How agentic AI helps prospective and existing students at DeVry

DeVry University has turned to agentic AI to solve a major pain point for its students.

Founded in 1931 as DeForest Training School in Chicago, DeVry serves more than 32,000 students with both online and hybrid educational programs in technology, business, and healthcare. About 83% of the private for-profit university’s undergraduate students are older than 25, with most holding full-time jobs, and about 50% are parents.

The problem? They’re doing most of their learning in the hours the university typically has no faculty or advisors available to help when they have questions.

“Our students are doing the vast majority of their learning at the end of their workday, after they’ve made dinner and put the family to bed,” says DeVry CIO Chris Campbell. “We see a lot of learning activities happening after 7 p.m. and up until about 1 a.m. Mountain Time.”

The university was also facing a crunch onboarding new students, with an average of 4,700 new students for each eight-week session. In the past, preparing new communications required manually moving lists between DeVry’s learning management system (LMS), the student community platform, and Salesforce Marketing Cloud. To address these challenges, the university implemented agentic AI with the help of Salesforce.

Core curriculum

DeVry is no stranger to AI. It’s used the technology in its classrooms for 10 years and started experimenting with NLP bots and gen AI use cases for internal use as soon as it became widely available. So in April 2025, it deployed its first AI agent.

Leveraging Salesforce Agentforce, DeVry developed DeVryPro, which provides prospective students with information about courses and the benefits of online learning at all hours. They can start a chat with the agent when it’s convenient for them, and it can respond instantly with current information about the courses, enrollment processes, and learning experience. DeVryPro can also help prospective students find, enroll in, and pay for courses.

For existing students, the agent extends support and care through the university’s portal, replacing DeVry’s legacy chatbot. Campbell says that, in addition to making round-the-clock, on-demand support possible, it’s also scalable, enabling DeVry to support those students without increasing service costs.

Abilities of an agent

DeVryPro pulls information from unstructured data across the university’s student handbook, academic catalog, and career services material to provide answers about a wide variety of topics. Campbell notes that human employees remain in the loop, so if students ask nuanced questions outside the agent’s scope, it routes the query to a student support advisor, and provides the advisor with context from the conversation the agent held with the student.

“We’re identifying the help students need, and when and where they need it, sometimes before the student knows they need that help,” Campbell says. “Think of it in terms of a digital faculty member that’s able to help drive the education and learnings of someone in real time.”

It takes a C-suite

Campbell says his experience with agentic AI shows that success requires the engagement of the entire C-suite.

 “You need all the stakeholders in every way, because you’re talking about fundamentally changing how the organization operates,” he says.

Organizational processes have to be reengineered with agentic AI as a core consideration. Campbell advises other technology leaders that process engineering is essential to success with agentic AI. Many existing business processes contain vulnerabilities that cause agents to stumble, so an agent can’t just be slapped on an existing process without consideration.

“A good way to fail is to apply agentic AI on top of a process that’s not all that optimal in the first place,” he adds. “We’ve deployed agentic AI and found the process it was built on held us back, and then we had to rebuild that process. So many processes in enterprises are built on centralized models that agentic AI blows up.”

He adds it can be beneficial to fail in this way, however, especially early in your agentic AI journey, as it’ll help develop the experience needed to succeed down the road.

“You want to look for choke points in the business process,” he says. “Look for breakage around SLAs or where you have long SLAs for what should be a relatively short task.”

Campbell continues that it’s commonly accepted wisdom that you must get your data in order before jumping into agentic AI. He concedes that data is tricky, but that’s not a good reason to wait.

“You’re never going to have the data perfect before you start, so you have to just start,” he says. “You have to test and learn, and be adaptable.”