Yahoo categorizes users into five levels based on their monthly active days, hoping to encourage frequent platform visits. However, the current retention rate is trending downward, making its enhancement a continuous and long-term product goal. Generative AI can generate images and texts, could this technology potentially improve Yahoo's current retention rate issue?
Consequently, three data team interns and one user experience design intern, which is me, gathered together to experiment with new concepts.
Following the NDA, the content below only discloses information that can be publicly shared.
In the initial phase of this project, three aspects needed to be understood in advance: users, technology, products. Let see our findings!
The project's primary focus was on the platform's users, with the aim of engaging a broader audience. Utilizing backend data and desktop research reports, I considered both platform and current trends, findings and opportunities are...
(1) "User goals are centered around this shopping experience."
(2) "Users require an extension of their emotional and psychological experiences."
(3) "Users anticipate a sense of identity recognition."
The lost user groups, depending on their level of departure, are distributed across different loyalty tiers. Due to resource limitations, we couldn't interview users externally, so we only interviewed known platform users around us. Nevertheless, from both internal and external desktop research and coffee chats, we discovered a common trend – users tend to seek resonance online.
Capable of generating various contents like images, texts, and videos, we primarily used the OpenAI API as our main tool, creating content adaptable to data. The competitive analysis also revealed many AI applications, mostly designed to reduce costs, enhance efficiency, and generate buzz. Therefore, these will become key considerations in our subsequent design process.
On the other hand, we researched the data from Yahoo! shopping backend to discover some findings that might help us to build an new meaningful design. Due to NDA, I can not show more here, but you can contact me if you want to know more, here is my email: xu060327@gamil.com
I divided the competitive analysis into two parts. Firstly, I analyzed direct competitors – e-commerce platforms, focusing on their current retention strategies. Secondly, I examined content platforms with limited resources, concentrating on current content-related topics and mechanisms that promote retention. Additionally, I conducted a design walkthrough to understand the current platform's design focus and potential possibility that could enhance retention.
・If the benefits offered by a new feature do not align with user goals, it will be challenging to retain users long-term.
・Without sustainability, it would be no different from regular promotional activities and would fail to enhance retention rates over the long term.
・External Differences: Most e-commerce platforms present varying time intervals as shopping perspectives for users, such as limited-time sales, daily deals, monthly specials, etc.
・Internal Differences: This shopping store has moved beyond time-based perspectives to life scenarios as a new shopping viewpoint, although it still primarily focuses on real-world contexts.
Additionally, long-term retention symbolizes user loyalty and trust towards the e-commerce platform. Short-term retention, however, often mixes in elements of novelty and excitement. Therefore, gamification is a key concept in this project.
"How might we provide a new experience that fosters resonance among users, encouraging them to stay on the platform?"
"How might we provide a unique, non-realistic setting as a new shopping perspective for users, achieving a sustainable and differentiated experience?"
As the sole designer in the team, I advocated for the users while facilitating consensus between business and technology. Therefore, I organized workshops and actively guided the direction of discussions, ultimately achieving a consensus on the execution plan.
One for product labeling and the other for design output. Yahoo Shopping already has a well-established design system and identity, such as the tiger mascot, which needed to be integrated into the process. Finally, I conducted another brainstorming session and selected three proposals. I brought these ideas and paper wireframes to the meeting for discussion and evaluation with the team.
We brainstormed numerous keywords related to everyday life experiences for our initial test labels. However, we ultimately decided to adopt GPT, as manually conceived labels often resulted in inconsistent levels, and scaling up in the future would likely make it difficult to maintain uniform standards.
We began our ideation with 'happy moments' from daily life, using joy as the foundation to construct subsequent design concepts. Although the outcomes of the workshop were not adopted, as a designer, I gained many insights from the outputs.
Based on preliminary research findings, we established evaluation criteria for the three proposals – attractiveness, sustainability, and feasibility. Ultimately, Proposal Two was selected.
All three proposals were developed under the same space-time concept. The new design concept offers users an unreal shopping scenario, where the 'tiger', representing the platform, invites users into a new scene – a world constructed by Generative AI.
Tiger is researching users, but there is still some information that is uncertain and we need users to help us with clues. As a thank you to our users, we are providing our current research findings as well as coupons that our users may be able to use.
Tiger's research data was scattered during his travels, and the user helped Tiger to select the valid data (product). Finally, Tiger successfully connects to the "persona" and provides some behavioral data about your time on the planet.
Tiger travels interstellar, flying to new places at regular intervals to find the user's avatars. Tiger will send back information about the user's life (product) on that planet, so that the user can learn what products exist in his/her life on different planets.
Proposal Two, with its eye-catching appeal in the meeting and its substantial development potential and feasibility, was chosen.
I refined the final result based on suggestions from various stakeholders, as follows.
This design concept requires the use of data and AI, so my collaboration with data engineers was divided into two parts: images and text. The focus was on ensuring a logical tone in the communicated design concept.
I am primarily responsible for assessing whether the outcomes in terms of tone and content are aligned with Yahoo's brand image, as well as the design concept. The engineers, on the other hand, integrate current product tags and user behavior tags to generate personalized narratives specific to the user.
I took the lead. By inputting current labels into GPT, it generated prompts for MJ. These prompts were then formulated and handed over to engineers for experimentation. To save costs, we used SD for image generation. However, the choice of SD model is crucial to the output, so I assisted in finding the right style and left the processing to the engineers.
Within the limited time of the internship, it was challenging to produce a complete solution. However, I still endeavored to think comprehensively about the design concept, paving the way for future work.
The platform's backend has varying amounts of data for users of different activity levels, which could affect the content displayed. Therefore, I identified key data scenarios to be considered for future design references.
To ensure the design concept's long-term viability, it needs to include diverse and expandable elements. Thus, I treated the user experience process as a framework, applying different examples to demonstrate the concept's flexibility.
－The proposal provides the company with a potential application of generative AI. Although verification on GSM was not feasible, the proposal still underwent multiple iterations through design reviews and usability tests.
－The effectiveness of data and AI applications in this concept, such as the appropriateness of label definitions and the number of users it can cover, was also confirmed by the data team.
Celebration of rapid Release!
Our design proposal garnered support from both within and outside the department during the results presentation.
It was decided to first release it in the form of an EDM, with a partial unveiling of the concept during the 2023 Double 12 event.
Check the release version ↗︎Big thank you all!
Extend my gratitude to my colleagues, mentors, the design team, data team, marketing team, and everyone who offered guidance and advice throughout the process.
more reflection
Please reach out to me if you want to hear more regarding my work.
.gif)