The constraint, stated plainly
For about two years I led design for a partnership serving a top-five US credit card issuer. Navigation, category pages, product templates, whole-page recreates — the surfaces where someone chooses a credit card. That work is covered by an NDA, and the screens are the client's property.
Which leaves a real problem: the interface is the least interesting part of this work and the only part I can't show. The decisions, the tests, the arguments I lost and won — all of that is mine to talk about. The pixels aren't.
So I rebuilt the pixels. Every screen in this portfolio is composed from a fictional issuer's design system, built from scratch for exactly this purpose.
Real, abstracted, and withheld
The distinction that matters most, since it's the one a skeptical reader is actually asking about:
| Element | Treatment |
|---|---|
| The problem | Real. Framed as I framed it at the time. |
| The research | Real. Preference surveys, card journey studies, click analysis. |
| The tests | Real. Live A/B, running weeks, on production traffic. |
| The results | Real, as relative lift. Direction and magnitude unchanged. |
| The decisions and reasoning | Real, including the ones that turned out wrong. |
| My role | Stated precisely on every case study. Sole designer where I was; lead where I led. |
| Client name | Abstracted — “a top-five US card issuer” |
| Product names | Abstracted to fenmore equivalents |
| Screens | Redrawn from scratch in a system I built |
| Absolute figures | Withheld. Approval counts, revenue, traffic volume. |
| Internal documents | Withheld entirely. Never shown, never linked. |
| Ticket IDs, data warehouse names | Withheld. |
Everything in the first group is what you'd be hiring. Everything in the second is a stand-in for something I don't own.
Why relative figures and not absolute ones
This one's worth explaining because it looks like hedging and isn't.
A percentage alone reveals nothing. An absolute count alone reveals little. Publishing both together lets anyone solve for the baseline — a lift of 14% alongside “+802 monthly approvals” gives you roughly 5,700 approvals a month, and from there you can back into traffic volume and conversion rate for a named issuer.
That's competitive intelligence about someone else's business, published by me, for my own benefit. So the portfolio picks one register and stays in it.
I have the absolute numbers and I'm happy to walk through them in a conversation. They just don't belong on a public page.
What had to survive abstraction
Redrawing a screen is easy. Redrawing it so it still supports the argument is the actual work, and it meant deciding which properties were load-bearing.
The counts
In the navigation case study, every link carries a number — All cards (40), Balance transfers (3). That's the smallest change in the project and the most transferable one: it tells you the size of what's behind a link before you spend a click on it. An abstraction that dropped the counts would have removed the evidence for the central claim, so they're preserved as plausible generic figures.
Card tile density
The category page case study is about crowding — fifteen products with identical visual weight, where scanning produces no signal. A tidied-up redraw would have quietly deleted the problem being solved. So the tiles stay as dense as they were.
Co-brand breadth
The strongest finding in the navigation work is that brand-first shoppers were the largest single group using the nav. That only reads as a real population if you can see a dozen-plus partner brands in the panel. The names are invented; the volume is accurate, because the volume is the point.
Abstraction that flattens the thing you're arguing about isn't abstraction, it's deletion. Deciding what to keep was most of the work.
fenmore is a token system, not a company
The fictional issuer in these screens is called fenmore. It has no story, no founding date and no market position, because it isn't a character — it's eighty design tokens and fourteen components, built so thirty-odd abstracted screens could be consistent with each other instead of thirty separate redraws.
Two things about it that mattered more than they should have:
The name was cleared. The first candidate was a real VC-backed fintech selling credit software to banks — which would have been a worse problem than the one it was solving. A fintech-literate reader seeing that name on credit card screens might reasonably have wondered whose product I'd actually worked on.
The build script refuses to ship a reskin that fails contrast. If any text token drops below 4.5:1 on a surface it's permitted on, generation halts. That caught two failures in my own alternate palettes — a muted text value two hundredths under threshold, and a monogram at 4.05:1 on its own chrome. Both would have shipped silently.
The full system is documented here — tokens, components, the contrast gate, and a live theme swap.
What I'd tell you in a conversation
The material that doesn't belong on a public page but is the first thing I'd bring to a real one.
- The absolute numbers.
- Approval counts, session volumes, the full funnel tables. Including the one figure I'd most like to publish and won't.
- The unabstracted screens.
- What the work actually looked like, and where the redraw diverges from it.
- The organizational detail.
- Who pushed back and on what, how the partner review cycle actually ran, which arguments I lost. The case studies say a compromised design went to test without me recording the delta — the specifics of how that happened are more useful out loud.
- What I'd do differently with the same brief.
- Two of these initiatives had the answer sitting in pre-test data and nobody looked, including me.