Published October 3, 2026 · Audits run August 13–14, 2026
The store
The Quest (thequest.ch) is a Swiss online marketplace for premium multisport and endurance gear. Collections of dozens of products — 38 in Shoes alone — a cart drawer, a checkout: the typical journey where small defects hide in plain sight because the team knows the store too well to stumble on them.
The audit
Tower ran a full walk of the store on its production infrastructure: an ephemeral machine, a walker agent browsing as a persona, then three analysts rereading the walk — one for trust and reassurance, one for wayfinding, one for the checkout funnel. Walkers judge nothing; analysts cannot return to the site. Agents stopped before payment, as they always do.
What the agents found
Checkout funnel · Wayfinding
The cart page returned a 404. After a successful add-to-cart, the mini-cart's button led to /cart — a page that did not exist. It was found independently by two analysts with separate mandates, reading the walk along two different paths. A defect like this costs orders every day it stays live.
Trust · Reassurance
A promise contradicted one click later. "Free delivery and returns" was shown under the add-to-cart button, and contradicted by the cart drawer that opened on the same page.
Wayfinding
A breadcrumb pointing nowhere. The breadcrumb named a category absent from the main menu, leaving shoppers no way back to it.
Product content
"No image" on more than 20 of 38 thumbnails in the Shoes collection — more than half the collection shown without a product photo.
Each finding arrived with its screenshot, an observation quoting the page byte-exact, a severity and an estimated conversion impact — ready to promote into a user story.
Were the findings real?
We ran a second audit on the same pages and submitted every defect from both runs — 11 in total — to the store's owner. All 11 were judged true. No false positive on this sample.
The interesting part was elsewhere: the two runs overlapped on only 43% of their findings. Since none of the findings were false, that gap was not noise — it was coverage. Each run saw real defects the other missed.
How much one audit sees
Two independent audits of the same pages are two samples of the same population of defects, which can be measured with a capture-recapture estimate:
Two caveats, both against us. The sample is small — 7 analyst defects observed, so treat these as orders of magnitude, not decimals. And the two runs used the same model, prompts and pages, so their misses are probably correlated: the true detection rate per run is likely lower than 65%, not higher. We publish the optimistic number with its caveat rather than a rounder one without it.
That measurement shaped the product: Tower runs on a schedule, deduplicates findings from one run to the next, and never declares a defect fixed just because one run stayed silent about it.
What this means for your store
- Your team cannot see these defects — not for lack of skill, but because they know the store. Agents browse it like a first-time shopper, every time.
- The costliest defects are often the dullest. A broken cart link is not a design question; it is a ticket. See how Tower turns findings into tickets on the e-commerce UX audit page.
- One audit is a start, not a verdict. Run it on a schedule. Read more about the process on e-commerce conversion rate optimization.