Ari
Conversational Interface
Ari is a concept chatbot application designed to make it easier for individuals with a history of alcohol use disorders to find the treatment options that are the most relevant to them. Instead of using complex guides called decision aids to seek treatment, Ari serves as a personal assistant to help individuals with alcohol use disorders stay on the right track towards recovery. This project was part of the Human Centered Design & Engineering (HCDE) graduate curriculum at the University of Washington.
Role: UX Designer
Time: 10 Weeks
Result: Interactive Prototype
Designing for conversational AI
As the designer for this project I was responsible for drafting end-to-end user flows, interactions, and prototyping the high-fidelity build. I also contributed to generative research and the usability study. For this project, our team worked with a scientific investigator in the medical field, and learned that documents called decision aids are designed to help guide addiction patients towards treatment options. However, these documents have been found to be long, complex documents that use industry jargon that the average patient isn’t familiar with; setting the stage for our project.
Design Problem
How can we design for individuals who are seeking information and support for alcohol treatment?
Research
Observation
Due to the difficulty of recruiting individuals with alcohol use disorders for research, especially ones who are looking for possible treatment options, we decided to observe Alcoholics Anonymous (AA) meetings. We came to this decision because AA is the most common treatment option used by people with alcohol use disorders.
Interviews
To learn more about this topic and understand how patients research treatment options, we decided to conduct a series of interviews with SMEs. We spoke to 16 people including addiction counselors, scientific investigators, and individuals with a history of alcohol use disorders who remained anonymous.
Secondary Research
Due to the plethora of research out there that has already been conducted around alcohol use, we decided to review qualitative research found in the U.S. National Library of Medicine’s PubMed database, and conduct a content review of both a decision aid and online treatment navigator.
Key Findings
Support networks serve as a way to seek treatment and improve patient morale.
There is a feeling of shame around seeking help publicly.
Patients have little knowledge of the treatment options that are available to them.
Decision aids lack personalization, and are daunting to look at, but the information they contain is incredibly valuable.
Anonymity is crucial.
Persona Highlight
Based on our research, we narrowed our target user group to patients diagnosed with alcohol use disorders, and are seeking treatment options that fit their lifestyle. We synthesized our insights into two personas that would guide us through the design process. Our primary persona, Jonah, wants to seek treatment options on his own, while Brenda is open to having others help her on her journey to recovery.
Jonah, IT Consultant
“My family encouraged me to seek help. Getting support from a network of people that care about you has given me confidence.”
Goals
Seek guidance from a network of people
Feel supported when making progress
Find help to overcome his obstacle
Brenda, Event Planner
“After falling back on an old habit, I feel ashamed to let others know. I want to overcome this challenge on my own.”
Goals
Feel she has accomplished something on her own
Seek unexplored treatment options
Remain anonymous when seeking treatment
White-boarding session using sticky notes and construction paper.
Ideation
It was during our ideation meetings that we decided on a chatbot called Ari as our design solution. The name Ari is gender-neutral, and also serves as an acronym for "Alcohol Recovery Information". The following points are areas our design sought to address:
Avoid stigmatizing language and common myths
Address common barriers to treatment
Facilitate trust by being friendly and inviting
Create a personalized recovery experience
Suggest relevant treatment options to users
User Flow
Once we had outlined the primary purpose Ari would serve from our ideation meetings, we needed to put together a user flow that we could test and iterate upon through user testing. The following user interface flow is designed to first welcome the user to the app with an inviting message, then survey them about their drinking habits and related preferences, and finally to present chatbot interactions. Once the user completes the survey, an initial treatment suggestion is issued along with supplementary information. The user can learn more about the suggested option if it seems interesting to them, or if they are not interested, they can ask for another suggestion. At any point, except during the survey, they can ask Ari free-form questions about treatment options or for other advice.
Paper Prototype
We developed a paper prototype using construction paper and sticky notes. The sticky notes were used as both user input and Ari’s messages to simulate an interaction between the user and the AI using the process flow before. The paper prototype was tested with participants before our digital prototype to gain quick insights so we could iterate and further critique the user interface.
Usability Testing
After outlining our interaction flow, and creating the paper prototype, we decided to test the usability of the design’s core feature set. Due to the constraints of the class, and the sensitive subject matter of our project, we decided to test our design with four students in our Master’s program. If we were given more time for testing, we would have liked to test the design with guidance counselors, since they frequently interact with our target users. The below points are the tasks we evaluated participants on:
Key Findings
Many users unsure how to learn more about suggested treatment options.
Many users unsure how to interact with the chatbot after completing the initial survey.
Some of Ari's vocabulary needs to be refined in order to use less stigmatizing language.
Visual affordance of bottom navigation icons didn't make sense to some users.
After finishing the initial survey, users wanted to explore the rest of the app to learn what they can do.
Some of Ari's vocabulary did not align with industry best practices.
Workflow & Interactions
Taking what we learned from the paper prototype findings, I designed both the high-fidelity mockups and prototype for Ari. Referring to color theory, the color blue was used throughout the app to evoke a sense of trust and expertise; two themes we found from our research to be very important to uphold. Additionally, Ari was given a face not only to represent brand identity, but to give the chatbot a more personal touch.
What I learned
Designing for a chatbot interface was an interesting experience, as it is not so much designing an interface as it is designing interactions. One of the key questions I had to keep asking myself was, “How will a user know what to do next?”. Without many things on screen to teach a user how to interact with Ari, Ari needed to be the one to engage the user and teach them the proper interactions. To accomplish this, Ari needed to consistently use language that was both personable and constructive. Overall, working on a chatbot design was a great experience, as it challenged me to think about design from a different perspective, and is something I hope to work on again in the future.