The End

I designed a carousel-based experience that scaled to support 65,000 themes, all personalized and ranked for each buyer across desktop and mobile.

Impact

Jul–Sep 2025

The Beginning

May 2024. I joined a new squad focused on Etsy’s homepage. Engagement analysis showed that discovery was a weak spot, so we started there.

In our first experiment, we removed distracting, redundant content, increasing overall engagement by 1.1%.

Next, we added new modules to expose buyers to more shopping ideas. The modules were dynamic, diverse, and fresh, but they weren’t personalized, and engagement was flat.

The Spark

The first spark of promise for discovery content was personalized themes.

The combination of engaging content, personalization, and design drove a 1% lift in homepage clicks.

The first themes dataset included titles but no descriptions, so I used Gemini to generate on-brand descriptions at scale.

The insight

Personalized themes were engaging, but the volume of content was too low to serve the broad interests of Etsy’s buyers.

The opportunity

We had a new, LLM-scaled dataset with 8,000 themes, matched and ranked for each buyer.

My challenge

The existing homepage design, with a mix of independent modules, couldn’t scale to support the new dataset. I needed to design an all-new approach.

Building confidence

In addition to the prototype, which we used in user research, I did a UI audit.

I compared my proposed approach with best-in-class ecommerce sites, other two-sided marketplaces, and streaming platforms.

I highlighted UI and UX patterns to prompt discussion with my team and stakeholders and gain internal support for my design.

Experiment results

The first personalized carousel experiment exceeded our expectations:

We blended unpersonalized content—like daily deals and gift ideas—with six personalized themes. Themes had a 2.5× higher CTR.

Overall, the new experience drove more than $9.4M in sales, giving us a strong foundation to build on.

The End, Again

The carousel-based experience scaled to support 65,000 themes, personalized and ranked for each buyer across desktop and mobile.

As we learned how people used the new experience at scale, we improved the design, content, and data, driving more than $11.7M in additional sales.

The problem

Etsy had search, category, and market pages, but each was built around a query or a category. None could show a curated, personalized group of items that cuts across categories, like a shopping recommendation or theme.

Key challenge

The pages had to work on web and in Etsy’s apps, and serve as destinations for push and email campaigns. My PM and I worked with our app and marketing counterparts to define the requirements and deliver them.

My role

I defined the content strategy, which set what appears where and in what order, and designed a flexible landing page system to carry it.

The Beginning

I partnered with two designers: a senior staff designer working across buyer experience and a staff designer working on discovery on iOS.

Together, we defined cross-platform experience principles along with a shared definition of discovery concepts, like 'speed bumps' that would help people explore content, and co-designed components, like related themes.

Eng collaboration

Since we were creating new pages, it was important to align on technical requirements and constraints with Engineering partners.

I created a feature priority guide to drive early conversations and maximize time for technical spikes and development work. Those conversations helped shape the design of the page the scope of our first experiment.

And when it came time to build the pages, I wrote CSS to solve tricky edge cases for new modules.

Experiments

In the first experiment with the new page, we 49% CTR on the listing grid and many people were scrolling to the bottom of the page. However, most people weren't engaging with speed bumps, which included related themes, searches, and items.

For our v2 experiment, I recommended removing the speed bumps and adding endless scrolling. Those changes increased CTR over 148% (to 62%) and people were 2.5× more likely to see all of the results on a given discovery page.

TK

Content for this slide: how the foundational design grew to support more types of content: we started with active missions and themes, then expanded to similar items and merch-curated content

Etsy: theme landing pages

As we developed personalized themes, we realized landing pages were a weak spot in the experience—there were no pages on Etsy that supported browsing themes.

I defined the strategy for the landing pages and designed a flexible system that expanded beyond themes to support active shopping missions, similar items, and campaign landing pages for email and merchandising.

Our first experiment showed strong engagement, with a 42% CTR, so we launched the page to all users.

Project reflection

What I’d do differently

Test the full alternative. We tested the old content as carousels and several designs of a single themes module, but not multiple themes modules in the old layout. I can show the carousel worked, but not that it was the only way to get there. Our constraint was a holiday code freeze, so we decided to launch and learn from real usage, knowing we were giving up that comparison.

What I learned

A preview and the page behind it have to be designed as one path. In one experiment, a denser module got 61% more buyers to click in, but landing page click-through fell 13% and fewer buyers reached a listing. Carousels fixed the preview by showing what each theme held, and deeper landing pages rewarded the click.

Mobile Web Experience

TK: bringing themes to mobile web…

Progression: Carousels → stacked carousels → 2×6 grid → 2×2 grid

What was next

TK: sparkle-based design, codepen demo of card-builder

My PM rotation

After the first personalized carousel experiment, my PM focused on expanding the dataset from 8,000 themes to 65,000. To support that effort, I signed up for a PM rotation.

The focus of the rotation was developing and launching Home Hub, a feature that condensed recent activity into a compact module at the top of the homepage.

I wrote a full PRD, defined measurement goals with our analyst, planned experiment scope and deliverables with the lead engineer, and analyzed the experiment results.