Most people arrive undecided. Not all of them.
The marketplace is where someone lands wanting a credit card without knowing which one. Dozens of products split by rewards structure, annual fee and credit tier. The nav is the first decision tool anyone touches — and the highest-traffic surface in the partnership.
Most visitors are doing a comparison task they've never done before, with vocabulary they don't own. “Cash back” versus “rewards” versus “points” are distinctions the industry treats as obvious and shoppers hear as noise.
But a large slice arrives brand-first — they want the airline card, the hotel card, the one their partner has. For them an intent taxonomy is an obstacle between them and a product they've already chosen.
Any structure serving only one of those populations was going to tax the other. That was the real problem, and it wasn't in the brief.
I split the problem into two altitudes before designing anything
Navigation debates get muddled because two questions get argued at once: what the nav bar is, and what happens inside it. Different constraints, different failure modes. Exploring them separately meant I could recombine them — which is how four genuinely different structures came out of it instead of four coats of paint on one idea.
Every direction below was worked up far enough to argue about. Select any one to see what it was for and what happened to it.
The highlight rail forced the most interesting small decision
Do the visual highlights show intents — cash back, travel rewards, flights with points — or products, the flagship cards by name with card art?
Not a styling preference. Lifestyle assumes people arrive knowing what they want to do. Card highlights assume they know what they want to buy. For a marketplace that exists because visitors haven't decided, I weighted intent — then the brand data complicated that answer.
Two structures went live, built on different theories
Same frame, same scroll position. Toggling is the fastest way to see what actually changed — and what each change cost.
—
What made Variant 1 what it was
Intent labels do the work. “Pay Less” isn't a product attribute — it's a goal, collecting no-annual-fee, 0% intro APR and balance transfers under the reason someone would want any of them. Naming the group after the motive instead of the mechanism is the cognitive-load hypothesis in two words.
Brand gets its own axis. Proprietary families and the full co-brand portfolio as a first-class path. Someone who wants the Southwest card shouldn't be routed through “Travel Rewards” to find it.
Every link carries a count. You know the size of what's behind a link before spending a click on it. Smallest change here, and the one I'd port anywhere.
The question I built Variant 2 to answer wasn't “does audience-first convert better.” It was: if this split is becoming permanent architecture, is top-level navigation where it earns its keep?
Framed that way the test was productive either way. A win validates the nav as the right home for a strategic priority. A loss says the priority needs a different surface — which beats the split getting implemented everywhere on the strength of strategy alone.
Four structures were build-ready. Two went live. The flat chip-row variants were paired with an expanded footer that put them out of engineering scope for the round.
Both variants won. I recommended the one that engaged people less.
Both delivered roughly +10% session yield, and Variant 1 came back with a 100% probability of outperforming control. So the headline metric was a tie, and the tiebreaker sat in the secondary metrics — which pointed in opposite directions.
Variant 2
Looked more engaging
Roughly a third more navigation interaction than control — the largest engagement figure anywhere in the test.
Variant 1 · recommended
Performed better downstream
+18% apply-now rate. Exit rate down about 60%. Deepest browsing of any variant, +11% pages per session against V2's +5%.
The number that settled it
Apply-now rate on product pages, among people who actually used the nav:
Variant 1 clickers
Control clickers
Variant 2 clickers
| Page type | Control | Variant 1 | Variant 2 |
|---|---|---|---|
| Product | 38% | 45% | 36% |
| Category | 16% | 19% | 16% |
| Interactive | 11% | 33% | 16% |
| Homepage | 19% | 20% | 21% |
Non-clickers barely moved on any variant, which is what makes the clicker split meaningful.
Variant 2's incremental clicks came from people who then didn't apply. It wasn't that engagement merely isn't conversion — V2 was manufacturing low-quality interaction. V1's clicks were qualifying.
Why the structure produced that
Audience-first navigation asks you to classify yourself before it gives you anything. The click data says roughly three-quarters of Variant 2's dropdown interaction went to the personal side — so for three of every four visitors, that first decision resolved nothing they didn't already know. They spent a click arriving where Variant 1 started them. It also split the inventory, so neither panel could show full breadth.
I argued the recommendation on the funnel evidence rather than the engagement figure, because the engagement figure was the easier one to present and the wrong one to act on. One quality check I made sure was in the readout: no material change in card mix between variants — the lift wasn't an artifact of steering people toward easier-to-approve products.
The brand axis became the most-used path in the nav
The strongest vindication wasn't the session yield. It was where people clicked.
In Variant 1, Card Brands was the single most-clicked item in the navigation — ahead of both category paths, by a clear margin. Individual co-brand partners surfaced in the top handful of clicks on their own.
Brand-first shopping wasn't a hunch I defended. Given a first-class path, it became the primary way people used the nav. Every structure treating brand as a subordinate branch of an intent taxonomy had been taxing the largest single group in the marketplace.
The business insight was real. It just wasn't a navigation problem.
Business shoppers clicked check-for-offers at a 38% higher rate than personal shoppers. Independent research pointed the same way: they want a product-specific pre-approval path — apply with confidence — before committing.
So the client's strategic instinct was correct. Business shoppers do need something different. It wasn't a nav split that cost the other three-quarters of traffic a click to deliver. I routed it to the business-card roadmap as a pre-approval flow.
That's the outcome I'd want to be judged on. Not that I killed an initiative on data — that a strategic priority arrived pointed at the wrong surface and left pointed at the right one.
Two findings that outlived the test
Travel dominates. 29% of all category-page navigation clicks were travel or airline — far out of proportion to their share of the nav. The next iteration I specified was reordering the hierarchy by actual engagement. The information architecture had been inherited from how the business thought about its portfolio, not how shoppers moved through it.
The record almost lost all of it. This test — the highest-confidence result in the entire dataset — was filed under an epic reading To Do, along with all eleven of its children. Fully run, rollout recommended, effectively invisible to anyone searching the project. That's the catalog case study.
What shipped
What I'd flag
The limitations I'd raise before someone else did.
- No research ran ahead of this.
- The design was reasoned from funnel and click behaviour, not from talking to anyone. The win replicated across a very large sample, but I can't tell you which specific change earned it. I'd want a research round in front of it next time — not to justify the direction, but so the result would be attributable.
- The variants weren't isolated experiments.
- Each bundled several changes. A cleaner design would have let me say why Variant 1 won instead of only that it did.
- Engagement metrics were reported without a caveat they needed.
- The nav-interaction figure sat in the same column as conversion metrics as though the two were commensurable. That framing is how teams talk themselves into the wrong variant.