evidoria

← Back to browse

Good practice

Georgia State University 'Pounce' — AI advising chatbot that reduces summer melt and boosts retention

United States of America · Atlanta · See the United States of America profile

Georgia State's AI text-chatbot handled 200 k+ student queries a year; three RCTs (2017–22) found 3 pp gains in summer matriculation and fall re-enrolment, and 5–6 pp more students earning B+ or above in a course. Strongest effects for Pell-eligible and first-generation students.

3.3 pp
Increase in timely summer matriculation (2017 RCT, n=6,828)
3 pp
Increase in fall re-registration (2018 RCT)
~1,300 students
Students retained via re-registration effect (2018 RCT)
5–6 pp
Increase in probability of earning B+ or higher (2021–22 classroom RCT, n=500)
32% to 54%
Six-year graduation-rate change (over 15 years)
7.6 $M
Federal grant to scale AI-enhanced advising (Jan 2024)

Details

Maturity
Scaling
Promoter
Georgia State University
Period
2016–present
Keywords
higher education, student services, AI chatbot, advising

Context

In 2016, Georgia State University partnered with AdmitHub (now Mainstay) to deploy 'Pounce', a text-message AI chatbot answering common student questions about financial aid, registration and course advising around the clock, routing complex queries to human advisors and sending proactive nudges at key decision points.

Results

Three separate RCTs document Pounce's impact. A 2017 RCT (n=6,828) found a 3.3 percentage-point increase in timely summer matriculation, with the largest gains for Pell-eligible and Black students. A 2018 follow-on RCT found a 3 pp increase in fall re-registration, translating to roughly 1,300 students who would otherwise have stopped out. A 2021–22 classroom study (500 political-science students) showed chatbot messaging raised the probability of earning B+ or higher by 5–6 pp and reduced course dropout (EdWorkingPaper ai22-564/23-734). GSU's six-year graduation rate rose from 32% to 54% over 15 years, with gaps for African American, Hispanic and Pell-eligible students narrowing by more than 30 percentage points — though this broader gain reflects a bundle of interventions, of which Pounce is one.

Conclusions

The RCT evidence specific to Pounce shows modest but real effect sizes (3–6 pp) rather than the dramatic figures sometimes cited for overall institutional improvement. In January 2024 the US Department of Education awarded GSU's National Institute for Student Success a $7.6 million grant to scale AI-enhanced classroom chatbots to a wider network of institutions.

Implementation

Indicative cost
Medium (€50k–€500k) — No per-institution price disclosed; the $7.6M federal grant funds scaling across a wider network of institutions, suggesting a medium-cost SaaS-style advising platform rather than a low-cost or very-high-cost system.
Time to results
Long (> 3 years) — Continuously operated and expanded since 2016, with a 2024 grant funding further scale-up; timeline_band set to long.
Staffing & skills
Human academic/financial-aid advisors handle escalations from the chatbot, IRB oversight for all three RCTs

Conditions for success

  • Proactive nudges timed to key decision points (deadlines, financial-aid verification, mid-semester check-ins)
  • Clear human-escalation path for complex queries
  • FERPA-compliant data handling

Where it fits

Scale
large public research university, US higher education

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful