Improving support for students and saving time for advisers
To solve that problem Coder and her department embarked on a project in June 2019 to use artificial intelligence to automate responses to routine queries so advisers could devote more time to one-on-one student support. After their call for proposals, three companies presented demos. Coder reports that Quantiphi, a Google Premier Partner for Data Analytics, Machine Learning, and Marketing Analytics, and Google Cloud stood out for their fast turnaround and passion for higher education.
“Google and Quantphi met all our requirements, including security and FERPA compliance,” Coder says. “That’s really important to us. We started in June and it was completely built and ready to go in September. That’s extremely impressive.”
After formalizing a service agreement with Quantiphi in February 2020, the department rolled out the innovative virtual assistant to automatically handle student emails about re-enrollment, change of major or campus, and deferments. Testing has already shown that the virtual assistant is about 86% accurate at identifying intents and providing the correct responses, which improves as the model processes more questions and learns from them.
With the virtual assistant, students email their questions to advisers through a secure platform with Single Sign On authentication. Dialogflow, Google’s natural language processing tool, then analyzes them, gathering relevant information like the student’s last degree, enrollment dates, and campus. The FERPA-protected student data remains on Penn State’s own secure, local student information system (SIS) database. Quantiphi trained Dialogflow on the criteria to use and designed a custom interface to generate clear, concise, and standardized responses for the defined categories of questions. Any questions outside of the training criteria automatically get sent on to advisers, with the student’s information pre-collected for them. “Our advisers are really excited,” Coder says. “They like that the information is all in one place. They used to have to click on many different screens.” The built-in reporting feature tracks how long each question takes to answer and compares it to the manual response times, creating a clear metric for time saved.