Tree testing an e-commerce admin: relabel or restructure?
By UXbeam, information architecture tools and services since 2021 · Updated August 9, 2026
One rename raised Direct success 19 points. The broader revision moved overall Direct 3 points.
Feature-dense products develop the same navigation problem: features with several plausible homes. Should automated review requests live with customer communications, with orders, or with analytics? UXbeam tree tested three versions of a seller-side e-commerce admin, Initial, Iteration 1, and Iteration 2, against the same nine tasks.
Three versions of one seller admin: trees of 60 to 80 items, nine tasks.
59, 67, and 38 panel sessions across the three versions.
Renaming one menu lifted its task from 27% to 46% Direct. The full revision moved overall Direct only from 30% to 33%. Iteration 2 (directional) reached 48%.
Where tree testing fits
The setup: a seller admin based on Shopify, Etsy Seller, WooCommerce and Ecwid Android apps
The tested product is a seller-side admin: the screens a seller uses to run a store, not the storefront a buyer sees. It covers orders and shipping, listings, inventory, customer communications, promotions, and sales analytics, in trees of 60 to 80 items. The feature set drew from four seller apps, each serving sellers counted in the hundreds of thousands or millions.
ShopifyMillions of businesses
Etsy Seller5.6M active sellers
WooCommerce4M+ stores
EcwidHundreds of thousands of merchants
One of those jobs is collecting product reviews. After a buyer receives an order, the store can automatically email them asking for a reviewWooCommerceWooCommerce DocumentationRemind customers to leave a review. The reviews it gathers shape how future shoppers judge a product: the first few displayed reviews measurably raise purchase likelihoodSpiegelSpiegel Research Center, NorthwesternHow Online Reviews Influence Sales. A seller who cannot find where to turn that on never reaches any of it.
Where does this live?
Admin findability is a daily cost. Sellers arrive with a job in mind: ship the order, check stock, answer a buyer. The work repeats, so every menu that makes someone stop and decode adds up; Nielsen Norman Group's intranet research calls navigation labels "the most costly words in the company"NN/gNielsen Norman GroupIntranet Information Architecture (IA) Methods. A tree test measures that cost before the structure ships.
Nine tasks, all from a seller's normal week:
- Pause notifications for the weekend.
- Share a new listing on Facebook.
- Find the closest post office to ship a new order.
- Set up an automated email asking buyers to review their order.
- Find which items are running low in inventory.
- Send a coupon to the most loyal buyers.
- Rebuild a dated store website.
- Open a second store to sell tutorials.
- Add a product video to a listing.
Participants saw only the labels and hierarchy, one task at a time. Each attempt ends Direct (right place, no backtracking), Indirect (right place after a detour), or Elsewhere (a confident finish somewhere wrong). The method is covered in tree testing, end to end; for phrasing tasks like these, see the Job-Shaped Task.
Iteration 1 renamed, split, and rebuilt three areas
Customer management center became Customer Notifications. The combined Manage orders/sales section split in two. The review-email area was rebuilt inside the renamed communications section. Seven positions left the tree, twelve arrived.
Customer management center
Notification settings
Pause
Customer Notifications
Notification settings
Pause
Notifications
Notification settings
Pause
The path to the same setting in each version. Only the menu name changes.
1Finding
Iteration 1 moved overall Direct only 3 points
Initial and Iteration 1 ran head-to-head on the same nine tasks: 59 and 67 panel sessions.
Overall Direct: 30% on Initial (144 of 477), 33% on Iteration 1 (187 of 568).
Three points. The average hides one large win, several flat tasks, one scoring artifact (method note), and two problems the revision left untouched. Tree testing separates changing a lot from changing the right thing.
Share of all attempts ending Direct. Iteration 2 ran later with some tasks reworded; read it as directional.
2Finding
One rename raised Direct success from 27% to 46%
The task read the same in both versions: pause notifications for the weekend. Only the tree changed.
On Initial, the setting lived under Customer management center; first clicks split three ways, and 27% ended Direct (16 of 59). On Iteration 1 it sat under Customer Notifications; 38 of 67 clicked it first, and Direct reached 46% (31 of 67).
The click pattern gives a strong clue: people scan for the word they already have in mind. Customer management center offered nothing to match. Customer Notifications did.
Initial first clicks split three ways:
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. Task wording was identical in Initial and Iteration 1.
3Finding
Two tasks stayed below 15%
Two tasks barely moved, and neither looks like a label problem.
The coupon task: 2% Direct on Initial (1 of 52), 6% on Iteration 1 (4 of 63). Both versions filed it under sales analytics; most first clicks went to communications, because sending a coupon reads as a message to buyers.
The inventory task crept from 12% (6 of 52) to 14% (9 of 63); Iteration 1 buried the alert one level deeper than most people searched.
The review-email task did worst: 9% Direct on Initial (5 of 54), 87% Elsewhere (47 of 54). No version cleared about 30% on it, and Iteration 1's gain partly reflects a scoring change (method note).
4Finding
48% chose the same alternative destination
Elsewhere is the most useful outcome in a tree test, because failures concentrate.
The initial tree held two nearly identical labels in the same menu: Auto send review request and Auto Sent review requests. Only the second counted as correct. 26 of 54 sessions (48%) finished at the first.
They were not lost; they picked the label the tree should probably keep. A concentration like that names the fix: merge the twins, or accept the one people choose. The share-listing task showed the same pull in miniature.
Only "Auto Sent review requests" was marked correct. Its near-twin, one line away in the same menu, collected almost half the sample.
5Retest
Iteration 2 reached 48% Direct
Iteration 2 restructured around the behavior: top-level menus took the plain names people kept matching (Notifications, Orders, My stores, Sales analytics), inventory alerts moved under Sales analytics, and the review label people kept choosing, Auto send review request, was marked correct.
Across 38 panel sessions, overall Direct reached 48% (158 of 328). Read it as directional: a smaller test, with several tasks reworded, so its numbers are not a controlled comparison.
Notifications reached 82% Direct (31 of 38), with 33 of 38 first clicks on the menu. Inventory reached 41%. Coupon jumped to 54%, but its wording changed too, so part of that jump belongs to the task. Review email stayed unresolved.
Percent Direct per task. Filled dots: Initial and Iteration 1. Dashed hot-pink: Iteration 2, directional. † scoring or task wording changed (method note). ‡ label echoes the task wording; see "When to relabel, when to restructure". * task reworded in Iteration 2.
When to relabel, when to restructure
Four things, each visible in the data above.
Relabel when the label is the problem. The notifications feature sat in an acceptable place behind a name nobody matched. One rename moved first clicks from a three-way split to a majority, and the task from 27% to 46%.
Restructure when people keep choosing a different home. The coupon and review tasks stayed weak through Initial and Iteration 1, with the same destination pattern each time: people looking under communications, and finishing at the unmarked review label. The cleanest structural win is inventory: stuck at 12% and 14% until Iteration 2 moved it under Sales analytics and it reached 41%. For coupon and review, Iteration 2 is directional evidence, and review has still not cleared 30%.
Read first clicks and destinations before trusting a rate. The three-point overall move was the net of a 19-point win, several flat tasks, and a scoring artifact. First clicks showed why the win happened; destinations showed what to fix next. A top-line rate alone carries neither.
Treat your best number with suspicion. Iteration 2's winning label, Notifications, appears in the task instruction itself. UXbeam's leading-term check flags exactly this overlap, because people can match a word without navigating by meaning. We report the 82%; on our own results screen it would carry a flag. Spotting and rewriting that overlap is covered in leading tasks in tree testing.
Relabel or restructure?
What UXbeam captures for every iteration
UXbeam is a tree testing and information architecture tool for testing navigation with participants and analyzing first clicks, paths and destinations. Each version in this study kept its own result set: task outcomes (Direct, Indirect, Elsewhere), first clicks, final destinations, complete navigation paths, and participant-level results. An iteration carries the tree and tasks forward, so the next test is set up in minutes rather than rebuilt. The charts on this page are that output.
Method note
- Participants were US-based adults recruited through a paid online research panel.
- Internal dry-runs were excluded from the reported counts.
- Counts are panel sessions per version: 59 (Initial), 67 (Iteration 1), 38 (Iteration 2). A small number of panel IDs appeared more than once within or across versions, so no participant total is reported.
- Per-task n varies because some participants stopped before finishing; each figure states its own n.
- Duplicate submissions of the same task within a session were removed before reporting; the earliest submission was kept.
- Initial and Iteration 1 were tested concurrently in April 2025; Iteration 2 in September 2025.
- Task wording was identical between Initial and Iteration 1 except for small edits to the review-email, coupon, and tutorials tasks. Several instructions were reworded for Iteration 2, so its task-level numbers are directional.
- Iteration 1 changed which destinations counted as correct for the review-email task (part of its gain is definitional) and removed one correct destination for the product-video task, so that task's apparent drop is partly a scoring change; it is excluded from improvement claims.
- Rates are observed percentages of completed attempts. No statistical-significance claims are made.
- This is UXbeam's own research on a client product, run to the same standard we recommend to teams.
See all nine tasks
| Task | Initial Direct | Iter 1 Direct | Iter 2 Direct | Note |
|---|---|---|---|---|
| 1 · Pause notifications | 27% (16/59) | 46% (31/67) | 82% (31/38) | Identical wording in Initial and Iteration 1. Iteration 2's label echoes the task word; see the caution above. |
| 2 · Share a listing | 21% (12/56) | 38% (25/65) | 38% (14/37) | Steady pull toward the listing screen in every structure. |
| 3 · Post office for shipping | 44% (24/55) | 39% (25/64) | 30% (11/37) | Task reworded in Iteration 2. |
| 4 · Automated review email | 9% (5/54) | 30% (19/63) | 22% (8/37) | Iteration 1 changed which destinations counted as correct; part of the gain is definitional. Never cleared about 30%. |
| 5 · Low inventory | 12% (6/52) | 14% (9/63) | 41% (15/37) | Moved under Sales analytics in Iteration 2. |
| 6 · Coupon for loyal buyers | 2% (1/52) | 6% (4/63) | 54% (20/37) | Task reworded in Iteration 2; part of the jump belongs to the task. |
| 7 · Rebuild store website | 51% (26/51) | 39% (24/61) | 54% (19/35) | Task lightly reworded in Iteration 2. |
| 8 · Store for tutorials | 33% (16/49) | 30% (18/61) | 43% (15/35) | Task lightly reworded in Iteration 2. Wrong finishes cluster on a plausible sibling category. |
| 9 · Add product video | 78% (38/49) | 52% (32/61) | 71% (25/35) | Iteration 1 removed one correct destination, so the drop is partly a scoring change; excluded from improvement claims. |
Percent Direct per task; n varies by task. All Iteration 2 figures are directional (38 sessions, several tasks reworded).
Sources
- Shopify, Q4 2025 financial results announcement.
Millions of businesses using Shopify. - Etsy, Inc., Q4/FY2025 results press release.
5.6M active sellers as of December 31, 2025. - WooCommerce, woocommerce.com.
4M+ online stores built with WooCommerce. - Ecwid by Lightspeed, Ecwid 101.
Hundreds of thousands of merchants in 175 countries. - WooCommerce Documentation, "Remind customers to leave a review".
Automated review-request workflow. - Jakob Nielsen, "Intranet Information Architecture (IA) Methods", Nielsen Norman Group, 2007.
Navigation labels and findability. - Spiegel Research Center, Northwestern University, "How Online Reviews Influence Sales", 2017.
Online reviews and purchase likelihood.
Study data: UXbeam tree-testing sessions and version history shown on this page.