Credit transfer is a point of friction for students moving between institutions and for the administrators that manage it, according to Daniel Knox, director of the Center for Data & Analytics at the National Association of Higher Education Systems (NASH), a partner organization in the CourseWise pilot. Knox, who previously served as assistant provost at the State University of New York (SUNY), said evaluation practices vary from campus to campus and aren’t always systematically tracked.
CourseWise, a new tool developed by the University of California, Berkeley lab studying Computational Approaches to Human Learning (CAHL), uses AI to learn from institutions’ past transfer credit decisions and recommend which courses at one college are most equivalent to courses at another. Administrators can then approve or deny the suggestions, minimizing time spent searching through course catalogs.
THE DATA LANDSCAPE
Heather Adams, a transfer consultant at higher-ed consulting firm Sova Solutions who has worked on transfer research and policy efforts for years, described the national landscape of higher-ed data as “a hot mess.”
“Everybody’s organizing it differently, everybody’s collecting it differently, everybody’s defining it differently, and practically no one’s collecting it on transfer,” she said.
CourseWise Chief Operating Officer Angikaar Singh Chana pointed out that the data may be housed in different formats — for example, a course catalog in a physical book, student transcripts in PDFs not easily readable by machines, and past course equivalence decisions in Excel files with different column headers. While staff were previously tasked with comparing information across these systems, he said, CourseWise aims to unify the data into one screen with simple suggestions.
For example, Texas A&M University requires students to take a course on Texas government, according to the university’s Assistant Vice Chancellor for Advising Isaiah Vance. A faculty member’s opinion on what is most important for students to take away from the class — the specific government events and practices in Texas, or a broader consideration of how state governments work — might impact whether a past course is accepted for credit or not.
“I know Georgia has a state government class, too,” Vance said. “Yes, the exact laws are a little bit different, but the way that state governments work in general is quite similar, regardless of what state you’re in.”
Course articulation can be time consuming for faculty and staff, and ultimately frustrating for students who struggle to navigate it or end up retaking courses with similar content, Adams said. According to a report by the U.S. Government Accountability Office that analyzed Department of Education data on student transfers from 2004-2009, on average, students who transferred institutions lost about 43 percent of the credits they had previously earned during the transfer process.
HOW COURSEWISE WORKS
CourseWise is built on more than a decade of research in the CAHL lab on how humans trust and interact with AI-generated course recommendations. Partners say its research origin makes it unique among ed-tech tools, offering more transparency than a proprietary company might and ensuring its features are based on documented trends and user feedback.
“There were published papers before there was a platform or anything that people could use,” Knox said.
Zachary Pardos, the head researcher at UC Berkeley who created the tool, initially built models trained on millions of existing equivalencies and enrollment records at the SUNY system. Early versions produced 10 to 12 suggested course equivalencies and occasionally surfaced poor matches, according to Vance.
“I remember there were some where somehow physics had included some carpentry courses,” he said.
As it stands now, the tool offers one primary suggestion for the course at a receiving institution that best fits the course a student completed elsewhere. Through validation testing on the SUNY data, the model has improved its suggestions significantly.
“That was what the algorithm needed to get the accuracy on par with human mappings,” Knox said.
To use CourseWise, institutions upload their own articulation histories and course information. Chana helps schools organize their information to ensure it is formatted appropriately, and CourseWise has published suggested standardization practices that make articulation data work with the platform.
At Texas A&M, three of the system’s 12 universities are preparing to move beyond the input stage to testing the tool directly, Vance said. At a college system or state level, users hope using the tool can reduce duplicated work.
“We’ve got 37 public universities in Texas,” he said. “If each of them are reviewing courses independently, that is a lot of wasted resources.”
THE FUTURE OF COURSEWISE
Developers of CourseWise are looking to grow its features in the future, including expanding the kinds of data schools can input to include PDFs and developing a student-facing tool to help them plan their degrees.
“It’s sort of helping address that rough handoff that somehow sometimes happens with advising,” Pardos said. “[A student could] continue using the planner once they’re at the receiving school, because the planner is cognizant of the degree requirement rules of the receiving school.”
Partners testing CourseWise also see broader implications.
Data from CourseWise may help institutions better understand students’ academic pathways and find trends in the courses they take, Vance said. He also sees an opportunity to expand reciprocal transfers, thinking beyond the pipeline from two-year institutions to four-year ones, to include how university credit would transfer to a community college, for example.
Simplifying the student experience could have ripple effects on concerns about enrollment in higher ed, as well.
“One of the things that we’ve really discovered is that speed matters. Things like admissions decisions, credit evaluations, the handoff between admissions and advising, those kinds of things, they take a really long time,” Knox said. “We’ve taken processes around those kinds of things, say, at a campus where it would take on average, 45 days, and got it down to 48 hours or 72 hours, and we see the enrollments go [up].”
