Designing Conversations with Bella Bot
to help prospective students access information efficiently on the Pratt website
Client
Pratt School of Information
Duration
9 weeks
Team
Sagarika Konanuru
Role
UX Research, Personality Design, Conversation Design, Chatbot Prototyping and Testing
The Opportunity
The admissions process can be daunting for many students - people tend to resort to emailing professors, or current students at universities to learn more about the course and opportunities the institution offers. There is a variety of information that students require during admission, and it is often difficult to design a very clean and simple information architecture for it due to overlapping information.
​
One part of the solution to putting prospective students at ease would be making information related to admissions decisions easily accessible. A chatbot that helps assist students in finding the relevant information would therefore be a more efficient and easy way to help students find information as well as help staff by reducing their repetitive email load.
The Design Process
01
Research
-
Expert Interviews
-
Digital Ethnography
-
User Survey
02
Design
-
Personality Design
-
Sample Scripting
-
Intents and Slots
-
Conversation Flowchart
03
Validate
-
Wizard of Oz Prototype
-
VoiceFlow Prototype
Research
Expert Interviews
I interviewed Quinn Lai (Assistant Dean, School of Information, Pratt Institute) and Meredith Brull (Assistant Director of Academic Programs and Services, School of Information, Pratt Institute) - who have over 10 years of combined experience in interacting with students during the admissions process for higher education institutions.
​
I believe that their expertise in communicating with prospective students’ would shed more light on the need for the chatbot, and what subject matter the chatbot should contain - as well as the challenges involved in creating such a bot.
Key Takeaways:
indicating that the bot should be reassuring, and also have a flow/error handling that would lead the user to a human that could resolve their queries, if the bot was unable to.
"a lot of answering these questions is about hand holding. So I think it's important to kind of integrate within the chatbot like, - 'But don't worry, you will talk to Meredith about Advisement and Registration. And if you have further questions, you can always…’"
which reinforces that subject matter about courses and job opportunities are some of the critical topics the bot should cover, to help answer prospective students’ questions.
"they want <to know more about> classes and material and … number two is probably employment, job placement, internship questions, that's like a general category … When students are considering where to go, you know … what is this going to earn me?"
further reinforcing the fact that the chatbot needs to be warm, but more importantly efficient, and concise in conversation to truly help the students.
"personality, in my opinion … young and friendly, but also concise … I don't even know if it needs a personality … go to the place where the links are … super efficient … not try to engage in any kind of conversation, just present information right away."
Research
Digital Ethnography
To understand the actual queries of current prospective students, I analyzed the email thread between the IXD Student Perspective Graduate Assistant and the current prospective students (Fall 2022) between the dates of Mar 10, 2022 and Apr 12, 2022.
Using the text from the mail thread with 143 emails back and forth with current prospective students, I created a data set indicating the intents and the count of certain utterances - mainly related to the NYC experience, faculty, housing, job opportunities, and courses.
Bubble diagram created to convey data insights from the email thread
As seen in the data visualization above, the most ‘talked about’ topics within the email thread are classes, job opportunities, and finally housing. Though the general NYC experience and finances are also spoken about frequently, there are lesser instances of these topics as compared to the rest.
​
These findings supported the insights from the expert interviews with Quinn and Meredith - with respect to the subject matter that the chatbot must contain.
Design
Personality Design
Using the data and research insights from the expert interviews, and digital ethnography I then began to design the chatbot’s personality, subject matter, and conversation flow.
Interaction Goals
Guiding
Guiding
by pointing prospective students to resources that would answer their inquiries
Accurate
to assure students that the bot is reliable, and provides correct information
Efficient
since students require the information they seek quickly, and in a frictionless manner
Level of Personification
The chatbot is meant to be a helpful and friendly chatbot that leads the users (including those who consider English a second language) to resources, I believe that giving the bot a name and personality traits would be helpful.
​
Since it is to be used by students from across the world - I stuck to a two-syllable ‘easy to pronounce’ name - Bella.
Low
Medium
High
Personality traits and Tone
From our research insights, we can say that Bella needs to reassure the students that they will reach their goal, and the information they are receiving is legitimate, leading us to understand that Bella needs to be perceived as knowledgeable and confident.
​
However, since it is a stressful time for students, to help them reach their goal with ease and reassurance, Bella also needed to be supportive, and understanding of the student’s situation - and therefore I incorporated that as well into the personality traits of Bella.
Formal
Casual
Expert
Novice
Warm
Cool
Excited
Calm
Design
Sample Scripting
To help start and understand the conversation flow, and incorporate the personality specifications I planned for Bella, I created short sample scripts of the different flows that the conversation could take - namely Courses, Job Opportunities, and Housing. These scripts also included phrases and utterances that students tend to look for during the admissions process.
Bot: Hello there! Congratulations on your admission to Pratt Institute. I’m Bella, the Pratt SI Student Perspective Chatbot! What can I help you with today?
Human: I’d like to know more about the courses provided in the IXD program!
Bot: Sure! The course is split into two tracks - research and design! It’s completely okay to not be sure right now. Do you know which track you’d prefer to take?
Human: I’m not sure yet!
Bot: No worries! You’ll have an academic advisor throughout your time at Pratt, who’ll help you figure out what you like better. For now, what track’s courses would you like to learn more about?
Human: Research Track!
Bot: Okay perfect, research track it is! Here’s a link to the mandatory courses and electives in the research track of the IXD Program!
Courses
Bot: Hello there! I’m Bella, the Pratt SI Student Perspective Chatbot! What can I help you with today?
Human: I’d like to know more about the housing options!
Bot: Sure! There’s on-campus housing and off-campus housing, what would you like to learn more about?
Human: On-campus housing
Bot: Alright! Less than 30 percent of the graduate student population at Pratt stays in on-campus housing - it is obviously a personal preference of safety, privacy, and financial condition. Would you like to learn more about the types of housing Pratt offers?
Human: Yes!
Bot: Here’s a link to the different kinds of housing that Pratt offers: https://www.pratt.edu/student-life/student-affairs/campus-housing/
Feel free to go through the link and drop back into the chat for any further questions!
Housing
Examples of sample scripting and conversation flow prototyping for course and housing-related inquiries
Design
Slots, Intents and Utterances
User intents and related slots were created to define utterances that users might mention to Bella. A majority of this data was also obtained from the digital ethnography study, and the student emails. The utterances were later used to conduct the prototyping test.
​
Click here for the full set of slots, intents and utterances.
​
A flow diagram was made to understand what conversation flows users can take with Bella and to map out all the different possibilities of flow and error handling with the correct intents and prompts. The flow diagram also helped me determine how many intents need to be created and what vocabulary to use for the utterances.
.jpeg)
Wizard of Oz Prototype
Validate
By mimicking the chatbot through text (low-fidelity Wizard of Oz testing), I could successfully test the flow of conversation between a user and Bella, the Pratt SI Chatbot.
To build the prototype, I simply followed the flow of conversation made earlier with LucidChart, and the prompts which we worked on prior to prototyping.
​
After the test with the prototype, I further asked the participant several questions, to get a better understanding of their experience with the prototype - and if they see the need for the chatbot in the first place.
​
​
Results
After the prototype test and expert interviews, I could identify that the Chatbot was missing critical subject matter related to Finances, which I later added to the conversation flow.
​
Another flow which allows the user to look at both the suggestions made by the bot, by adding a ‘Both Intent’ that then redirects them to learn about one first, then the second option was also essential to the error handling process.
Voiceflow Bot
Validate
By using the software Voiceflow, that allows for no-code chatbot creation to create a high-fidelity prototype, I could successfully test the flow of conversation between a user and Bella, the Pratt SI Chatbot.
To build the prototype, I simply followed the flow of conversation made earlier with LucidChart, and the prompts which we worked on prior to prototyping - and built the chatbot with the features on the Voiceflow platform.
​
Similar to the prototype, after the testing, I asked the participant several questions, to get a better understanding of their experience with the prototype.
Results
In terms of flows, the 4 topics that were implemented (with the Finances flow added) in the chatbot for Prototype Testing 2 definitely covered most of the questions that the students tend to have.
The participants mentioned that they found the chatbot to be warm and very reassuring and they believed the information to be accurate since they were directed with links to pages on the Pratt website to find the information they were looking for.
​
By changing the prompt to include the utterances I anticipated from the participants - sometimes within quotes - the participants subconsciously chose within the utterances prompted within the chatbot.
Conclusion
Bella was designed to put prospective students at ease by making information related to admissions decisions easily accessible and assisting students in finding the relevant information in an efficient and easy way, as well as help Admissions faculty in reducing their repetitive email load.
​
Here are some steps that I’d like to implement in the future for Bella:
​
-
Build a prototype with a developer - something that I realized while prototyping on VoiceFlow was that there were limitations within the software - like being unable to bypass some parts of conversation, or branching conversation with slots - and I’d like to understand better what limitations I’d need to overcome while creating an actual product version of this chatbot.
-
Make this chatbot easily replicable - Since the conversation flow and subject matter is something that nearly all prospective students at all schools seek prior to admission, making this easily reproducible would be advantageous to multiple schools, as well as the different stages of admission - post deposit as well.
-
Redesign the chatbot to be accommodate better accessibility, both in terms of graphics with color contrast and fonts as well as by creating an optional voice user interface as well to accommodate ESL students, and those who cannot type.
​
If you'd like to interact with Bella, visit the VoiceFlow prototype here.
© May 2023 by Sagarika Konanuru.