The Human Future of Admissions Technology

The responsible use of AI in college admissions workflows is underpinned by two key elements. First, responsible use of predictive models requires an understanding of the “why” behind the predictions. A higher education leader must be sure they can point to the factors driving a prediction at both the cohort and student levels, and vendors need to build their tools with this need in mind. 

This transparency builds confidence and drives meaningful dialogue; the waterfall visual in Liaison’s Othot platform is a great way to see which factors are driving key metrics like enrollment or retention. For instance, a change in financial aid strategy – perhaps providing more merit aid than in previous years – could contribute to higher expected enrollment. The waterfall visual quantifies exactly how much of the prediction comes from that component, demystifying the assumptions underlying overall figures.

Second, responsible leaders pair AI with expert judgment. This applies to both predictive models and to the many applications of generative AI across the industry. A process built to use AI-generated results alongside expert judgment in context can engage, rather than sideline, knowledgeable professionals.

Institutions need to first realize that predictive models can often be like looking in the mirror. The patterns uncovered can be used to drive decisions in familiar ways, or to pursue a new strategy if desired. To avoid oversimplification, don’t skip the step of asking, “Do these patterns reflect how we want to be making decisions?” This analysis is all predicated on having an explainable tool and an informed team to guide the process. 

After gaining initial insight at the surface, savvy teams drill deeper to see how patterns persist or differ for key student populations. What might be true in aggregate might not be true for an individual student or a subset of students.

For example, an out-of-state student who makes even one campus visit has cleared a much higher bar in time, cost, and commitment than an in-state student who lives an hour away, making that single visit a significantly stronger enrollment signal for that population. In Othot, segmenting by residency can reveal that a campus visit is among the top predictive features for out-of-state yield specifically. Taking it a step further, teams can look at an individual out-of-state student who hasn't yet visited and use the what-if analysis to see how much their likelihood score would change if a visit were to happen, turning that insight into a clear reason to pick up the phone.

Guardrails are key when using AI to help make decisions related to enrollment. Something simple, like a change in campus visit data collection practices mid-year, can have massive ramifications for predictive model outputs. 

When using AI to inform enrollment decisions, institutions need to prioritize building data monitoring processes or work with vendors that can give them this support. The Othot Data Science team reviews data changes on an ongoing basis with automated tooling that can flag issues before they significantly distort model performance or recommendations. The team often catches changes in, for instance, major names or fund codes that, if unnoticed, would make downstream analysis unreliable. This is an area worth investing in: campuses constantly navigate staff turnover and system migrations which can drive downstream data risk if not mitigated by the right tools and team.