Product Design
of an AI Application
Overview
I joined Abzu as its second designer with two main responsibilities, help design their Reason application which was in its early stages of development, and help with the mentoring and growth of the design team.
Role
Senior Product
Designer
UX Research
UX Strategy
UX/UI Design
User Testing
Time
Feb - Aug
2023
6 Month project
The Reason App
Reason app was an artificial intelligence tool aimed at supporting scientists with their research, in particular with the statistical analysis tasks.
The way it worked was that a user (research scientist) uploaded their data and asked a research question. The AI analyzed it and recommended the best series of steps to come up with an answer to the research question.
This tool was quite interesting as commonly research scientists are not experts in statistics and usually need help from bioinformaticians to complete their research.
Initial Strategy
I started with research, first a UI Audit, whenever I am new to a project I like to start with a UI audit as it allows me to explore the application get familiar with all the different pages and features and at the same time I’m fining easy-to-fix problems with usability, typography, alignments or consistency. These are low-hanging fruit items that the team can fix fast and already start making the application better.
The second part of my research was to map the main user flows, I want to understand what steps is the user taking to complete their main goals in the application.
Then I move into understanding more about the user, to properly design for someone I need to understand them, so my plan as a new person in the team is to gather the people with the most knowledge on our users and get them to tell me everything they know about the users, but also everything they don't know and all of the assumptions they have made.
Finally, I want to validate all of the knowledge I have gained, so I prepare to talk to the real users, via surveys and interviews with a focus on understanding who they are, what are their needs, what are their pains, and how they like to work. This way we can learn if our assumptions are true.
UI Audit
As expected the initial UI Audit helped me understand more about the application but it also discovered problems with the:
Accessibility
Navigation (Architecture information)
Alignments
Layout
Consistency
Missing guidance
Information Architecture
During the AI Audit, I like to map the architecture of the application to understand where everything sits next to each other, this usually brings forth some problems. In this case, I could see there are a lot of layers between the homepage and an analysis. I expect the user will spend most of their time on the analysis. So I could already start thinking about reducing this distance.
User Flow Analysis
After mapping the main user flows I get the same feeling as from the IA, there are too many steps before reaching the main action of the product, creating an analysis.
User Assumptions workshop
To understand more about the users but also what the team knew I organized a small workshop. I facilitated the workshop and invited two software engineers, the CEO, a product owner, and Marta my UX design colleague. The goal was for me to understand the users, but also to get everyone on the team to have the same idea about the users, and lastly, I wanted to get a list of assumptions that we needed to test to make sure our product was serving our users in the right way.
So I asked everyone to individually list the assumptions they had about our users, and as a group, we mapped them into the matrix below, where the most uncertain and impactful assumptions made it to the top of our testing list.
Impact / Risk to product success
High
High
Low
Uncertainty
TO TEST FIRST
To test later if needed
Validating the assumptions
To get the best information we planned to get some quantitative and some qualitative methods to validate our assumptions:
Quantitative:
Design a survey and aim to get a high number of replies.
Target Abzu contacts who fit the user persona
Offer a free trial of the platform once it is ready.
Qualitative:
Design an interview aimed at getting more in-depth knowledge of our users.
Invite Abzu’s contacts who fit the user persona (5 to 10)
Offer a free trial of the platform once it is ready
Findings
Even though the survey answers were not as numerous as we wanted we were luckier with the interviews where we managed to have 6 in-depth interviews and together they brought a lot of good light into our user persona:
Users do have a hard time with statistical analysis and often recur to bioinformaticians for help.
Users work in teams but do most of the work themselves and use the rest of the team for reviews.
Users' main pain point is in data gathering and preparation.
Users' data is private so they need to trust the tool before uploading anything.
Users are intrigued by the potential of working with AI but also slightly skeptical.
Ideation Strategy
I proposed to focus on the analysis part of the application, this is what will give our users the most value, other areas of the application like file management sections or user settings are not as impactful and have been resolved many times by multiple design teams. This way we focus on innovation and what brings value.
The plan was to start by defining new user flows, and then build those up to low-fidelity mockups that we could test internally with our team. After the first round of feedback refine the mockups into the first iteration of an interactive prototype that we can start testing with our potential users.
User Flow
User goal or problem
User action
What is the user trying to accomplish? What is the problem to be solved?
What actions are they currently taking? How are they doing it? What information are they seeking?
The first change to the original application was bringing the analysis as close as possible to the login so that the users could start working and thus gain value from our application as fast as possible.
Low Fidelity Mockups
The home page was redesigned to follow the user flow, now the main actions were to be front and center, and an additional sidebar was added to ease the navigation and file management.
The first mockup here starts to show an idea of how the step-by-step workflow of the analysis will guide our users in finding the answer to their research questions.
Feedback + Ideation Workshop
I gathered the same participants as in the assumptions workshop and presented them with the new designs. My goal with this workshop was first to get some initial feedback on the designs but I also wanted to explore their heads for new ideas so I asked them to sketch a few ideas for how they imagined the analysis could go.
First Iteration Prototype
After applying the feedback and synthesizing some of the new ideas, we arrived at our prototype.
This prototype was now ready to be tested with our potential users, who had agreed from the surveys and interviews to continue in contact with us and help us with tests of this solution.
Iterative loop
The project followed this loop for a few cycles, every time improving in usability, simplicity, and overall value to our users.
Final Result
After several rounds of feedback, the application was ready and we were working with the development team to get it ready for our users. In the last version, we were following the feedback that users wanted to get faster into the answer to their question and review the steps afterward, instead of following a step-by-step approach.
The pivot ending
In parallel with our design and development, there was an effort to recruit users, 500+ were already on an early access list.
However, at the same time, we entered into a partnership to develop a demo project for a big European Airport. To use AI to improve the prediction of the taxi times of airplanes.
It was decided that this would be more beneficial to the goal of getting another round of funding for Abzu so the Reason project was dropped and we focus on this project next.