// Case Study
SCOPE
Discovery architecture across multiple core surfaces
ROLE
Drove the diagnosis, strategic reframing, and architecture
TEAMS
Product · Design · Research · Engineering · L10n · Curation
Work shown is from my tenure at Pinkoi. To respect confidentiality, visuals and product screens were recreated, and metrics have been simplified or obfuscated. All perspectives are my own and do not reflect the views of Pinkoi.
CONTEXT
Curation, meeting scale
Since the 2019 brand refresh, Pinkoi had grown from a niche marketplace for Asia’s independent designers into a broader design platform, adding major brands such as Philips, Nintendo Switch, and Logitech. By 2024, it served 6.25 million members and more than 50,000 brands across 93 countries.
The Pinkoi app, originally designed for a curated selection, struggled to scale alongside this explosive growth in products and brands. An expanding range of products and unrelated experiments competed for attention in prime real estate, making exploration increasingly difficult. Endless, themeless product grids could no longer meet users’ expectations.
Pinkoi had become all product, no discovery.
The Challenge
Restoring wonder without breaking everything
My goal was to reclaim Pinkoi’s position as an “inspiration leader,” where people discover personalized items that inspire purchases. Browsing Pinkoi used to spark wonder. People would stumble on surprising products that made them stop and look. As inventory grew, creating those moments became much harder.
I needed to restore that sense of discovery without disrupting a fast-growing marketplace or rebuilding the product from scratch.
What needed to change:
Turn discovery from dull to delightful
Support every shopper in finding their own path
Highlight what makes Pinkoi unique
Discovery sat between opposing forces: business growth pushing for more inventory, user experience demanding clarity, and structural constraints limiting what we could surface. As inventory expanded under structural constraints, the discovery experience tilted, making it harder for buyers to find delight and for sellers to gain exposure.
Surgical strategy over full redesign
Rather than rebuild everything, I chose a precise approach that could compound while managing risk. The work needed to improve discovery in a live marketplace within the capacity of one buyer squad, with only brief support from a second team for a single initiative.
Instead of one large launch, I structured the effort as a series of focused initiatives that could build on one another over time.
First Signs
Listening to exit moments
"Nothing catches my eye" had become one of the most common complaints. The question was: where exactly was discovery breaking down?
What the feedback revealed
To understand why users were leaving, I dug into months of pop-up survey responses, focusing on open-ended feedback about browsing challenges and user workarounds. Three recurring issues emerged:
Repetitive content
Users kept seeing the same products and brands across sessions, making discovery feel stale.
Lack of personal relevance
Recommendations felt disconnected from users’ interests, leaving them wondering why those products were being shown.
Loss of inspiration
Users reported that browsing no longer sparked the curiosity and delight Pinkoi was known for.
The Discovery
Unique isn't unique anymore
The discovery pain points revealed a deeper problem: brand erosion.
The expectation–experience gap
Pinkoi had always “owned” unique. The ability to find customizable items, international designs, and surprising gifts was our strongest differentiator. But a gap had opened between what users expected and what they experienced:
☺︎ Users expected: "I come here for inspiration and unique gifts."
☹ Users encountered: repetition, irrelevance, and lack of inspiration
We researched how longtime members’ perceptions of Pinkoi had shifted:
Members still expected Pinkoi to be their go-to source for niche brands, customizable products, and novel finds. Yet they perceived more than half of its inventory as available elsewhere. Users increasingly saw Pinkoi as just another e-commerce platform.
When asked which words no longer described Pinkoi, 26% of members selected “unique,” a defining part of its brand identity.
The problem went beyond unsatisfying recommendations. Pinkoi was struggling to deliver its promise at the very start of the buyer journey, before casual browsing could become genuine interest.
Deeper Insights
Lost the magic
Previously, I had visualized the customer lifecycle as two critical loops, growth and retention, to help the team understand what sustained each. I used this framework to examine how the discovery problems were affecting both new visitors and returning buyers.
Both loops depended on discovery creating delight: giving new visitors a reason to explore, buyers a reason to return, and people something worth sharing. Bringing people to Pinkoi was not enough if the experience gave them little reason to continue.
Conceptual model: Buyer Growth & Retention Loops
The data showed where those loops were breaking down.
Discovery breakdown across all users
49% of direct visitors left without clicking a single item. Those who stayed averaged just 1.5 item views per session. People were arriving, but exploration was shallow.
The same pattern appeared among our highest-intent users, suggesting the problem went beyond attracting the right audience.
The retention freefall
Core buyers were declining while sleeping buyers, our inactive buyer segment, surged across all major markets. Our most engaged customers were drifting away, and we were struggling to reactivate those who had already disengaged.
The problem extended beyond first impressions. Even buyers who had previously found value in Pinkoi were losing the habit of returning. After years of building loyalty, we were watching it evaporate.
I synthesized these findings into a broader model of the risk: as buyers lost delight and sellers lost exposure, Pinkoi risked losing the distinctive sellers and products that gave people a reason to return.
Conceptual model: a downward cycle linking discovery, seller exposure, and brand erosion.
Homepage: losing prime real estate
The homepage showed how little of the discovery experience was reaching users. Only 20–30% of visitors scrolled beyond the fold, while click-through rates were just 1% even in prominent placements such as hero banners and featured topics.
The homepage had become a battleground for competing content, making it harder for users to find a clear path into discovery.
The real retention driver
The biggest insight came from analyzing what brought users back. More than discounts, promotions, or product variety alone, it was freshness: the expectation of novelty and surprise.
People returned when they trusted they would discover something new. This connected the retention problem to the brand promise gap: Pinkoi was losing not just its distinctive positioning, but the expectation that each visit would offer something worth finding.
When people could no longer count on that spark of surprise, they lost a reason to open the app.
Reframing the Problem
A single pathway traps different user types
Finding what was broken wasn't enough. I needed to understand why it stayed broken, which meant looking at internal dynamics, not just user behavior.
The hidden bottleneck
Stakeholder interviews revealed why the problem persisted. Teams treated the first fold as the only reliable path to exposure, creating a zero-sum battle for slots.
Even with data showing banners had limited impact, the belief in homepage dominance kept driving short-term decisions. Features that should have worked together competed instead, fragmenting the experience and confusing users.
The app reinforced this bottleneck. Only one of five tabs was accessible without signing in. Visitors who tried to explore beyond Home hit an authentication wall, while teams competed for space on the one surface everyone could access.
Users who explored through multiple functions had conversion rates 2.8 times as high as those who did not, yet only 17% of users explored this way. Broader exploration was associated with higher conversion, but for most users, effective exploration was hard to achieve.
The hidden bottleneck
We couldn’t just optimize within existing constraints. Mediating between teams might have eased the competition for slots, but it would not have changed the structure that created it.
The question shifted from “How do we fix discovery?” to “How might we give every user a path into rediscovering what made Pinkoi special?”
I reframed the opportunity around multiple discovery paths, giving teams room to contribute and users ways to explore that matched their intent.
This became Inspiration Compass ↓
Solution
Inspiration Compass
Inspiration Compass replaced single-entry discovery with a multi-path architecture, shifting from passive product display toward content-led commerce.
We opened discovery from one accessible tab to three purposeful pathways.
Home became the welcoming front door, combining themed storytelling with a streamlined Categories hub for direct browsing. A “Japanese Minimalism” story could connect ceramics and textiles into one narrative, giving teams a cohesive way to present products and users a theme to explore.
Discover became the inspiration hub, a magazine-style editorial space where curated stories met contextual shopping. Users could explore design philosophies, lifestyle inspiration, and the stories behind unique products.
Feed deepened personalization, with recommendations shaped by users’ browsing patterns. Returning users could pick up where they left off.
Goal-oriented shoppers could go straight to search. Those browsing for inspiration could explore categories in Home or editorial stories in Discover, while returning users could follow personalized recommendations in Feed. Each tab connected to the others, creating pathways rather than dead ends.
From product grids to contextual storytelling
Within these pathways, we replaced endless product grids with stories that gave products context and activated purchase intent.
Discover was the centerpiece. Editorial stories like “a cozy haven for cat lovers” presented lifestyle scenes instead of isolated products, helping users picture how products could fit into their lives.
We extended this approach into Shop the Look, which let users explore tagged products within a complete aesthetic.
A product tile shows an item. An editorial story shows a life it fits into.
Curation that learns
Supporting these experiences was infrastructure designed to scale personalization while preserving Pinkoi’s curated identity. At its core was a style taxonomy that captured cultural nuance and enabled recommendations across aesthetics.
Hybrid curation blended algorithmic scale with curatorial judgment:
AI clustered products by style and suggested groupings.
Curators refined the vocabulary and validated choices.
User behavior closed the loop.
The result was personalization that felt both smart and tasteful.
Guest experiences in Feed let visitors sample what personalized recommendations would feel like, which reduced friction and built early trust.
Style-based discovery was phase one. We built the infrastructure for future expansion into occasion-based and mood-based discovery, though those experiences were not completed before I left Pinkoi.
Shop the Look: explore tagged products within a complete aesthetic.
Results
Measurable impact across the system
Ecosystem impact
Adoption varied at first, but over three months, the experiences began reinforcing one another:
Home: Cohesive, theme-based storytelling through topic cards increased click-through rate 9.5×
Discover: Once friction was reduced, users browsed an average of 12 items per session, an 8× increase that showed editorial curation sparked meaningful exploration.
Feed: Gamified onboarding lifted engagement by over 300% and doubled preference completion, giving the personalization system better data to work with.
Users who explored Discover also returned more frequently, with revisit rates climbing from 15% to 43%. This created a reinforcing loop: users entered through inspiration, explored more deeply, and returned with clearer preferences, strengthening both engagement and personalization.
Optimization insights
Discover’s initial adoption revealed an entry-point problem: only 2% of users clicked into editorial stories. Once inside, however, engagement was strong: 34–37% clicked through to product pages, and users browsed an average of 12 items per session.
The opportunity was to make those stories easier to discover. We ran quick tests with localized story titles, reduced image height, tag hints, and weekly notifications. Within weeks, product page visits from Discover rose 10%, and adoption increased from 19% to 26%.
User validation
User feedback echoed the patterns we saw in the metrics:
"I keep coming back to the Discover tab. There's always something I didn't expect to love."
“Love the update! I finally feel excited to open the app again, it’s refreshing.”
"After completing the onboarding tasks, my feed felt spot-on, like it really gets me."
"It's nice to see more small brands again. It feels more like the Pinkoi I remember."
Users were rediscovering why they loved Pinkoi.
External validation
Apple featured Pinkoi as App of the Day across Taiwan, Hong Kong, Macau, and Japan, spotlighting Discover. App Store exposure increased by 84%.
Brand recovery and looking forward
Users formed new habits. They returned to Discover, completed preference settings in Feed, and browsed more deeply into categories.
The infrastructure (adaptive curation, style taxonomy, content-commerce integration) would improve over time as more users explored and expressed preferences. Personalization would get sharper and discovery more relevant.
Pinkoi began restoring its position as an inspiration-led marketplace, with infrastructure built for sustained growth.
👀 Behind the Strategy
Turn discovery from dull to delightful for every shopper
I started with three core questions:
What makes discovery delightful?
How do you design for every shopper?
What contexts shape discovery?
I mapped discovery’s broader context to understand how user behaviors, content systems, and organizational dynamics shaped the experience.
This map showed that discovery problems weren't isolated UX issues but symptoms of deeper structural tensions between what users wanted, what the business needed, and what the product could support.
Turning discovery gaps into design questions
I translated these tensions into four “How might we” questions to guide my strategy:
Together, these questions formed a strategic arc:
Bring forward uniqueness → Support discovery paths → Activate desire → Build lasting trust
#1 Noisy → Tailored
Turning noise into tailored journeys
Navigation scarcity was strangling discovery.
Only 20–30% of homepage visitors scrolled, and CTR in prime homepage real estate was just 1%. Authentication walls blocked access to 4 of 5 tabs, stopping 50–80% of users before they could build momentum.
Think of it like a narrow river during a flood. When water volume exceeds capacity, it overflows. But when users hit navigation constraints, they didn’t overflow into exploration. They just left.
Why one path couldn’t work
Our user research had identified two primary discovery behaviors:
Goal-oriented shoppers arrived with specific purchase intent and expected efficient ways to narrow down to the right product. Their typical journey: Search → Product → Similar Products → Shop. They avoided the homepage, finding its recommendations irrelevant and repetitive.
Window shoppers came to browse and be inspired, expecting Pinkoi to surface novelty and surprise. Their typical journey: Homepage → Categories ↔ Products → Similar Products. Some browsed homepage content first, while others jumped directly to categories.
Within these two groups, we observed four distinct subtypes:
Targeted seekers
Came with a precise product in mind and used search almost exclusively.
Refiners
Started with a vague idea and relied on progressively narrowing results.
Taste-driven browsers
Revisited favorite categories, styles, or brands, collecting inspiration for later.
Casual wanderers
Browsed widely with low intent, and shifted themes easily while seeking surprise.
Our single-entry architecture couldn't serve these four different behaviors. Discovery needed to adapt to how people actually explored.
Creating opportunity, not managing scarcity
The architecture concentrated competing content and user needs into a single entry point, making discovery feel noisy, repetitive, and irrelevant.
My solution centered on a 2+2 navigation structure:
2 tabs always open to visitors (Home & Discover), showcasing Pinkoi's uniqueness and value upfront, even before login.
2 tabs powered by user data (Feed & Wishlist), with guest experiences such as recommendation samples and onboarding so visitors weren’t faced with an authentication wall. Feed was labeled “For You” in the interface.
The Me tab remained account-only for profile and utility functions.
While Wishlist was part of the long-term 2+2 vision, our first step expanded visitor access from one entry point to three tabs: Home, Discover, and Feed.
This framework created opportunity instead of managing scarcity: purposeful spaces for teams and personalized pathways for users.
The 3-tab architecture
I developed the navigation strategy and worked side by side with our product designer and PM through brainstorms and reviews to define each tab’s role and expand visitor access.
Home: Redesigned for promotional campaigns and category-driven window shoppers. It balanced business objectives with user-friendly storytelling, giving teams purposeful campaign space while preserving browsing flow.
Discover: Created to showcase Pinkoi’s design marketplace identity. Magazine-style editorial stories sparked inspiration for window shoppers and gave the curation team a dedicated space to tell those stories.
Feed: Enhanced personalization for returning users. AI-driven recommendations were paired with guest-accessible sample content that demonstrated value before sign-up, building early trust. Many visitors continued with a lightweight preference step to fine-tune their Feed.
Early preference cues (“Pass” / “Love it”) helped users shape their Feed with minimal effort.
Home opened the door, Discover sparked inspiration, and Feed deepened engagement. Each tab worked independently while contributing to a more connected discovery experience.
Bringing uniqueness back to the surface
Underneath the homepage battle was a fight over identity.
Random product grids made Pinkoi feel like any other e-commerce site, burying the unique designers and customizable items that defined us.
Our creative goal became clear: elevate design stories that showcase Pinkoi's heart.
I drove the decision to create a new homepage structure and partnered with our product designer and PM on the redesign, while aligning business teams and securing executive buy-in.
This structure brought forward what's truly one-of-a-kind and supported different discovery paths through two complementary entry points:
Campaigns landing: Shifted from scattered product placements to cohesive, theme-based storytelling. A “Sustainable Living” theme, for example, could bring bamboo kitchenware and organic textiles together rather than present them as isolated listings.
Categories hub: Created a direct entry point for window shoppers who preferred to browse by interest, replacing endless homepage scrolling with clear category pathways.
To support this, we introduced new curation guidance for business teams, changing the question from "What products need promotion?" to "What story showcases our unique marketplace?"
The new homepage structure gave competing business needs clearer places to live while bringing Pinkoi’s identity back into focus. Instead of blending into generic e-commerce, Pinkoi once again felt like a curated design destination.
#2 Passive → Activated
What activates desire?
Multiple discovery pathways solved the navigation bottleneck. But users still needed content that sparked desire, not just entry points.
The harder question: what turns passive scrolling into "I need this!" moments?
We worked with researchers and data scientists to study inspiration patterns. The findings were stark: 67% of Pinkoi users showed no behavioral signs of inspiration. Of all users, 51% bounced immediately, while another 16% showed minimal or unfocused engagement.
Our behavioral analysis identified five distinct inspiration states, each representing a different level of engagement and focus:
G0 — Non-Starters
Barely engaged with any products, rarely progressed beyond landing
G1 — Uninspired Users
Browsed scattered, unrelated products without focus
G2 — Inspiration Builders
Gradually narrowed to similar products, showing emerging focus as inspiration crystallizes
G3 — Inspired Shoppers
Consistently viewed related products with strong intent
G4 — Intent Continuers
Returned to saved items and similar products, closest to purchase
These inspiration states were associated with conversion. As inspiration began forming in G2–G4, users started adding products to wishlists and carts, with most purchases happening in G4. G0–G1 users left quickly, often within 10 pageviews, with little engagement.
The inspiration formula
From these findings, we developed a framework with three ingredients:
Inspiration = Context Setting + Keywords + Uniqueness
Each ingredient played a distinct role:
Context Setting helped users imagine the product in their lives.
Keywords gave vague desires concrete language users could search for.
Uniqueness elevated perceived value beyond commodity alternatives.
For struggling users in G0–G1, Context Setting and Keywords were critical to helping inspiration form. For engaged users in G3–G4, Uniqueness helped turn interest into purchase intent.
Product tiles alone could show users what was available. These ingredients helped explain why a product might matter to them.
Why Discover could spark "I need this!" moments
The Inspiration Formula gave us a rationale for pairing editorial storytelling with contextual shopping: help users imagine a possibility, find language for it, and recognize what made the products distinctive.
This clarified Discover’s role in the three-tab ecosystem.
Home remained the campaign front door for promotions and first-time visitors, while Feed deepened personalization for returning users. Discover would activate inspiration through magazine-style stories, editorial narratives, and modules like Shop the Look.
Editorial content delivered all three inspiration ingredients. We built contextual shopping into Discover, showing products within lifestyle scenes or complementary arrangements so users could imagine how they fit into their own lives:
Context Setting through lifestyle stories: Instead of showing isolated products, editorial content created scenarios such as “evening wind-down routines,” helping users visualize products within a desired lifestyle.
Keywords through narrative flow: Editorial stories gave users language for interests they struggled to articulate. A story about “workspace sanctuaries,” for example, could introduce terms such as “minimalist desk accessories” or “productivity rituals.”
Uniqueness through curation: Editorial selection signaled that a product deserved attention. Stories highlighted the design details and qualities that distinguished it from commodity alternatives.
Discover wasn’t just another tab. It completed the ecosystem by giving our curation team the space to activate desire through storytelling, turning passive browsing into motivated exploration.
#3 Defensive → Adaptive
Teaching the system taste
The Inspiration Formula gave us a way to activate desire, but curation alone couldn’t scale delight. Manual editorial judgment showcased Pinkoi’s identity, but couldn’t adapt to millions of individual tastes or rapidly expanding inventory.
Without adaptability, we risked losing our differentiation again.
We needed a discovery system that could learn without losing what made Pinkoi special. That required a shared aesthetic vocabulary, expressive enough for human nuance and structured enough for machine learning. Categories and attributes described what a product was, not why someone might love it.
I recognized style as that vocabulary. It expressed Pinkoi’s design identity and enabled personalization across product categories. This became the foundation for an adaptive curation system that blended algorithmic scale with curatorial taste.
Style foundation & hybrid curation workflow
Defining style systematically was harder than naming aesthetics we intuitively recognized. Many products sat between aesthetics, expressed multiple styles, or had no clear stylistic anchor. Style is inherently fluid.
Some styles, like “Streetwear,” worked in one domain but didn’t translate to others. Creating separate style systems for each category would have simplified classification, but added operational complexity and risked confusing users.
We focused on Fashion and Home & Living, where our research indicated style mattered most to user decisions and choice overload was highest. These categories also had the visual diversity and product volume we needed.
Unlike concrete attributes such as color or material, style carries cultural, aesthetic, and emotional meaning. Wabi-sabi, for example, requires curatorial interpretation, not just visual pattern recognition.
This required both human insight and system logic.
I drove a hybrid curation workflow with three components:
AI for scale: Clustering models scanned thousands of products and generated preliminary aesthetic groupings for curator review.
Human for context: Curators refined these groups by clarifying terminology, resolving overlaps, and aligning styles with Pinkoi’s brand.
Feedback as learning: Low-confidence classifications were flagged for human review. Curator corrections fed back into model refinement, while reviewing the groupings helped the team calibrate its own judgments.
The curation team and I created an initial set of 53 style classes and human-labeled datasets for each. After multiple rounds of calibration and validation, we evaluated each style using two dimensions: Breadth of representation and Accuracy of fit.
Through this human–machine calibration, we refined the taxonomy from 53 to 37 styles. Before launching to users, we confirmed each style had 1,000+ products, enough for exploration without losing coherence.
This taxonomy became the foundation for style-based discovery.
Bringing style to users: the MLP approach
With the taxonomy ready, the question was: what was the smallest user-facing expression that could demonstrate value quickly?
Style tags became our Minimum Lovable Product (MLP): a simple pilot that let users experience style-driven discovery. They also offered a way to support gift-finding, one of Pinkoi’s highest-value use cases.
Early user behavior validated the approach: users explored more deeply, responded to stylistic framing, and recognized style as a meaningful lens for navigating Pinkoi's inventory.
This informed a phased roadmap:
Phase 1: Style-based discovery through Shop by Style.
Planned phase 2: Occasion-based discovery through Shop by Occasion.
Future vision: Mood-, aesthetic-, and persona-based discovery.
Style-based discovery was the first expression of the system. We built the infrastructure for further expansion, but occasion-based and mood-based experiences were not completed before I left Pinkoi in 2024.
We didn’t have to choose between brand consistency and user relevance. We built a foundation designed to strengthen both.
A final note
I led the vision, and it came to life through teammates across product, design, curation, engineering, research, and data. Grateful to have worked with teams who cared as much about getting it right as I did.
