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 2.0 changed structurally that 1.5 didn't — what you were willing to break
- Which components carried the clarity gain, for the detail crops
- What you handed off vs. held with the junior designer
- How the partner review cycle worked on this one
- Screens: control/1.5/2.0 three-up, 2.0 full page, detail crops, winning research variations
This audience arrives apprehensive, not evaluating an aesthetic
The card is for people who don't have a credit history yet. First-timers, rebuilders, people who've been declined. They aren't judging a visual language — they're working out whether they'll be approved, whether they're being taken advantage of, and whether they understand the terms.
The question wasn't whether to modernize. It was how far.
Push the visual language far enough and you gain contemporary polish while losing institutional trustworthiness. A page that looks like a fintech startup is a liability when the reader's live question is can I trust this company with my credit. So rather than pick a point on that spectrum and defend it, I tested two.
Research said clarity beat novelty. Twice.
Preference surveys ran on both creative rounds. The same finding came back both times: the winner was the option that read as clearest, not the one that read as most current. Top adjectives across both rounds were “informative” and “easy-to-read.”
Those aren't exciting words. For this audience they're the correct ones — what someone says about a page that answered their question without making them feel stupid.
Why the agreement matters more than the lift
+23% isn't the largest number in this portfolio. What makes this the most useful project is that research and live performance pointed the same direction — which, across the eighteen initiatives I later catalogued, happened almost nowhere.
When preference research and live behaviour agree, you've learned something transferable. When they disagree, one of your instruments is measuring the wrong thing and you don't yet know which.
What I'd flag
The limitations I'd raise before someone else did.
- The test was designed to watch for one risk, and the risk showed up.
- 2.0 won application yield decisively, but its approval rate slipped about a point and a half while 1.5 held flat. At this volume that's inside noise — but the direction is worth taking seriously, and the next test should separate application lift from approval quality instead of reporting one number that blends them.
- The record said it failed.
- The epic for this work reads Abandoned. It shipped, it tested, and it won.