Usable Knowledge is a trusted source of insight into what works in education — translating new research into easy-to-use stories and strategies for teachers, parents, K-12 leaders, higher ed professionals, and policymakers. Usable Knowledge is produced at the Harvard Graduate School of Education by Bari Walsh (senior editor) and Leah Shafer (staff writer). Contact us at firstname.lastname@example.org.
Smarter Tech to Smooth the Path to College
How personalized texting (from a virtual counselor) can efficiently reduce summer melt
After college acceptance letters arrive, the complexity and sheer number of tasks required to actually enroll — complete FAFSA, submit a final transcript, pay a housing deposit, obtain immunizations, among many other things — thwart the plans of many high school grads to matriculate. Between 10 and 40 percent of intended students fall into this “summer melt” pattern, and aspiring first-generation students, who are more likely to lack the prior knowledge and support to complete these steps, are particularly susceptible. Guidance counselors and admissions officers can provide valuable assistance, but it can be nearly impossible for them to be with every student, every step of the way.
Research has shown that sending students text messages with tips and reminders throughout the admissions process is a successful intervention. Now, a new study shows how that approach can go one step further, by using an artificial intelligence system — a virtual admissions counselor — to send personalized texts that reflect an awareness of which tasks students have finished and which they still need to complete.
The study found that with this data-informed, personalized guidance, students are more likely to enroll in college, while real counselors have more time to help students with complex circumstances.
The study, conducted by education policy analyst Lindsay C. Page and educational psychologist Hunter Gehlbach, examined the impact of an artificial intelligence (AI) system on enrollment at Georgia State University (GSU) in the fall of 2016.
In April 2016, GSU began to use Pounce, a program developed by AdmitHub, to send targeted text messages to admitted students about what they needed to do prior to matriculation. For example, the system would ask if the student was bringing a car to campus. If the student responded “no,” he wouldn’t receive any additional info about parking and registration; if “yes,” he would.
Pounce tailored its messages based on the data GSU had on each student. A student who had already submitted her FAFSA would not receive FAFSA-related outreach, for instance, but a student who had not yet submitted it would receive reminders and suggestions.
When the program was unable to answer student questions, it forwarded them via email to university admissions counselors. The counselors’ responses were then incorporated into AdmitHub’s system, with the goal of fewer staff interventions over time.
Messages were sent to 3,745 admitted students. 3,744 others, in a control group, received regular communication from GSU admissions. Approximately one-third were aspiring first-generation students.
The AI intervention had a powerful effect on matriculation at GSU.
- Students who received messages from Pounce were 3.3 percent more likely to enroll than those who received regular communications outreach — translating to a 21.4 percent overall reduction in summer melt at GSU.
- These students were also 6 percent more likely to complete college loan counseling, 3 percent more likely to attend orientation, 4 percent more likely to submit a final high school transcript, and 3 percent less likely to have a FAFSA verification hold on their financial aid.
- Significantly, while just as effective as other texting interventions, this campaign was more efficient. Pounce worked at a fraction of the price of other text-based interventions; because the system developed over time, it required far less human involvement.
Getting to College – and Persisting
While researchers have known for several years that text reminders can improve college enrollment, a key takeaway from this study is how effective these messages can be when they come from a system that can continually refine its outreach strategy, said Page in a recent talk at the Harvard Graduate School of Education.
Because Pounce had access to GSU’s student data, it could tailor messages to each student’s individual needs, so that she would perceive the program’s messages as relevant and therefore engage with them. And because Pounce’s database expanded as admissions officers intervened with new responses, the system continuously became more accurate and targeted.
Of course, guidance counselors and admissions counselors will always play a vital role in helping students apply to and enroll in college, as Page noted. But if counselors can make use of AI programs like Pounce to answer frequent, commonly recurring questions for large numbers of students, they will likely have more time and capacity to assist the students with complex needs.
And this type of data-informed outreach could go beyond mitigating summer melt. It “has the potential to improve college persistence and success even after enrollment,” said Page. “Colleges know what administrative steps students need to take and whether they have done so; they can know how students are doing academically, and they can act on this information to reach out students proactively.” Artificial intelligence systems could “connect students to supports and resources in response to early signs of difficulty” throughout their college careers.
- A summer melt handbook, from the Center for Education Policy Research at Harvard University, which delves deeper into the problem of summer melt and helps schools design interventions to combat the problem
- More on how text messaging can reduce summer melt
Get Usable Knowledge — Delivered
Our free monthly newsletter sends you tips, tools, and ideas from research and practice leaders at the Harvard Graduate School of Education. Sign up now.