Fairer Gig-Driving
Designing a web-app / website to understand the gig-drivers’ opinions and create applicable models using machine learning.
My Role
Working with another designer and a team of developers in this project, my role was to:
My Roles:
UX Researcher
UX Designer
Motion Designer
2019 Summer - Fall
Minkyung Lee
Rachel Lee
Nathan Jen
Brent Hong
Hemank Lamba
Michael Becker
Interviews with gig-drivers to learn about their understanding matching algorithm.
The user flow of the website experience with regards to process of machine learning.
Part of user interface, illustrations, and motion graphics to reduce mental & cognitive load for the participants.
There are millions of people working for Uber, Lyft, Doordash, etc, as gig-drivers. For many of them, it is their full-time job.

These gig-driving systems rely on artificial intelligence to connect the drivers to the passengers. Most people assume that it is based on distance, while it’s actually not, and many drivers are dissatisfied with their experience regarding the matching algorithm.
Gig drivers often think the matching algorithm was unfair, but find it difficult to propose practical change.
With my teammates, I went through online threads and interviewed 4 gig drivers to learn about their understanding around the algorithm.

We learned that there were many complaints for moments where the drivers don’t understand why they got a certain match, or didn’t get one.

Despite their frustration, it’s difficult for them to suggest applicable ways to improve these systems due to the algorithm being opaque as well as many drivers’ not having much knowledge about how it works.
Designing the tool
We wanted to create a tool that can help the drivers better communicate their thoughts about making a fairer matching algorithm.
We weren’t working with a company to provide us details about their algorithm, so we initially made an assumption: There are more factors that get taken into account when within a radius from the passenger.

(Later, we learned this was correct by talking with a driver who has done a study with Lyft.)

With this in mind, we designed a process where the user creates a fair algorithm by customizing the variables which algorithm considers for making a match.
The user is given a set of variables and can choose or add variables they think the matching algorithm should take into account.
Then the user has to go through a series A/B testing in scenarios created according to the variables they chose.
Finally, a customized matching algorithm is born! It learned from the choices of the user how to weigh different variables to reach a conclusion.
User Testing
"Hmmm... the process can be a little more exciting."
A teammate and I conducted user testing with paper prototypes & rough versions of the website. Going through this process with participant drivers as well as ourselves, we noticed two major findings.

Through user testing, I saw an opportunity to incorporate visual assets in this process to enhance understanding and make the process feel less tedious for the participants.
“Can you say that again?”
The concept of algorithm and this tool was not so easy for the participants to grasp quickly just through verbal explanation.
Too many tests...
After going through the first few A/B testing scenarios, the participants made decisions regarding mainly 1-3 variables out of many they chose initially.
Enhancing the understandability and making the experience more joyful using motion and illustrations.
I illustrated visuals to explain the concepts of the process and added animations for loading time, where previously no information or just heavy text were shown. I also added icons to lessen the cognitive load to read heavy text all the time, but associate visual information with the variables.
Limitations of another team can be an opportunity for intervention.
Close collaboration with developers took place during this project for deciding the order of events and interactions. I found it helpful to learn what is capable and reasonable on the engineers’ side in making design decisions because it narrowed down the broad number of choices designers had to make.