Perspectives

There’s a word behavioral economists use that most enrollment leaders have never heard, but would immediately recognize: satisficing.

It means settling for “good enough” instead of finding the best option — not because you don’t care, but because you don’t have time to evaluate everything. Herbert Simon coined it in the 1950s to describe how humans make decisions under constraints. It’s not a flaw. It’s a coping mechanism.

And it is the single most accurate description of how enrollment teams operate today.

The real problem isn’t effort. It’s capacity.

A typical admissions counselor manages 800 to 1,200 students. Each one has a different academic profile, communication history, financial situation, family dynamic, and timeline. The counselor also has travel weeks, events, committee work, and the pressure of hitting goals that shift mid-cycle.

No one in this job is being lazy. But the math doesn’t work.

When you’re managing that kind of volume, you triage. You check the CRM, scan for red flags, and move on. You send the same message to 40 students because writing 40 different messages isn’t possible by Thursday. You go with your gut on who to call first because there’s no time to analyze the data.

That’s satisficing. And it’s happening in every enrollment office in the country.

The industry’s response has made it worse.

For two decades, the answer to enrollment complexity has been: more tools. More dashboards. More workflows. More predictive scores. More automations. More “next steps.”

Each one adds a layer of information the counselor is supposed to process. But processing isn’t the bottleneck — thinking is. A counselor who already can’t evaluate 1,000 students deeply doesn’t benefit from another dashboard showing them the same 1,000 students from a different angle.

More data without more thinking capacity is just more noise.

What would actually help.

The solution to satisficing isn’t more willpower or more information. It’s better thinking infrastructure — systems that do the deep analysis humans can’t do at scale, so humans can focus on the decisions and relationships only they can handle.

Imagine a counselor starting their day and instead of scanning a dashboard, they get a briefing: “Here are the three students in your caseload who need attention today, why each one matters, and what the best approach is for each, written in your voice.”

That’s not automation. That’s augmentation. The counselor still decides. Still writes the email. Still makes the call. But the thinking that used to take 45 minutes of CRM archaeology has already been done.

Now multiply that across every counselor, every day, for every student in the funnel.

This is the cognitive ceiling.

We talk a lot about the enrollment cliff. But the more immediate problem is the cognitive ceiling — the gap between what enrollment teams are asked to accomplish and what’s humanly possible to think through.

The cliff is demographic. The ceiling is operational. And the ceiling is what teams hit every single day.

Satisficing is the symptom. The ceiling is the cause. And no amount of dashboards, automations, or predictive scores will raise it — because those tools don’t think. They display. They sort. They trigger. But they don’t reason about a student’s situation, weigh the trade-offs, and recommend what to do next.

The question has shifted.

Two years ago, the question was “Will AI work in enrollment?” That question is settled. The research is clear, the early adopters are seeing results, and the tools exist.

The question now is: “How will we do it?”

And the answer matters enormously, because not all AI is the same. A chatbot that auto-responds to inquiries is AI. A system that reasons through every student in your funnel every day, weighs their signals, considers your institutional voice, and delivers personalized strategic guidance to each counselor — that’s a different thing entirely.

The first replaces a task. The second raises the ceiling.

What this means for your team.

If you lead an enrollment team, you already know the satisficing problem — you just may not have had a word for it. You see it when counselors default to batch communication. When territory strategies don’t get updated mid-cycle. When leadership questions take weeks to answer because the data is scattered.

The fix isn’t asking your team to work harder. It’s giving them thinking infrastructure that matches the complexity of what they’re being asked to do.

That’s what breaking through the cognitive ceiling actually looks like. Not more tools. Not more data. Better thinking — at scale, in real time, for every student, every day.

I spend a significant amount of time talking with enrollment leaders about AI. Some are moving quickly and others are skeptical. I wanted a resource that treated both sides seriously: what the evidence says AI can improve, where the risks are real, and the questions enrollment teams should ask before adopting it.

This approach draws on research and guidance from EDUCAUSE, NACAC, the U.S. Department of Education, NIST, the International Energy Agency, and academic studies, not on vendor claims. Skeptical perspectives are included on purpose, because addressing those questions often lays the foundation for smooth adoption.

AI is already here, though its applications have materially different consequences, and enrollment leaders need to understand those distinctions well enough to make deliberate choices. Please share this freely.

Two kinds of AI, and why the difference matters

“AI” as a blanket term covers a vast landscape, and most disagreement about it comes from various perspectives. A useful split is between AI that replaces a task and AI that supports a person’s judgment.

The labor research backs the distinction. A 2025 Stanford Digital Economy Lab analysis of payroll data and revised in August 2026 found employment for workers aged 22 to 25 in AI-exposed jobs ran 19% below trend. The declines were concentrated where AI substitutes for work; where it complements work, employment held or grew (Brynjolfsson, Chandar and Chen). The authors call these early indicators, not settled causal findings.

For an enrollment office, the practical question is not “AI or no AI.” It is which tasks, if any, a team is comfortable handing to a machine, and which it wants a machine to help elevate people and their impact.

Where AI is helping

Some of the strongest field evidence so far shows AI substantially improving human performance, especially for newer or less-experienced staff, when it is designed to augment rather than replace them.

• Newer staff get up to speed faster. In a study of 5,179 customer support agents, AI assistance raised issues resolved per hour by 14% on average and by 35% for the least experienced. Agents with two months on the job performed like six-month veterans. Customer sentiment improved and attrition fell (NBER, Brynjolfsson, Li and Raymond). Support work is not admissions counseling, but the pattern of lifting newer staff is relevant to offices with turnover.
• Small teams can cover more ground. After budget and staffing cuts, many offices carry more students per counselor. Tools that help sort who needs attention now, and who is fine for the moment, let people spend limited hours where they matter most.
• Leaders get answers faster. Questions about the funnel that once waited weeks for an analyst can be explored in hours. The value is in asking better questions more often, not in the tool deciding.
• Less time on drafting and repetition. A first draft of a routine message or a summary of a student’s history frees time for the conversations only a person can have.
• Staff are already using it. In a January 2026 EDUCAUSE survey of 1,960 higher education professionals, 94% had used AI for work in the prior six months, and 56% had used tools their institution did not provide (EDUCAUSE). A deliberate, approved approach is often safer than the unmanaged use already happening.

The common thread: benefits show up when AI hands a person better information and the person stays in charge of the decision and the relationship.

The risks and limitations

These risks are documented, not hypothetical, and none of them disappears because a tool is well built. Each can be reduced with the right practices.

The bias finding deserves the most attention. Predictive tools learn from past outcomes, and past outcomes reflect past inequities. A tool that is not checked for this can quietly steer attention away from the students an institution most wants to serve.

Environmental and community impact

The environmental cost of one person’s AI use is small, but the combined build-out of data centers is large and lands hard on specific communities. Both matter, and staff who have seen the local effects are raising a fair concern.

The big picture

Data centers used about 415 TWh of electricity in 2024, roughly 1.5% of global use. The International Energy Agency expects that to more than double to about 945 TWh by 2030, just under 3% of global use, with the U.S. and China driving nearly 80% of the growth (IEA, Energy and AI). The IEA also notes AI could save energy elsewhere, but says the net effect is not guaranteed (IEA).

The individual picture

Google reports its median Gemini text prompt uses 0.24 watt-hours, about nine seconds of television, and 0.26 mL of water, about five drops (Google Cloud). These are company-reported figures; outside researchers have questioned what they leave out, and image, video and heavy analytic workloads cost more.

What this means for a decision

Declining to use a tool in one office will not change the regional build-out, and using one will not meaningfully add to it. What an institution can control is its choices: prefer vendors who disclose where and how they host, favor focused tools over heavy open-ended use, and include environmental questions in procurement alongside cost and privacy. It is also reasonable for staff to hold these concerns and name them openly as part of the conversation.

Common staff concerns, answered honestly

Most staff concerns about AI are reasonable, and some are partly right. Here is what the evidence supports and where it is still unsettled.

The people raising these concerns are often the ones who care most about students. Their questions belong in the evaluation, not outside it.

Guardrails that make adoption work

The professional and legal standards enrollment offices already follow give a sound starting framework for AI. Three are most useful.

NACAC’s ethical guide

The 2026 edition asks members to “promote ethical practices in relation to the use of artificial intelligence” in line with “transparency, integrity, fairness, and respect for student dignity,” and to protect student data privacy “across all platforms, applications and technologies” (NACAC Guide to Ethical Practice).

FERPA

A vendor that handles student records can act as a “school official” only if it performs a service staff would otherwise do, stays under the institution’s direct control over how records are used and kept, and follows limits on redisclosure (34 CFR 99.31). Those conditions belong in the contract, not just the sales conversation.

NIST AI Risk Management Framework

A voluntary federal framework, released January 2023 with a generative AI profile added July 2024, organized around four functions: Govern, Map, Measure and Manage (NIST). It is a practical checklist for campus IT and governance committees.

Taken together, those standards suggest a practical set of working rules for enrollment teams:

1. A person makes every decision that affects a student. AI informs; it does not decide.
2. Nothing AI drafts reaches a student without a person reviewing it.
3. Results are checked by student group at least once a cycle for uneven accuracy.
4. Staff know what data the tool uses and can explain a recommendation in plain words.
5. The institution owns its data, can export it, and can end the contract cleanly.
6. Security is verified through standard documentation such as SOC 2 reports and the HECVAT, with single sign-on where campus policy requires it.

Questions to ask any AI vendor

A good vendor will welcome these questions and answer them in writing. Vague answers are a signal in themselves.

Purpose and people

• Does the tool act on its own, or does it inform a staff member who acts?
• Which student-facing actions, if any, happen without a person approving them?
• How do you expect our team’s daily work to change, and what training do you provide?

Accuracy and fairness

• Can staff see why the tool made a given recommendation?
• How do you test accuracy across student groups, and will you share those results for our data?
• What happens when staff disagree with the tool, and does that feedback improve it?

Data and privacy

• Will you sign FERPA school official terms, and what exactly do you do with our records?
• Is our data used to train models for other clients?
• Can we export our data and have it deleted when the contract ends?
• Can you provide a SOC 2 report and a completed HECVAT, and do you support single sign-on?

Cost and environment

• What is the full cost, including setup, integration and staff time?
• Where is the tool hosted, and what do you disclose about its energy use?
• What would a limited pilot look like, and how would we measure it?

A practical path to adoption

The most reliable way to earn buy-in is to involve skeptics in the evaluation and let results, not promises, make the case.

1. Start with a real problem. Name one specific pain point or opportunity, such as too many students per counselor or slow answers to leadership questions, before looking at any tool.
2. Put skeptics on the evaluation team. Their questions will find weaknesses early and make the decision more credible to everyone else.
3. Ground rules first. What the tool will and will not be used for, including performance evaluation, agreed before anything is bought.
4. Check the output side by side. Staff compare the tool’s suggestions with their own judgment and record where they differ.
5. Review fairness and environmental questions openly. Share results by student group and the vendor’s hosting answers with the whole team.

This is an interesting article by Johanna Alonso in Inside Higher Ed on the University of Tennessee’s efforts to attract more rural students.

We worked on a very similar problem about two years ago with another R1 state flagship. The question, coming from the state legislature, was pretty straightforward: why were in-state rural students enrolling at a lower rate than non-rural students?

The institution had already surveyed this extensively. We took a different approach — using our ensemble machine learning models, trained on historical enrollment data, to look for behavioral pattern differences between rural and non-rural students. In this process we reviewed 3 years of enrollment history, ~150 native enrollment data elements and ~350 newly engineered data points for every student.

The finding

Perhaps the most interesting finding was that the institution’s bulk communications appeared to work differently for rural students. The standard post-admit workflows and campaigns — how many emails were sent, opened, clicked and responded to in some form — were a noticeably weaker predictor of whether a rural student would enroll.

The same flow that served as a strong signal for the broader population wasn’t doing the same for rural students.

A thesis worth testing

This didn’t tell us exactly why rural students were enrolling at lower rates. But it did give us a thesis worth exploring: what if rural students needed (or wanted) a different post-admit communications and interaction chain?

Like a lot of large institutions, this one moved essentially everyone through the same post-admit flow. A new branch of this chain — built specifically for rural students — could mean different content, voice, volume, channel and pacing, built around how rural students were actually engaging with the institution.

The bigger point

There’s no simple answer here, and the Tennessee story has its own tensions. But sometimes the objective isn’t to wait for the definitive answer. It’s to use the data to build a new thesis, test it, and see if you can move the outcome.

Underneath the comms flows and probabilities, and behind the surveys, there’s a lot of previously unseen behavioral data that can help enrollment teams work with much more precision on the objectives that matter most.

I was in Chicago this week at Roosevelt University. Grateful to Scott Clyde and Mike Dolen for being part of the conversation, and to Roosevelt for hosting.

We talked about AI. The tools, the frameworks. That conversation matters.

But the tech isn’t the hard part.

The hard part is that when the pressure is on, every instinct tells you to go back to what you know. We’re too busy to change course. We already invested in this. The playbook feels safe, or at least less risky. Staying the course feels like control. It’s not. It’s just comfort.

That’s the real hurdle right now. Not choosing the right AI tool. It’s the willingness to ask whether the strategies you’ve relied on for years still make sense.

Drop your tools.

Adam Grant opens Think Again with a story about smokejumpers caught in a wildfire. The ones who survived dropped their tools and ran. The ones who didn’t held on to equipment built for a different situation. They weren’t undertrained. They just couldn’t let go.

The question Grant keeps returning to is “How do you know?” Not as a challenge. As an invitation to separate what you’ve tested from what you’ve repeated.

The willingness to put something down that used to work is what separates adaptation from exhaustion.

This is a great read about the fight to save Antioch College by Jonathon Podolsky in Medium — and I appreciate the reference to The Signal Solution.

I’ve seen this movie: I spent several years working in high-urgency higher ed turnarounds, often being the villain in the story — challenging unsustainable marketing spending, accelerating cost reductions, optimizing programs, sections and schedules, killing capital projects that didn’t substantively change outcomes, restructuring debt, searching for auxiliary revenue, challenging longstanding pricing methodologies, and sitting across from skeptical accreditors.

None of those were popular moves. But there are very few levers that can change trajectory quickly, especially in an industry that doesn’t exactly have a reputation for moving quickly, and where every down enrollment year compounds.

You have to find a way to grow enrollment.

Jonathon highlights something that sounds obvious from the outside, but can feel almost impossible from the inside: you have to find a way to grow enrollment.

In those situations, my focus almost instantly shifted to yield. The math is pretty compelling. A 15% yield rate means 85 out of every 100 admitted students said no. Find just 3 more students among those 85 and you’ve grown enrollment 20%, without increasing your cost of new student acquisition.

That’s why I appreciated Jonathon’s reference to Teege Mettille’s and my book, The Signal Solution.

The old playbook isn’t coming back.

The old playbook of driving more top-of-funnel volume and expecting enrollment growth to follow doesn’t work the way it once did. And it’s not coming back.

We have to transform how our technology and our teams drive enrollment. That means getting dramatically better at understanding and competing for the students already in the funnel, and giving our people the intelligence to know where and how their attention can make the greatest difference.

Strategic advisor, not task executor.

This is where AI can fundamentally change admissions. Not simply by sorting, automating, or telling a counselor who to call next, but by thinking deeply alongside the team. Understanding the full context of an individual student. Connecting signals across time and systems. Helping uncover the nuance behind what a student may be doing, thinking, or signaling.

That is a massively strategic role for AI in enrollment: strategic advisor, not just task executor.

Colleges are confronting foundational challenges around cost, pricing transparency, outcomes and value, and now an AI-driven teaching and learning environment.

The playbook built for a different era should not be expected to perform the same way in this one. Admissions has to transform and elevate.

Enrollment still solves. The question is whether we are willing to build the admissions organizations capable of solving it now.