AI Predictive Model Partnership Dramatically Raises CUNY Graduation Rate

• By Kate Lucariello
• 10/12/23

Seeking to improve graduation rates at City University of New York (CUNY), a three-way partnership between Google, DataKind, and CUNY's John Jay College of Criminal Justice (JJCCJ) built a predictive AI tool that resulted in a dramatic graduation rate increase — from 54% to 86% in just two years.

The AI tool will now be extended to six more CUNY schools to help improve their rates as well.

With the support of a grant in 2021 from Google's philanthropic arm, Google.org, nonprofit data science organization DataKind and JJCCJ built the AI predictive model based on data from thousands of students identified as most likely to drop out. Many of these were not traditional students, but faced challenges such as being first-generation, working while studying, or struggling to raise families while going to school.

The model looked at 75 risk indicators, such as grade variations and attendance patterns, to generate a risk score for each student. Those students numbered around 200 out of roughly 750 students assigned to each adviser.

With that score, advisers were able to focus their attention and resources, such as one-to-one support, on those students at greatest risk of not completing their degrees.

Following the success of the project, Dean Dara Byrne summarized four takeaways other institutions can use in building or incorporating AI to solve problems:

1. Co-create a solution: This builds transparency, collaboration, and mutual trust to help integrate the use of AI successfully.
2. Start small: The project began with one college in order to "get the model right," Byrne said, and "to bring the right historical data to the table and to keep a laser focus on results, while closely monitoring for risks like bias in the model."
3. Use AI as an aid: It can help advisers, but cannot replace them when it comes to giving personal support and helping students make good decisions.
4. Seek help: Organizations and companies like DataKind and Google can help strapped institutions fund and develop AI models and support sharing what was learned for the benefit of other institutions who want to help their students succeed.

"This project has fundamentally reshaped how I think about building a culture of belonging — informed by data, and powered by community," Byrne said.

Visit this page to read Byrne's full blog post.

To view a video for more specific information, go to this YouTube page.

Learn more about DataKind's work to help empower communities in the U.S., and visit Google.org to read about its philanthropy programs.