ShopGoodwill

Problem Space

Our target users were first-time secondhand buyers aged 18–29, already fluent in Depop, Poshmark, and eBay patterns. Two structural issues on ShopGoodwill prevented users from browsing efficiently and saving reliably. The 200+ product categories sat in a flat vertical list with no hierarchy, so users had to already know where something was categorized to find it. Search often failed to return results because of exact-match spelling, which left the flat list as the only fallback. Separately, the Favorites heart icon and Favorites tab weren't placed where users look because these interactions failed to match user mental models of e-commerce platforms.

My Role & Constraints

Phase 1 — Structure, Not Labels

A card sort opened Phase 1 to test whether the existing category labels matched user mental models. It returned something bigger: participants flagged "Bulk" and "Miscellaneous" as placeholder categories they couldn't assign cards to, and independently proposed merges the existing hierarchy didn't allow. The taxonomy didn't match how users cluster these items. That reframed Phase 1 from a labeling exercise to a structural one.

No existing tree-test tool fit both ShopGoodwill's 200+ categories and a student budget — UXtweak capped at 15 responses, Optimal Workshop ran $199/month — so I built one that captured everything the paid platforms did. Two rounds of tree testing, with a site map revision between them, surfaced three problems in V1: Books under Entertainment hit a 41-second average time on task, Antiques and Collectibles were consistently confused, and the flat hierarchy made users hesitant at every level. Elevating Books to a top-level category was my call. V2 confirmed it.

Phase 2 — Testing Against a Competitor

Phase 1 ended with a restructured site map and a redesigned prototype that hadn't been user-tested. Phase 2 picked up that validation gap: could first-time users in the target demographic actually operate the redesign against a platform they already trusted? I sequenced three methods to answer — heuristic evaluation to pick the test flows, moderated testing against Depop to baseline the Favorites problem, and unmoderated testing on the redesign itself.

On the moderated side, participants saved items and located the Favorites Tab on both ShopGoodwill and Depop. ShopGoodwill took 23.77 seconds; Depop took 7.6. Same story on the Favorites Icon — 34.44 seconds vs. 18.14. Users took more than three times as long on ShopGoodwill as on a direct competitor.

I then designed and ran an unmoderated Maze study on the redesigned prototype, matching the moderated tasks so the before/after comparison would be clean. Navigation tested clean: participants located the Favorites Tab in 2.5 seconds, down from 23.77. All 8 participants located items through the restructured categories. The Favorites Icon result did not follow.

A Sampling Shift, Not a Design Failure

Favorites Icon time increased from 34.44 seconds to 68.66 in the unmoderated round — the wrong direction. All 8 participants added items to cart before looking for a save-to-favorites option. The moderated group had been recruited from users familiar with heart-based saving; the Maze pool skewed toward users whose mental model for "saving for later" was the cart. The redesign improved the experience for the heart-icon users but hadn't developed interactions for the cart-first users. The final iteration added "add to favorites" on the cart page — meeting users where they actually went.

Phase 2 — Redesigned Prototype

Favorites in the global nav, a clearer heart affordance on listings, and — added after the cart-first finding — a save-to-favorites affordance on the cart page, where the sampling-shifted group actually went.

Reflection

I'd build a more functional Phase 2 prototype before launching the unmoderated study. The non-functional cart page meant 8/8 participants hit a dead end I couldn't cleanly separate from genuine task failure.

I'd also screen unmoderated participants more specifically for e-commerce familiarity — the Maze pool skewed the results in a direction we could only interpret after the fact, a control I'd want in front of the study, not behind it.