Conversational AI in Education
“The decline in first-time enrollment at community colleges was a staggering 21 percent. Black, Hispanic and Native American first-year students showed even steeper drops in a November report, between 28 and 29 percent.”
After digging into the experiences faced by many college students, I discovered that a problem arose with enrollment as the coronavirus pandemic began. Typically, when the economy would slow down, the enrollment rate for community colleges would increase, but the opposite was true in 2020. With this problem in mind, I was tasked with creating a conversational chatbot that would help assist post-secondary schools to increase the rate of enrollment of first-time students.
Roles
User Researcher and Conversation Designer
Tools
Miro, Sketch, Voiceflow
User Research
To start off my research, I created a survey that would get feedback directly from students. My goal in this survey was to see why students used their school website and what they could and could not accomplish. Some of the questions included were:
1. What are the qualities you looked for in the school you chose?
2. Before enrolling, what were the main reasons you visited your school's website?
3. What were some of the things you couldn't accomplish when visiting your school’s website?
After looking over the responses from those surveyed, I saw that a lot of students visited the website to find out deadlines, cost of attendance, financial aid, and what the school offered.
One student even mentioned that the website was too cluttered and it was hard to navigate to find what they wanted. I saw this as an opportunity for Conversational AI to simplify the search for information on a school’s website.
Online Research
As I continued my research, I began to hear the experiences of other college students. For many, the search to find what college they wanted to attend was an overwhelming process. With students feeling this way and the change in the economy due to the pandemic, enrollment at community colleges began to decrease significantly. Here are some staggering facts I found:
“The decline in first-time enrollment at community colleges was a staggering 21 percent. Black, Hispanic and Native American first-year students showed even steeper drops in a November report, between 28 and 29 percent.
— Lanahan, from The Hechinger Report
“This significant decline in enrollment is particularly worrying for educational equity. After all, community colleges are often access points for low-income students, students of color, and adult students.
— Fishman & Nguyen, from New America
Definition & Synthesis
As I gathered my information, I put the information and student’s utterances into an affinity diagram. This allowed me to visually see the commonalities that students were facing as they visited their school’s website and how the chatbot could be leveraged.
Problem
Due to the economic crisis stemming from the pandemic, students have faced loss in income, access to amenities, and guidance, which has caused first-time enrollment at community colleges to decrease significantly. Students feel overwhelmed and alone in the process it takes to find the right school and enroll.
Solution
This chatbot is being designed to serve post-secondary schools by providing a contactless, user-friendly way to guide potential students to enroll at their school. How might we develop a chatbot to converse with future students to assist in the enrollment process and decrease the overwhelming process of looking for schools to attend?
Bot Persona & Needs
With the problem and solution clearly defined, I moved into creating the chatbot persona. The chatbot would need to both align with the company’s brand, tone, and business needs and have a personality that would speak to the user. In each use case, the chatbot also had specific needs it would express to the user.
Dialogue & Testing
For each FAQ, I would go through the same design process of creating dialogue, rewriting, testing, and iterating. As I would test the chatbot, I would pick up on specific utterances from the user and modify the structure of the dialogue.
Voiceflow & Continual Iterations
Once I hit a good stopping point of the basic layout of the FAQ dialogue, I would transfer my design into Voiceflow. Having the chatbot’s dialogue in this format, allowed me to clearly define the user’s intent and utterances and create repair flows for no input and no match.
Creating dialogue between user and computer is a constant process of analyzing what is/isn’t working and requires adding to the chatbot’s NLP and NLU. If you have questions about this case study or would like to chat, I would love to get connected!