More incoming students complete college enrollment

Pounce, built on AdmitHub’s platform. [1]

The authors describe supervised learning with convolutional neural networks and a university-specific knowledge base. [1]

The article does not identify a release number or the deployed network’s weights. [1]

AdmitHub. [1]

Pounce selected relevant reminders and answered questions about enrollment. [1]

The SMS assistant used Georgia State’s student records and knowledge base to tailor messages about unfinished tasks. [1]

Inputs included student questions and completed enrollment tasks. Outputs were text replies, reminders, and requests for staff help. [1]

Pounce sent messages and routine replies automatically. Questions that it could not answer went to university staff. [1]

Pounce checked which tasks remained, sent reminders, and answered questions. Counselors handled unresolved questions by email through the platform. [1]

Staff supplied answers and reviewed conversations. Their corrections helped train the system and expand its knowledge base. [1]

The assistant used machine learning to interpret questions and choose answers. [1]

The trial compared the complete Pounce service with usual enrollment support. It did not isolate machine learning from reminders and staff assistance. [1]

Not assessed

Pounce increased enrollment among students who had already committed to Georgia State. [1]

Increased access to education. [1]

Observed in a randomized evaluation. [1]

Incoming college students. [1]

Admitted students who had already committed to attend Georgia State. [1]

The trial assigned 7,489 students. The subgroup with the reported enrollment gain contained 1,948 students. [1]

The effect was concentrated among students already committed to Georgia State. The overall enrollment effect was not statistically significant. [1]

The outcome was enrollment for fall 2016. The trial did not measure degree completion. [1]

Reducing the loss of committed students before enrollment was the intended benefit. [1]

The service required staff to maintain answers and resolve difficult questions. Students could opt out of messages. [1]

AIBD-0215-E01 [1]

Students were randomly assigned within groups defined by whether they had already committed to Georgia State. [1]

Researchers used university administrative records and National Student Clearinghouse enrollment data. [1]

There were 3,745 students assigned to Pounce and 3,744 controls. All had a U.S. mobile number and had agreed to receive texts. [1]

Enrollment, financial-aid tasks, and other requirements for entering college. [1]

Georgia State’s usual enrollment support. [1]

In the committed subgroup, enrollment rose by 3.3 percentage points from a control rate of 84.6%. [1]

The trial did not separate automated conversation from personalized reminders and counselor support. [1]

The gain applies to the committed subgroup at one university. Overall enrollment rose by 1.2 percentage points, which was not statistically significant. [1]

Of students assigned to Pounce, 96.2% received texts, 85.1% replied, and 6.6% opted out. [1]

The article does not link to participant-level data or analysis code. [1]

Platform price and a modeled annual budget. [1]

The platform cost US$7–15 per student per year, plus staff work. The authors illustrated a Georgia State budget of about US$53,000 a year. [1]

U.S. dollars at the prices discussed in the 2017 article. A separate price-base year is not given. [1]

Per student per year, or per university year for the illustrative budget. [1]

The price range and annual budget are estimates, not an audited statement of Georgia State’s payments. [1]

Universities pay for the platform and contribute staff time. [1]

The figures are costs. The paper does not give a measured net financial benefit. [1]

The subscription excludes staff time for setup, monitoring, and answers that require a counselor. [1]

The authors compare the price with earlier counseling and text-message interventions. Those programs were not cost arms in this trial. [1]

The illustrative annual budget applies a US$15 per-student price. Additional tuition income is discussed but not quantified. [1]

The authors suggest that added tuition would exceed the platform cost, but they do not publish a return or payback calculation. [1]

AdmitHub built the assistant, Georgia State provided student information and counseling, and academic researchers evaluated enrollment. [1]

Georgia State University in Atlanta, Georgia, United States. [1]

AdmitHub supplied the platform. Georgia State supplied its institutional knowledge and student records. [1]

The service needed SMS access, current student records, a maintained knowledge base, and counselors for unresolved questions. [1]

The trial involved a U.S. university and students with U.S. mobile numbers. It does not document international data transfers. [1]

Not assessed

Universities need accurate task records, students who can receive texts, and staff who can answer unresolved questions. [1]

Two later Georgia State trials reported 5–6 percentage-point increases in grades of B or higher. Their courses and students differed from this enrollment trial. [3]

Later Pounce trials addressed coursework rather than summer enrollment and reported improved course grades. [3]

Not assessed

Not assessed

Not assessed

Not assessed

Not assessed

AIBD-0215-source-0001, AIBD-0215-source-0002, AIBD-0215-source-0003 [1] [2] [3]

The 2017 trial supplies the enrollment findings and system description. WABE adds a student’s firsthand account. A later paper concerns coursework. [1] [2] [3]

English. [1] [2] [3]

The trial was conducted with the developer and university. WABE’s student interview adds independent reporting, while its trial figures come from the original paper. [1] [2] [3]

The enrollment findings originate in the 2017 trial. WABE adds a student’s account, and the later paper evaluates coursework. [1] [2] [3]

The published enrollment article and WABE’s original report were read. The later course study was inspected only through its title page and abstract. [1] [2] [3]

Pounce helped already-committed students complete enrollment. [1]

The sources are linked to the findings that they support. [1] [2] [3]

A peer-reviewed randomized trial supports the subgroup effect. [1]

Not assessed

Students’ access to college. [1]

Personal: 2 [1]

More already-committed students completed college enrollment. [1]

Observed in a randomized evaluation. [1]

Not assessed

Not assessed

Not assessed

Independently studied [1] [2] [3]

A randomized evaluation of text-message support for admitted students. [1]

Not assessed

Higher education. [1]

Pounce helped admitted students complete the steps required to enroll. [1]

Messages ran from April through August 2016. The study measured enrollment that fall. [1]

Georgia State used Pounce for incoming students during summer 2016. [1]

SHA-256: 67e54baea04d…

Not assessed

Not assessed

Not assessed

6 revisions. Latest: 25 October 2026.