Tree Testing · Case Study
How tree testing improved an e-commerce admin's navigation
By UXbeam, information architecture tools and services since 2021 · Updated October 7, 2026
Three navigation versions. Nine everyday seller tasks. Direct completion rose from 30% to 48%, and the paths showed what to improve between tests.
An online seller relies on their admin to run their business end to end: shipping orders, updating listings, checking stock, reaching customers and managing promotions. These are not occasional errands. They are the recurring work of keeping a store operating.
If an inventory alert is hard to find, a seller might miss the chance to restock before the next order. If the review-request settings are buried, a valuable customer follow-up might never be enabled. Even when a feature exists, a menu that makes people guess where to look delays the job they came to finish.
UXbeam tested three versions of a seller-admin navigation with participants. Their clicks revealed confusing names, functions filed where participants did not expect them, and two nearly identical review-request destinations. Each comparison gave the team a more specific question to answer and a more deliberate next version to test.
The results at a glance
| Initial | Iteration 1 | Iteration 2 | |
|---|---|---|---|
| Overall Direct | 30% | 33% | 48% |
| Completed sessions | 59 | 67 | 38 |
| Direct attempts | 144 / 477 | 187 / 568 | 158 / 328 |
Direct means reaching an accepted destination without backtracking. UXbeam also records Indirect (the right destination after a detour) and Elsewhere (finishing at a different destination).
Share of all attempts ending Direct. Iteration 2 ran later and tested the revised tree together with several revised task instructions.
Inside the seller admin
The tested prototype was a seller-side admin console, the application merchants use to operate their stores. It brought orders and shipping, listings, inventory, customer communications, promotions and sales analytics into one navigation. The study examined whether people could reach those functions before the menu structure was carried into an interface.
Its feature inventory drew on four established seller platforms: Shopify, Etsy Seller, WooCommerce and Ecwid. Together, they show the range and reach of the tools this kind of admin must organize, from shipping a purchase to maintaining a catalog or communicating with buyers.
ShopifyMillions of businesses
Etsy Seller5.6M active sellers
WooCommerce4M+ stores
EcwidHundreds of thousands of merchants
Consider automated review requests. After an order, a seller might schedule an email asking a buyer to review a purchase; WooCommerce documents this workflow. Reviews can matter to sales: Spiegel Research Center reported purchase likelihood 270% higher for a product with five reviews than one with none in its research. A seller still has to locate the feature before they can use it.
Where would you look for that setting: Customer communications, Orders, or Sales analytics? All three might sound plausible until you see how people actually navigate the menu.
Where does this live?
These menu labels carry real consequences. Jakob Nielsen called navigation labels “the most costly words in the company” when discussing employee intranets. In a seller admin, an unclear label can interrupt an order, delay a stock check or send someone hunting for the customer-message settings. Tree testing reveals those wrong turns while the hierarchy is still inexpensive to revise.
Where tree testing fits
What participants were asked to do
Nine tasks represented a seller's workweek: pausing notifications, sharing listings, finding shipping information, sending automated review requests, checking low stock, sending loyalty coupons, rebuilding a store website, opening another store, and adding a product video. Participants received one job at a time and navigated only the labels and hierarchy. Their first clicks, backtracks and final destinations showed where expectations lined up with the structure and where they did not.
Across the three versions, the tested hierarchies ranged from roughly 60 to 80 navigation entries. Each version had its own fixed tree. Task summaries and results appear in Study details and full results below.
01 · The baseline exposed several different problems
The Initial tree recorded 30% Direct overall. Some routes were considerably harder than others:
- Pause notifications: 27% Direct. The setting sat under Customer management center, but participants' first clicks split across that category, Add new and the orders section.
- Check low inventory: 12% Direct. The alert was harder to locate than its everyday purpose suggested.
- Send a loyalty coupon: 2% Direct. Many participants looked under communications, while the answer was filed under analytics.
- Set up an automated review email: 9% Direct. Participants often ended at a plausible-looking destination that the test did not accept.
The team had more than a low overall score. It had the beginnings of a diagnosis: one problem looked like unclear language, another like unexpected placement, and another like two destinations competing to describe the same function.
02 · The first revision made notifications easier to find
Iteration 1 replaced Customer management center with Customer Notifications, split the combined orders-and-sales area, and reorganized review-email settings. The Initial and Iteration 1 versions were tested in parallel with separate participant groups. The task Pause notifications for the weekend kept the same wording in both versions.
Customer management center
Notification settings
Pause
Customer Notifications
Notification settings
Pause
Notifications
Notification settings
Pause
The iterations tested three names for the same path.
| Notifications task | Initial | Iteration 1 |
|---|---|---|
| First click on the intended category | 34% | 57% |
| Direct completion | 27% (16/59) | 46% (31/67) |
The old category made participants interpret an organizational phrase before they could find the setting. The new label named the function more directly. More participants chose that branch first, and more completed the task without backtracking.
Initial first clicks split three ways (n=59):
Iteration 1 38 of 67 first clicks on the renamed menu:
Iteration 2 33 of 38 first clicks on Notifications:
The rename concentrated first clicks: the intended menu drew 34% initially, 57% in Iteration 1 and 87% in Iteration 2.
Another task, sharing a listing, rose from 21% to 38% Direct. Rebuilding the store website fell from 51% to 39% Direct in Iteration 1. Across all nine tasks, Direct completion moved from 30% to 33%. That modest overall change concealed useful progress on specific routes, together with other problems the next iteration still needed to address.
03 · What the wrong destinations revealed
Automated review requests provided the most revealing example. In the Initial tree, two neighboring destinations appeared under the same menu:
- Auto send review request
- Auto Sent review requests
Only the second was accepted by the test. Yet 26 of 54 participants (48%) chose the first. Overall, 47 of 54 (87%) finished Elsewhere on that task.
Those clicks pointed to a very specific problem. Participants had not simply wandered around the tree; almost half had converged on the same alternative label. The team could resolve the duplicate destinations and correct the accepted-answer setup instead of endlessly renaming unrelated categories.
Only "Auto Sent review requests" was marked correct. Its near-twin, one line away in the same menu, collected almost half the sample.
The remaining low-performing tasks told different stories. Inventory was still at 14% Direct in Iteration 1 after a move that placed the alert deeper than many participants expected. The loyalty-coupon task reached just 6% Direct. First clicks favored communications over analytics. These were clues about where to reconsider the route, not merely which words to polish.
04 · The second revision brought more seller tasks within reach
Iteration 2 simplified the top-level menus to names such as Notifications, Orders, My stores and Sales analytics. Inventory alerts moved into the revised analytics structure. The review-request destination participants had repeatedly selected was recognized as an accepted answer. The research team also refined some task instructions as it learned how to test the seller jobs more clearly.
Overall Direct completion reached 48% in this tested version, up from 30% in the Initial test. The per-task results make that progress more tangible:
| Seller task | Initial | Iteration 1 | Iteration 2 |
|---|---|---|---|
| Pause notifications | 27% | 46% | 82% |
| Find low inventory | 12% | 14% | 41% |
| Send a loyalty coupon | 2% | 6% | 54% |
| Share a listing | 21% | 38% | 38% |
On the notifications task, 33 of 38 participants began directly in the Notifications menu. The inventory route, barely improved by the first revision, also became easier to reach in the later tested setup. The coupon result rose in a version that refined both its location and the way the task was worded.
Direct rate per task. Filled dots: Initial and Iteration 1. Dashed hot-pink: Iteration 2, which revised both the tree and several task instructions.
Not every problem was solved. The automated review-email task stood at 22% Direct in Iteration 2, and finding a post office at 30%. Those paths gave the team specific work for the next revision, while the easier-to-find notification and inventory functions showed what should be preserved.
05 · How to use the patterns in your own navigation
Across the three versions, the team could make increasingly specific decisions: rename a menu that people missed, relocate a feature they kept looking for elsewhere, reconcile two nearly identical destinations, and refine task instructions that did not capture the intended job clearly enough.
How to read the pattern
Overall Direct completion rose from 30% to 48% across the tested versions. More importantly for the next design decision, the path records identified which labels and placements helped participants reach a tool and which routes still sent them elsewhere.
From tested routes to interface design
A tree is a stripped-down navigation prototype. Once its important routes have been tested, designers can carry those labels and groupings into a sitemap, wireframes and the interface itself. That gives a product team something stronger than a workshop guess: observed routes for the jobs that matter, plus a list of the functions still likely to slow people down.
UXbeam records separate results for each tested version, including Direct, Indirect and Elsewhere outcomes, first clicks, destinations, full navigation paths and participant-level detail. Researchers can see what changed and where people still struggled without losing the earlier evidence.
Make everyday store work easier
When an order needs shipping or stock needs checking, the seller should not have to hunt for the right screen. Test those important jobs with participants before the navigation is built into the interface. UXbeam shows where they start, turn back and finish, so your team can resolve obstacles before launch.
Study details and full results
The prototype drew on seller functions from Shopify, Etsy Seller, WooCommerce and Ecwid. Three navigation versions were tested against nine seller jobs with US-based participants recruited through a paid online panel. Internal dry runs were excluded. Initial and Iteration 1 were fielded during April 2025 with separate participant groups; Iteration 2 was tested in September 2025. Between Initial and Iteration 1, task wording was adjusted for review emails, coupons and tutorials, and accepted destinations changed for review emails and product videos. Iteration 2 also refined several task instructions; per-task counts differ where participants did not complete every task.
See all nine tasks
| Task | Initial Direct | Iteration 1 Direct | Iteration 2 Direct |
|---|---|---|---|
| Pause notifications for the weekend | 27% (16/59) | 46% (31/67) | 82% (31/38) |
| Share a new listing on Facebook | 21% (12/56) | 38% (25/65) | 38% (14/37) |
| Find the closest post office to ship a new order | 44% (24/55) | 39% (25/64) | 30% (11/37) |
| Set up an automated email asking buyers to review their order | 9% (5/54) | 30% (19/63) | 22% (8/37) |
| Find which items are running low in inventory | 12% (6/52) | 14% (9/63) | 41% (15/37) |
| Send a coupon to the most loyal buyers | 2% (1/52) | 6% (4/63) | 54% (20/37) |
| Rebuild a dated store website | 51% (26/51) | 39% (24/61) | 54% (19/35) |
| Open a second store to sell tutorials | 33% (16/49) | 30% (18/61) | 43% (15/35) |
| Add a product video to a listing | 78% (38/49) | 52% (32/61) | 71% (25/35) |
For the method behind the study, see How tree testing works, Writing tree-test tasks, and Avoiding leading tasks.
Sources
- Shopify, full-year 2025 results: millions of businesses use Shopify.
- Etsy, full-year 2025 results: 5.6 million active sellers at year-end 2025.
- WooCommerce: 4 million+ stores built on WooCommerce.
- Ecwid by Lightspeed: hundreds of thousands of merchants across 175 countries.
- WooCommerce documentation, “Remind customers to leave a review”: an example automated post-purchase review request.
- Medill Spiegel Research Center, “How Online Reviews Influence Sales”: the 270% purchase-likelihood comparison for products with five reviews versus none.
- Jakob Nielsen, “Intranet Information Architecture (IA) Methods”: the cost of ambiguous navigation labels.
Study evidence: UXbeam's version-level tree-test records and the results presented above.