How We Fixed Product Discovery With Tree Testing and Card Sorting
In a previous post, we detailed the course and results of a UX audit and cognitive walkthrough for our main persona, Anna. The expert study identified three core problems:
- Product categorization: Many categories were unclear, with items potentially fitting into 2 or 3 different categories.
- Filtering: In many categories, filters were too general and inadequate for finding specific products in a long list.
- Search: The search engine was poorly optimized, particularly for synonyms.
This article covers how we resolved the categorization problem using tree testing and card sorting, two standard UX research methods that gave us hard data on where users got lost.
Analyzing the Menu Structure
We started with a detailed analysis of all categories and subcategories in the store. We identified categories that were particularly unclear, where products could be assigned to multiple categories or lacked a corresponding subcategory.
The underlying issue was that the store's information architecture reflected warehouse logic, not customer mental models. Baby food could be in "Food" or "Mom and Baby." Hair brushes were in "Beauty Accessories" instead of "Hair." Children's toothpaste could be in "Hygiene" or "Mom and Baby." These are not edge cases. These are everyday products that a parent buys regularly.
We selected the most problematic categories and subjected them to tree testing.
Tree Testing: Measuring Where Users Get Lost
We replicated the store's menu tree structure and prepared a list of 35 problematic subcategories. These were tested with 100 users through a survey using a tree testing tool. Respondents were tasked with finding specific items in the menu structure (e.g., "Find hybrid nail polish").
The testing provided data on three key metrics per subcategory:
- Time to find: How long users took to locate the item
- Success rate: Whether they found it at all
- Backtracking rate: Whether they entered a category, realized the item was not there, and had to go back
The results for all 35 subcategories were below the store's average, validating our hypothesis of suboptimal category matching. Users frequently backtracked, meaning they entered a category expecting to find a product but realized it was placed elsewhere. They often pointed out incorrect subcategories, and the search for items took longer than average.
This pattern is a textbook example of Hick's Law in action: when two categories seem equally valid for a product, decision time increases and users either guess wrong or give up entirely. Each backtrack is a moment where the user can abandon the task. Over a shopping list of 6 items, those abandoned moments compound into a smaller cart. We documented this effect in detail in our UX audit case study.
Card Sorting: Rebuilding the Menu Around User Expectations
To rebuild the menu according to how users actually think, we conducted a card sorting study. We prepared a list of main categories and asked respondents to assign subcategories to the ones they considered most appropriate.
Card sorting removes the design team's internal bias. The team that built the store knows where everything is. They navigate by memory, not by logic. External users navigate by expectation, and those expectations often diverge significantly from internal organization.
The study gave us a data driven map of where users expect each product type to live. We used this to redesign the store's information architecture from scratch.
Validating the Redesign With a Second Round of Tree Testing
To confirm that the redesigned menu actually improved navigation, we ran a second round of tree testing on the new structure. Same 35 subcategories, same task format, fresh respondents.
This before/after approach is critical. Without it, you are guessing whether changes helped. With it, you have direct comparison data.
The improvements were clear:
- Users found items 2x faster on average
- Backtracking dropped by 29%
- Users were significantly more likely to select the correct category on the first attempt
These are not marginal gains. Finding products twice as fast with 29% fewer wrong turns means users spend less time navigating and more time adding items to the cart. For an e-commerce store where heatmap data showed that navigation friction was the primary driver of smaller cart sizes, this fix addressed the root cause directly.
What Product Teams Should Take From This
Tree testing before card sorting. Tree testing tells you what is broken. Card sorting tells you how to fix it. Running card sorting without first identifying the specific problem categories wastes respondent time on categories that already work.
100 users is enough. Tree testing does not require thousands of participants. 100 respondents gave us statistically reliable data across all 35 subcategories. The signal was clear even at that sample size.
Test the fix, not just the problem. The second round of tree testing on the redesigned menu was what proved the changes worked. Many teams stop after identifying problems. The validation step is what separates an opinion from evidence.
Internal teams cannot test their own navigation. The store's team was surprised by the results. Categories that seemed obvious to them confused 100 external users. This is the single strongest argument for external UX research: you cannot see your own blind spots.
Related Case Studies
UX Audit: Why Customers Buy Less Online covers the full audit that identified navigation as the core problem.
How Heatmap Analysis Increased E-Commerce Sales by 33% details the cart visibility and feedback improvements that came from the same project.
Improving Online Shopping for Persona Anna documents the persona development and cognitive walkthrough that preceded this tree testing study.
Radiant Capital Hack Explained: $4.5 Million is an unrelated but technical deep dive by our blockchain developer Krzysztof Cywiński.
