2024

Trailblazing the first consumer-facing Genarative AI feature for Uber

Overview
This AI assistant is designed to help users pinpoint their exact wants, spark inspiration, narrow down decisions and simplify the process of grocery shopping and more. This feature is unique to Uber Eats, as it allows users to express their desires and needs directly through their own words.
My role
Responsible for the end to end product design for Beta release collaborating with content and conversational designers. Ran a in-depth study with 16 pilot users to get further insights for next steps.
Team
Design lead (me)
Product Director
1 Content Designer
1 UX Researchers
2 Data Scientists
8 Backend, Mobile, QA Engineers
Problem statement
Today, finding and ordering food and items can be a tedious and time-consuming task.

- Mid-intent users have specific and often complex needs (dietary preference, price range, brand loyalty) and no way to search for something that meets all of them.
‍
- Low-intent users have decision paralysis. The feed is personalized but can’t narrow down recommendations enough to help users over this hurdle.
‍
- Grocery shopping often starts with a high-level job-to-be-done more complex than just ordering a meal, e.g. throwing a birthday party..
Research Insights
Longing for "Just For Me" and Wishing for Easier Interactions

High personalization need

Users want very personalized recommendations based on past behavior and history. "It should only recommend vegetarian options knowing I only ordered from vegan places"

Prompting is hard...

GPT-like experiences require high effort. Users want help to shape prompts and would only engage with AI when it's a complex task that regular Eat's flow can not help.

Decision-making is harder

The most typed organic query types are open-ended “What should I eat” and “Get me deals”. 

Final Designs
Here is what we tested with 5% of all US IOS public users
Discovery & FTUX
Three entry points—near search on home, on browse, and on the search zero-state —tie the AI assistant to search to set that expectation.
Inspirational zero state jump-starts the conversation
Get ideas: Discover personalized options
Find deals: Search food discounts and offers
FInd things fast: Search multiple things at once
Reorder favorites:  Effortlessly reorder your favorite dish with pre-filled customizations
Help users articulate the wants and needs
Take advantage of of the chat interface to ask follow-up questions when needed.
The assistant takes into account: Unique food characteristics, Order history, Time of day, Weather
Complex searches
To make the AI assistant additive to a user’s experience, we’ve focused on complex searches not possible elsewhere.
Failing gracefully and learning fast
We started by asking, “What can go wrong?” and drafted 20+ deny list responses. These cover:
Safety issues, including allergies and food intolerances
Misuse and abuse of the technology
Unsupported features
Impact & next step
The valuable learnings of the launch led to fruitful new tracks of works
After the public launch, around 8% of users would tap into the AI Assistant albeit almost non-existing marketing of the feature. Interesting data and queries were captured daily that kept broadening our eyes for this new powerful technology. To fully leverage this learning opportunity, I ran another UXR of 16 in-depth interviews with 50/50 retained and churned users. All insights pointed towards a need for seamless integration with search.
Design system
Meticulously contributing to the company's AI UI design system
Self Service Procurement