case study 03

The page that didn't move.

Card listing pages for a top-five US card issuer Sole designer Live A/B test Shipped
Problem
Fifteen cards per page, all with identical visual weight. People didn't pick wrong; they didn't pick at all.
What I did
Redesigned filtering, card tiles and product differentiation across four page templates.
Result
+14% session yield — the largest absolute gain in the round. Won on three of four templates.
What it proved
Business shoppers are a different population, and one page not moving is worth more than the average that hides it.
Before and after — full category page at identical scale img/cat-hero-before-after.png

The crowding was the problem being solved. Scannability has to be visible at thumbnail size or the case study's central claim goes unillustrated.

Shell page. Structure, masthead and at-a-glance are final. Body copy is in the markdown draft; screens and the items below are still needed. Delete this block when the page is written.

  • What “clearer card tiles” actually meant — what you added, removed, or reweighted
  • Whether the business page was ever picked back up
  • Whether the completion drop was ever looked at
  • Duration and team shape
  • Screens: before/after full page, filtering states, tile detail, business page beside a winner

The failure mode isn't picking wrong. It's not picking.

Category pages are the middle of the funnel. Someone has narrowed from “I want a credit card” to “I want a travel card” and now faces fifteen products that look approximately the same: a rectangle, a name, an APR range, and a bulleted list written by a compliance team.

+14% session yield, and the largest absolute gain in the round

Three weeks live. Application starts per session rose about 17%, and the redesign came back with a 100% probability of being the best option. It won clearly on three of the four URLs it shipped to.

Before / after — full category page
The crowding was the problem being solved — scannability has to be visible at thumbnail size.
Filtering — resting, open, applied, reset MOTION · img/cat-filtering.mp4 + .webm
One of three changes, and the only one a still can't carry. Four-second silent loop.
Card tile — before and after, detail crop img/cat-tile-detail.png
Where differentiation was won or lost.

On the fourth page it did nothing, and I wrote that up instead of burying it

The business card page came back flat — about a point down, inside noise. The average across all four still looked great. I could have reported the +14% and moved on; nobody would have gone looking.

Two separate tests both saying business behaves differently isn't noise. It's an unscoped user segment.

Business page beside a winning page img/cat-business-vs-winner.png
What's structurally different about the one that didn't move.

The win leaked at a specific place, and it wasn't underwriting

Application starts rose about 17%, but the share of starts that became completed applications fell by roughly three points. Approval rate on the applications that did complete held steady.

So underwriting wasn't the constraint. The application form was.

The full scope

All four templates at small scale img/cat-four-templates.png
Four surfaces, not one. Three moved and one didn't.

What I'd flag

The limitations I'd raise before someone else did.

No research ran ahead of this, and it cost me the explanation.
Three changes shipped together, so I know the bundle won and I can't tell you which one earned it. The win is real; the learning is thinner than the number suggests.
One file held three initiatives.
This work, the navigation redesign and an early product-page pass all lived in the same Figma file — and ended up with three different verdicts.

Next: Clarity beat novelty. Twice. — a starter card recreate where preference research and live behaviour finally agreed.

Partnership work under NDA. Screens are abstracted and metrics are relative. What's real here → · Privacy