User research Interaction design Conversation design
UW Medicine
Cierra Sisters
As part of the design team, I completed analysis of literature review user interviews, worked closely with stakeholders to synthesize user requirements and iterate on messaging that resonated with the users the most.
CONTEXT
In 2020, Cierra Sisters, a breast cancer survivor and support organization, and UW Medicine partnered to address disparities in breast cancer mortality and screening among Black women. Through their early research, they discovered that compared to their white counterparts, Black women experience a higher mortality rate when they face breast cancer.
DISCOVERY
My contribution to this project began with a stakeholder meeting, where I gained initial insights into previous research, as well as their needs and expectations for the chatbot. Here are some of the key inputs they provided.
DESK RESEARCH
Right after the initial meeting, since I was new to this field, I used the available research papers to educate myself about the reasons Black women mistrust the system and don’t seek information about mammograms. Additionally, I tried to understand what past research has discovered about Behavioral Interventions methods and their efficacy.
In some communities, there's a reluctance to discuss personal health issues, especially those related to breasts and other intimate body parts.
Some Black women may hold inaccurate beliefs about the causes, symptoms, and treatability of breast cancer, leading to fatalistic attitudes or a sense of invulnerability.
Additionally, concerns about the cost of treatment, even with access to free or low-cost screening programs, can deter some women from participating.
Studies have shown that Black women are more likely to get mammograms when encouraged by family and friends, especially if they know someone affected by breast cancer.
Studies highlighted that information that highlight the effectiveness of mammograms in detecting cancer early and improving survival rates can encourage screening.
Interventions that resonate with Black women's cultural values, beliefs, and concerns has been shown to be more effective.
After consolidating these findings and sharing them with the team, from the two previously greed upon intervention methods, I decided to focus on the efficacy-based messaging to maximize the impact of the messages.
QUALITATIVE CODING AND ANALYSIS
Before I joined, the team had created both fear-based and efficacy-based messages and ran a co-design session with participants from the Ciera Sisters to get their ideas. However, they did not generate any insights or iterate further on them. I rewatched the co-design sessions and coded the qualitative data that the participants generated.
GUIDING PRINCIPLES
After incorporating reviewing the insights of the co-design session reviews, I crafted a set of design goals that guided my content-design decisions.
For the first iteration, I focused on the content of the messages and their information architecture. I tried to form the content in a way that proactively answered the subcategories mentioned above.
Even though the steps were clearly laid out, users got overwhelmed with the options and thought it was very robotic.
Users did not know whether they could ask more questions or they just had access to what was presented to them.
Even though users liked the idea of hearing why they’re getting the message, they did not like their data to be readily accessible in chatbot.
Following the first round of feedback, I developed multiple iterations for each component of the chatbot. To ensure the design process remained focused and aligned with user needs, I created a happy path diagram. This tool helped me stay on track and prioritize addressing the primary requirements of the users within the designs.
After another round of iterations, I once again presented the results to the team and Cierra Sisters representative to get their feedback and ideas for enhancement. Despite some improvements, three main issues came up consistently.
Answering in numbers, letters or a combination of both did not help users feel they’re conversing with an intuitive bot. In fact, it added to their confusion because they felt a wrong response type would take them completely out of the flow.
The number of combinations that days of the week, number of hours, and the number of locations would offer to the users were a lot. The linear chat conversation did not offer them an adequate speed in which felt comfortable to the users to navigate that.
To alleviate the issues above, I read Twilio’s documentation to understand what response types and features they offer and I came across their parsing ability. After confirming that UW Medicine’s engineering team were able to write simple scripts and create a database to utilize this feature, I used the happy path I created before to generate classifications and data mapping.
Since parsed words could fundamentally function as classifiers, emphasizing the classification of message types—along with keywords for certain subcategories—was crucial in showcasing the bot's capabilities.
When creating the diagrams, and sample databases, I realized that ever person may use different words and sentence structure to convey their message. Therefore, I recommended creating a separate database with all the entries that users entered, but did not receive a response from the chatbot.
IMPACT
After this version was shared with the Cierra Sisters co-design participants, user rated this revised version 12% higher in terms of its engagement, trustworthiness, and friendliness compared to earlier versions.
Both qualitative and quantitative data indicate that these improvements will foster stronger user trust, increase engagement rates, and create a more supportive and approachable experience for individuals navigating mammogram scheduling and breast health education.
