The AskIT chatbot focuses on creating an interactive platform for New York University (NYU) students to get answers to their questions on various functional areas across various locations of NYU. The chatbot was designed to make a conversation enjoyable and drive efficiency, providing 24/7 service from different geographical locations. By launching this chatbot, we were able to decrease the traffic to customer service desk by 9%.
Role
In my role as UX Designer, I worked closely with a Product Manager and engineers. I was responsible for the end-to-end design process, from user research, conversation flow, wireframing, prototyping, and UI design.
This case study is a glimpse on how I learned conversational UX design. The engineering team taught me a lot about chatbot frameworks, such as Dialogflow and Elasticsearch, and how the chatbot processed the input through Artificial Intelligence.
Within 4 months of development, I was able to create a chatbot that wasn't just functional, but also fun to interact with.
The challenges
Students should focus on their studies and use the school facilities to leverage their education. Many students found it distracting when they couldn't use the school services. That's why many of them reach out to NYU's service center whenever they encountered some issues because they couldn't find the information on NYU directory pages. The inefficiency of the service center increased the traffic of customer calls by 40% which could decrease the latency between users' queries to give out an appropriate and timely response.
Design objectives
The chatbot should help the students to get their answers in a timely manner. It should work in tandem with human agents to reduce the incoming traffic rather than replacing human intervention. Besides that, since NYU's campuses are available across the globe, it should provide 24/7 service from different geographical locations. Contextual communication is also necessary to leverage help through building connections with students.
NYU AskIT Chatbot was designed to support students' education through leveraging help across the user journey into one simple platform.
User research
The quality of the conversation depends on the knowledge base and user inputs. NYU had a detailed database for all the information the students needed. However, it wasn't well-structured. My first attempt was to restructure the knowledge base. I conducted some interviews with several service center staff and students to understand their behavior in resolving the issues. Understanding the way humans interact to resolve an issue brought more context to how we restructured the knowledge base.
I mapped the data categories in the knowledge base to be as contextual as possible.
Designing conversations
I found that designing a chatbot was more similar to scripting a movie than designing any other digital platform. It required the knowledge of contextual inquiry to understand real-time conversations between the bot and the user.
Similar to creating a movie script, building contextual conversations is about maintaining the continuity of the conversation flow. I created the conversation flow to help the team build empathy to make the bot more human.
Chatbot personality
To bring more values to the bot, I noticed we had to design the conversation that could align with the inputs from students to make it more engaging. To keep the conversation going, I chose a tone and built the whole persona around it as the bot personality. The personality of the bot that is friendly, calm, smart, and passionately talk about technology would bring a fun experience when students interacting with the bot.
Visual design
Engaging students in fun communication should be from both conversations and visuals. To bring more personality to the bot, I created visual guidelines to show NYU brand's personality and make the experience more engaging. Since the target users were NYU students, I chose to use emojis to the communication direction. They weren't just making the visual more colorful, but emojis also added emotions to the communications. The symbols also helped the students to interpret the information faster.
The bot personality could be seen not only from the conversations but also from the visual aspects. Since we had more flexibility with the platform, I designed some illustrations to bring the conversations into a more fun way. The illustrations aligned with the information that the bot provided.
Conversation flows
The intent of the chatbot was to help students resolved their issues regarding school services as effective as possible. We noticed that providing suggestions based on the conversation would expedite the search. As the first design iteration, I designed 3 main paths of how the bot would give suggestions based on the inputs.
Happy path: I designed the conversation flow by "follow one with the next" categories. The suggestions showed one after another based on the user's previous selection.
Happy path: Showing the suggestions based on message inputs. I created conversation categories based on keywords, so the bot could provide the nearest possible suggestions.
Dead end: To prevent the user to get stuck in the conversation, I designed a dead-end path that showed popular suggestions when the bot unrecognized the user inputs.
Below is the full happy path of the conversation flow.
Detail matters
I believe that small details speak more to differentiate a great experience from the ordinary one. Conversation and the visual aspects could be the main focus when designing a chatbot. But, it's more about how to deliver the small micro-interactions to leverage the user's satisfaction.
A slight delay in conversation is natural to make the chatbot act as a person. I designed the typing animation to make the bot more human.
Clarity would enhance users' trust. It was necessary to identify the user if the bot was active or if there were any connection issues by adding read or delivered information.
It was necessary to notify the users of what was happening in a time frame. We created color gradients to notify the time flow in the conversation.
Users' inputs are important to train the bot. To enhance the quality of the data, we asked follow-up questions to make sure the conversation fell within the context of users' queries.
Visual aspects are important to engage users in the conversations. I designed the illustrations of the bot to bring more storytelling aspects to the interface.
Users' feedback
Creating a chatbot for the first time was a really big challenge. We had to validate our solutions to know if the bot conversations were understandable and to find out any usability issues with the bot. I conducted user testing with 5 users with different backgrounds towards their activities at school.
From the testing, we were able to prioritize what we needed to focus on for the next iteration. Many users still found it confusing how the bot responded based on their inputs and most of them didn't understand the technical terminology that was used in the information.
Below are our next steps what we had to do based on the testing findings.
Restructure the knowledge base, by simplifying the information, regrouping information from users' perspectives, adding other necessary categories, and updating the data with a user-friendly tone and terms.
Enhance the conversation flow by adding more use-cases and knowledge to the bot, adding more categories, and designing more dead-end cases.
Learning
Unfortunately, I wasn't able to contribute more since I left the team in 2019 because I graduated from NYU. Looking back, there are a lot of things that I'd like to improve. I'd improve the visual and interaction design by creating the animation more playful, and improve the experience by integrating the bot with a real person.
Takeaway
A chatbot is no longer a bizarre communication platform for us. But the way human interacts with the chatbot is still something to be continuously learned. A conversation is a basic human communication tool. It's natural for humans to do it unconsciously, but not for machines. Like a human, a machine needs some time to learn the conversation. In order to make machines learn the conversation, I believe the conversation designers have to abstract our thought patterns and unconscious habits into something that can be interacted in an organic way.
I also think that conversation is something that is not absolute. It is subjective based on the person who's interacting with. An important thing that I learned is that conversation is an infinite loop. In my personal opinion, what matters the most when designing conversations is to always practice writing the flows and scripts. Learning the users' conversation behavior will also benefit us to write how the bot's personality will meet the goals.
