How York University provides instant, personalized self-service to 50,000 students
York University needed to make a deeper personal connection with its students — all 50,000 of them. “One of the challenges for any large university is student satisfaction,” says William Gage, Associate Vice President, Teaching and Learning at York University. The university worked with IBM, using Watson Assistant to create SAVY, a virtual assistant that gives customized answers to a wide range of student questions about university life, from what’s for lunch to career advice.
“IBM guided us through the whole process, making sure we take on a human-centered design approach, making sure that we hear students’ concerns and ideas,” says Donald Ipperciel, Chief Information Officer at York University. The business owner for SAVY was the Division of Students, not a data team. The team identified pain points to address and created empathy maps and personas for SAVY.
Getting the right people in the room
“One of the things we did really well, right from the beginning, was identify the stakeholders and bring them to the table early,” says Gage. At the start of this project, he says, “there were a lot of people who were uncomfortable with the notion of artificial intelligence and machine learning, and really a lack of awareness of the potential.”
A diverse group of about 150 students helped to guide development and give SAVY its own biography and personality. York University is a bilingual institution, so SAVY needed to work equally well in English and in French. Ipperciel says the diversity of voices was crucial to the successful development. The project was initially focused on academic advising, but students led the team to other important cases, such as where to find bathrooms, and which dining options offer halal food. “You won’t get that question if your group is not diverse enough,” says Ipperciel.
York added a list of defined entities to SAVY, which help it to answer follow-up questions. When SAVY says an event is happening at ATK, for example, a student can ask “What’s ATK?” and get the answer.
Providing the right assistance, not just the right answer
While some virtual assistants simply scrub source pages and churn out answers, SAVY was built with much more sophistication. Say a student asks, “How do I drop a course?” the best response isn’t just “Here’s how,” says Ipperciel. If a student asked this question in a face-to-face meeting, their advisor wouldn’t jump to that step — they would ask why. Is the student having difficulty academically or perhaps financially? Do they feel like they’re not fitting in? “You want to dig deeper,” says Ipperciel. “And that’s what the tool does. That’s why we correct people when they say this is a bot. This is much more than a bot.”
SAVY also uses information about the student to provide a better answer. Students log in so SAVY can access relevant parts of their student profile, including what programs they’re in, what courses they’ve taken, and whether they’re international students. So a student can ask an open-ended question like “What course should I take?” and SAVY can give a relevant answer on the spot without asking the student other questions.
To provide this level of customization, York University also needed to preserve student data privacy. It holds itself to a higher standard than simply complying with regulations. Students must feel cared for but not surveilled.
Improving the university experience through continuous optimization
Students were impressed by how quickly SAVY learned, and how it could understand a question that had stumped it just a week before, says Gage. This helped students feel more confident and engaged with the program. It also enabled them to ask SAVY questions they might have problems asking in person. “The goal of SAVY is not to replace humans,” says Ipperciel. “On the contrary, often SAVY helps direct students to a real person. But in some cases, maybe students don’t want to talk to a person.” SAVY lets them ask without shame or embarrassment.
Student advisors were grateful for SAVY too. They had been spending a lot of their time answering the same straightforward questions over and over, and students might wait in line just to get a 30-second answer. Now advisors are freed up to spend more time on students with higher-touch issues.
Within its first year of operation, SAVY reached a confidence rate of 84%, which Ipperciel expects to rise with the start of a new year and returning questions. Meanwhile the team continually improves SAVY, increasing accuracy, monitoring popular questions to make sure they get answered.
Exploring more use cases for SAVY
The team is also expanding SAVY’s scope. They’ve already expanded the assistant into IT support, building on the existing IT ticketing system. They’re looking to increase SAVY’s features, so that instead of pointing to a page for changing a password, SAVY can change it directly.
And in the long term, they hope to apply SAVY to career planning, drawing from the experience of over 300,000 York alumni. In the future, a student could ask SAVY what courses to take in pursuit of a specific career, or conversely ask what career their desired courses would lead to. SAVY could give answers based on the actual careers of York alumni, says Gage.
They’re also exploring a similar virtual assistant for staff and faculty. “That’s going to be a whole different set of questions,” says Ipperciel, “and we’ll have to work with the HR side, just as we’ve worked with the Student Services side.”
SAVY is also delivering insights back to the university, says Gage. The student questions, he says, are “a treasure trove of knowledge that we can use to help shape new systems, even external to SAVY.” These questions have helped the academic division improve its own processes and provide better information to students.
To succeed with AI, Gage and Ipperciel agree, you need enthusiastic buy-in from stakeholders. “If the business side is not going to be involved, don’t even start the project,” says Ipperciel. If you’re serving students, involve students. If you’re addressing HR, work with HR. “Do not look at this as an IT project.”
“You have to commit to the project,” says Ipperciel. “A half-baked effort will not lead to half of the success. It will lead to 100% failure. The users won’t see value in the tool, and they won’t come back. You’ll just have another bot in the bot graveyard.”
To build a successful and transformational AI project, involve your target users and other stakeholders, find what excites them, and constantly consult them in development. You’ll find yourself solving more problems than you imagined.