit pace against last year. All of it matters. But it hands the hardest part right back to the person. A counselor looks at a 0.62 and still has to work out why this student is wavering, what it means, and what to actually say on Tuesday’s call.
Let’s put a name on that work, because we all do it and we rarely call it anything. I call it enrollment judgment. It’s what happens when we go into committee. It’s what our most experienced people are doing when they sit down with a newer counselor. It’s what happens when we sit in our offices with the dashboards, the reports, and the metrics, trying to work out what it all means. It’s some of the most important work in enrollment, and it’s always been done by hand.
Thinking support does the heavy lifting behind that judgment. It forms a hypothesis about why a student is behaving the way they are, weighs it against other explanations, connects it to what the institution can actually offer, and suggests a move.
So analytics says: this student’s engagement dropped 40% after the financial aid letter.
Thinking support says: this looks like a value question, not a cost question. Her pattern looks like students who enrolled once someone connected the program to outcomes they cared about. Lead with that, not another scholarship.
You can disagree with that second one. You’re starting from a reasoned position instead of a number.
Analytics scales data. Thinking support scales judgment. One is the instrument panel. The other is a sharp colleague reading the panel with you. And it never really finishes, because the student keeps changing and so does the read on them. A report is done when it’s delivered. Thinking isn’t.
Here’s what that looks like in practice. A counselor with 30 years in enrollment read a three-paragraph AI assessment of a student in her funnel. The thesis it proposed wasn’t the path she’d been planning to take. But as she read it, she realized it had picked up on a situation she hadn’t accounted for. It changed how she saw the student, and it changed her strategy.
She followed the approach. She built a much deeper, more substantial connection than she would have otherwise, matched what the student actually needed with the unique value of the institution, and secured the student.
How do I know the story? Because right after that, she walked into her Vice President’s office, told her what had just happened, and said something to the effect of, “This is going to change everything.”
That’s not the AI making the call. That’s a 30-year veteran’s judgment, made sharper.
The advantage has moved below the language model
A couple of years ago, if you wanted a written read on a student, someone on your team had to sit down and do it. Now you can drop a CRM export into ChatGPT and get three confident paragraphs back in about ten seconds. It reads well. It sounds like somebody who knows the student.
I get the appeal. It’s actually where we started as well. But it doesn’t work the way that we need for high-stakes enrollment. Think about what the model is actually looking at. Opened emails. A campus visit. A FAFSA flag. At best, the model is looking at too limited data and has to infer too much. At worst, it’s making assumptions based on hundreds of other institutions running old playbooks. The same milestones that told the school in our book’s opening story they were on track, right up until they missed by 250 students. The LLM doesn’t fix that. It just writes it up nicer.
That’s precision theater with better grammar.
When Teege and I wrote The Signal Solution, the problem we kept running into wasn’t a lack of data. It was a lack of dimensionality. We were counting milestones instead of seeing behavior. That’s still the problem, and a language model doesn’t solve it on its own. We fixed this by combining ensemble machine learning and massive feature engineering, hundreds of behavioral signals built around timing, pace, and change, with governed and gated LLM analysis, built specifically for the work of enrollment and the decision paths of students. The models tell the student who’s speeding up from the one who’s quietly drifting. The LLM works out what that means for each one. It’s slow, unglamorous work, and it’s the only reason the reasoning on top is worth listening to.
The writing is easy now. The seeing isn’t.
Why it matters more now
The economics of enrollment judgment have shifted. We’re asking smaller, less tenured teams to interpret more signal per student than ever before, and the traditional answer, another report or another dashboard, adds to the load instead of carrying it. Automating the task, rather than training the thinking, gives a false sense of scalability. You have more contacts, but not more understanding of the student on the other end of them.
Until recently, reading each student individually at scale just couldn’t be done. It takes serious behavioral analytics and a language model working together, and that pairing is only now possible. And as everyone adds agents and automation, sending the email faster stops being a differentiator. Knowing which email, to whom, and why is. And task-based outreach doesn’t prepare a counselor for the conversation that comes next. I don’t believe auto-tasking is the path to connecting effectively with today’s and tomorrow’s student. When everyone has agents, it’s the thinking that differentiates.
What this changes, and what it doesn’t
I’m not saying analytics are going away. But trying to run enrollment on our analytics, reports, and trends the way we always have is like driving across the country with the parking brake on. You’ll get there eventually, and it feels familiar. But you’re burning a lot more fuel than you need to, and you’re leaving serious horsepower on the table: better decisions, smoother operations, a stronger team. You can now mitigate the enrollment leakage that happens when team members turn over. A 26-year-old can operate with the collective wisdom of the entire team. A 30-year vet gets a second perspective that reinforces their strategy or proposes another option.
Even at the leadership level, with trends, market shifts, and high-level performance, we’re still relying on ourselves or our strategy teams to work out what it means, what the implications are, and what our options are. The compression of that time, from data to implication to creating and assessing strategic options, is where the massive gains are being made. And it’s happening at every level, from the VP setting strategy all the way down to the individual student.
I’m also not saying good directors and analysts don’t already think this way. The best ones absolutely do. The problem is that there aren’t enough of them, and they can’t do it for every student, every week.
And I’ll say the uncomfortable part too. AI reasoning can be confidently wrong. If you’ve been burned by a tool that sounded smart and wasn’t, your skepticism is earned. The answer isn’t to promise accuracy. It’s to make the reasoning visible, make it easy to push back on, and keep people in the decision.
The question worth asking
So if you lead an enrollment team, here’s the question I’d sit with. Not “do we have the analytics?” or “do we have next actions?” You almost certainly do. And you almost certainly have the data. Across the hundreds of models we’ve built, we’ve only had to tell a school no twice because their data wasn’t enough to build high-confidence behavioral models. Twice.
The question is: are we unlocking all that is possible from our data, our systems, our investments, and our teams? For the institution, for the team, and for the betterment of each individual student, who gets a deeper, more thoughtful approach to every interaction.
And this isn’t theoretical. It’s capability that’s here now, already in use. We can turn it on in days, and it immediately changes how a team understands the work of enrollment. It goes straight at the cognitive ceiling.
That’s what we’ve been building VantagePoint to do since day one. But whatever tools you use, I think the next few years belong to the teams that figure out how to scale judgment, not just data.