Exploring the future of the Redbubble uploader
The problem
Redbubble's uploader is a primary functionality for artists, but it had grown fragmented and inefficient over time. Artists were forced to manually configure dozens of products, rely on inaccurate previews, and repeatedly fix placement issues, resulting in poor-quality listings, frustration, and costly downstream customer refunds.
From a business perspective, a well-configured listing was worth 186% more than a default one, making upload quality a meaningful growth lever.
Research
By synthesising prior research and new interviews, I identified that artists didn't actually want more control. In interviews, artists consistently reused defaults, skipped over advanced controls, or described configuration as "busywork" that they had to push through in order to publish their listings. They were asking for less effort and more trust in the outcome. This reframed our initial understanding of the problem from adding configuration tools to automating decisions wherever possible.
I translated this research into a small set of design principles that guided all subsequent exploration:
- Design for non-experts, and educate in context
- Treat uploading as part of the creative flow
- Reduce cognitive load through simplified decisions
- Automate wherever possible to unlock creativity
- Build confidence through accurate, trustworthy previews
These principles helped me evaluate concepts consistently and avoid incremental fixes that didn't meaningfully reduce effort or increase confidence.
Ideation and experimentation
Because this experience affected every artist on the platform, I explored a wide range of solutions and interaction models. Over the course of the project, I created and tested dozens of concepts across different levels of automation, sequencing, and control to understand where confidence broke down.
Product fit assessor
Rather than relying on artists to judge product suitability manually, I worked with data science to define a scalable product-fit assessor that matched images to appropriate products, and blocked unsuitable combinations. While this removed some artist control, this was an intentional choice in favour of marketplace quality and trust. This shifted quality control upstream, reduced artist effort, and prevented poor customer experiences before they even occurred.
Auto-placement
I also worked with data science to create an auto-placement algorithm that took into account required sizes, the edges of the image, and the best size for the image to be on each product. Designing this required balancing visual integrity with scalability across nearly 100 product types. As a result, uploaded images were placed in the best position for each product without cropping any important parts, meaning that artists don't need to change anything and every product looks great.
Results
As the work progressed, it became clear that delivering the uploader at the quality bar we were aiming for required deeper platform investment and migration work. When company priorities shifted, the initiative was paused before full rollout.
Despite this, the project:
- Established a shared, measurable understanding of what "high-quality listings" mean at scale
- Introduced automated fit and placement concepts adopted in later creator tooling
- Delivered a validated MVP used to guide future uploader investment
For me, this work reinforced that meaningful experience improvements often come from shifting responsibility from users to systems, especially in complex workflows. We cannot effectively solve user problems through UI tweaks alone, but rather, need to address underlying technical issues and constraints.