Vettd.ai partnered with an association of colleges and their admissions departments to create a solution to their most pressing problem – too many post-graduate applications and not enough time to evaluate them thoroughly.
This association initially approached Vettd.ai to help explore AI as a possible means for making the admissions process more adaptable to environmental changes, such as the cancellation or postponement of standardized testing sessions during the COVID-19 pandemic. With oversight over dozens of colleges, each with their own unique admissions process, the association is responsible for making sure each school has the tools they need to succeed.
On average, each college receives about 5,000 applications for a class of 165 and has a team of two to three people who evaluate each one. The applications are 15 pages long. Evaluators pride themselves on thoroughly evaluating every application, however this is incredibly time consuming and only represents a small part of what they're responsible for. They end up with less time to engage their most promising applicants. On top of that, the sheer amount of information makes it difficult to pick promising applicants out of the group that are neither clear winners or bad fits.
In the middle of an application cycle, Vettd worked alongside over a dozen colleges to support their decision-making with AI. What we discovered together was that not only could an AI model speed up their evaluations, but it could actually make predictions that no one thought were possible.
During the admissions cycle, Vettd created a proof-of-concept model capable of predicting the likeliness that a specific student would enroll at a given college. This is a measure referred to in admissions as matriculation. The model was trained using two years of personal statements and the corresponding student outcomes.
The initial model was promising. This meant that within personal statements, there are hidden indicators to show not only that a student is a good fit for the college, but also that the student is likely to enroll at the college. Given the complexity of admissions decisions and all the factors that go into making an enrollment decision on the student side, this was a truly astounding discovery.
Next, we turned our attention to predicting who should be interviewed. With the same historical data, we were able to create a model that gave admissions teams an idea of who they should be interviewing first. This model was even more accurate than the first and introduced a huge added benefit. It identified promising applicants that were overlooked in manual review. These alternative student selections could lead to the creation of a more diverse class - a huge focus for admissions teams.
This interview prediction model represented a solid starting point for integration into live admissions workflows.
The implications of these AI models are far reaching beyond this subset of colleges. During their next admissions cycle, these predictions will be readily available inside Liaison’s Centralized Application Service (CAS), the software they rely on to review applications. With this intelligence at their fingertips, the admissions teams will be able to: