Georgia State University

Objective

Reduce summer melt, meaning the share of admitted students who confirmed their attendance in the spring but do not show up at the start of term. At Georgia State, this rate had risen from about 12% to nearly 19% according to Campus Technology, and mail, email and phone produced few results. The university wanted to follow each admitted student individually on their steps without hiring the team that this volume of conversations would have required.

The deployment

Pounce is an SMS conversational assistant built by AdmitHub (now Mainstay) for Georgia State University and launched in April 2016. It writes to admitted students during the summer between admission and the start of term: deadline reminders, information on financial aid, housing or course registration, surveys, and instant answers to questions asked at any hour (tuition payment date, sending ACT scores, the FAFSA form). The nudges are not the same for everyone: the system cross-references the university's data on the progress of each file and only pushes on the tasks where the admitted student is behind. Questions the assistant cannot handle go to the vendor's team or to Georgia State staff. The first season was run as a randomized trial: of more than 7,000 admitted students with a US mobile number, half received Pounce and the other half the usual communication by email and mail. The trial, led by Lindsay Page and Hunter Gehlbach and published in the journal AERA Open, measures among admitted students committed to Georgia State better completion of required steps and a 3.3-point higher probability of enrolling on time. Campus Technology, drawing on the vendor case study, reports for the treatment group a summer melt 21.4% lower than in the control group, nearly 200,000 messages exchanged, and fewer than 1% of the 50,000 student messages requiring staff intervention. The university then extended Pounce to the enrollment communications of about 40,000 students on its Atlanta and Perimeter College campuses, then adapted it into a course assistant from 2021. Two caveats. The strongest figure remains the one from the 2016 trial: the later scale-ups are not measured against a control group in the sources consulted. And the drop in summer melt from 19% to 9% cited by the university in 2024 is not attributed to the chatbot alone.

Results Proof A

None of the categories A to D literally describes the case: the core evidence is a randomized controlled trial published in a peer-reviewed journal (AERA Open, 2017), led by academic researchers and reproduced by Georgia State's institute (S1). It is primary, uninterpreted evidence of the causal effect, stronger than a vendor case study (B), hence the A rating by equivalence; it is corroborated by an official university press release (S4, extension to about 40,000 students), by the Mainstay case study (S5) and by the press (S3, S6). Caveat: the measured effect dates from the 2016 season and the vendor took part in setting up the trial.

1. 1
   Task inventory marketing

   Admissions list the required steps before the start of term and their deadlines, and write the university's own answer base.

2. 2
   File synchronization data team

   Each admitted student's progress data is passed to the assistant so it knows which tasks remain to be done.

3. 3
   Targeted SMS nudge AI

   Pounce sends each admitted student the reminders and information matching the tasks they have not finished, plus surveys.

4. 4
   Answering questions AI

   The admitted student writes in natural language; the assistant answers the vast majority of messages immediately.

5. 5
   Escalation human

   Unresolved messages go to the vendor's team or to Georgia State staff (fewer than 1% of student messages during the trial).

6. 6
   Reading intent and measurement marketing

   Changes of decision reported by admitted students in the conversation give admissions a continuous view of expected enrollments; the season is measured by actual enrollment rate.