Students do not leave because of finances or grades. They leave because belonging did.
Vincent Tinto published his departure model in 1975. Nearly five decades later, institutions are still not using it correctly.
Tinto's core finding was deceptively simple: students leave when they lose their sense of integration into the social and academic fabric of the institution. Not when they run out of money. Not when their GPA drops. Those are outcomes of departure, not causes. The departure itself begins the moment a student starts to feel that the institution is not a place for people like them.
This is called social identity threat, and it operates largely outside conscious awareness. A first generation student who walks past the financial aid office three times without going in — not because they do not need help, but because they are not sure they are allowed to ask. A transfer student who has attended for a semester but cannot name a single person on campus who knows their name. These are the early stages of departure. Your data system does not see them yet.
"The financial shock is often the event that triggers the withdrawal form. But the decision to leave was made weeks earlier, in a hundred small moments of not belonging."
Your Early Alert System Is Watching the Wrong Things
Most institutional early alert systems are built around academic indicators: missed assignments, declining GPA, attendance below a threshold. These are useful. They are not sufficient.
By the time a student appears in your early alert dashboard, they have typically been disengaging for 6 to 8 weeks. The behavioral and relational signals that predict departure appear well before any academic indicator surfaces. But they require human attention to catch, not data entry.
The practical implication is uncomfortable: the most important retention work happens in the spaces your systems cannot see. It happens in the advisor's office, in the residence hall, in the moment a faculty member learns a student's name and uses it again the next week. These are not soft interventions. They are the interventions the research says work.
What Predictive Analytics Gets Wrong About Belonging
AI powered retention tools have proliferated in higher ed over the past five years. Most of them are built on the same logic: aggregate behavioral and academic data, generate a risk score, trigger an outreach when the score exceeds a threshold.
The problem is not the technology. The problem is the theory of change underneath it. Predictive analytics can tell you a student is at risk. It cannot tell you why — and the why is everything. A student who is struggling financially needs a different intervention than a student who is struggling to find their place. The risk score looks the same. The solution is not.
The most effective use of AI in retention is not as a replacement for human judgment, but as a triage tool that frees advisors to spend more time on the relational work that actually changes outcomes. That requires a very different implementation philosophy than most vendors are selling.
Until next issue — build systems that see what the data cannot.