After writing about the placement redesign where a click-through drop turned out to be fine — users were simply exploring elsewhere on a platform where that's a legitimate thing to do — a related case kept nagging at me. The same reasoning, applied carelessly, can excuse almost any regression. Here's a redesign where that excuse would have been wrong, and the two questions that would have caught it.
The Setup: Removing a Step That Wasn't Actually in the Way
A product page had a "See more" link that took users to a dedicated grid — a simple, thumbnail-dense, vertically scrolling page showing everything in a category. The redesign replaced it with an inline horizontal carousel embedded directly on the page, cutting out the extra tap entirely. The intuition behind the change was straightforward: fewer steps between the user and the content should mean a smoother, more seamless experience.
Impressions on the now-inline items went up, since they were visible without any click at all. Clicks on those items did not keep pace — click-through rate on the carousel came in lower than the old "See more" page's click-through rate had been.
The Tempting Read
Having just been through the placement redesign where a CTR drop turned out to be a non-issue, the easy explanation was sitting right there: "of course CTR is down — more content is visible without a click now, the denominator grew, and users who don't click the carousel are still exploring the page, just not this specific row of it. Fewer navigation steps is objectively a smoother experience." That explanation would have closed the discussion. It also would have been wrong to accept without checking.
Why Fewer Steps Isn't Automatically Better
The "See more" page wasn't friction for friction's sake — it was doing real work. A dedicated, thumbnail-only grid that scrolls vertically is a genuinely easy interface for browsing many items quickly: users already know how to scan a vertical grid, they can see more items at a glance, and vertical scrolling is the platform's default, most practiced gesture. A horizontal carousel shows fewer items in the same screen space, requires a less-practiced swipe gesture on many platforms, and doesn't afford the same fast scanning. Removing a click can just as easily remove a better interface as it can remove an unnecessary one — "the user has to tap once more to get here" and "the user has to tap once more to reach a page that's actually easier to browse" are very different situations wearing the same description.
This is the same trap from a different angle: I'd correctly resisted "the metric moved the wrong way, therefore it's bad" in the previous case. Here, the risk was the mirror mistake — "the redesign removed a step, therefore it's good" — without checking whether the step being removed was actually the part doing the user a favor.
The Question That Actually Settles It
The test I should have reached for immediately: what was this specific redesign actually for? If the goal was to redirect users toward a different surface or action entirely, then a drop in on-surface engagement can be legitimately offset by gains elsewhere — that's the exploration-platform argument, and it has to be verified by tracing where the displaced attention goes, not just assumed. But if the redesign's own stated purpose was to make browsing this specific content easier and more direct — which "cut out the extra step" clearly was — then the metric that actually answers "did this work" is engagement with the content itself. A drop in click-through on the very surface the change was meant to improve isn't offset by "well, it's an exploration platform." That reasoning only applies when redirection was the point. Here, it wasn't — the point was to make this exact interaction better, and by its own click-through numbers, it didn't.
A Second Wrinkle: Click and Like Aren't the Same Signal
Digging into the session data surfaced a second complication. Click-through on the carousel was down, but the like (save/wishlist) rate on carousel items was up. At first glance that looks like it could net out — engagement just moved from one action to another. It doesn't net out cleanly, because a click and a like aren't measuring the same intent.
A click, in this context, generally signals "I want to look at this specific item more closely, right now" — active, in-the-moment consideration. A like signals something closer to "I want to remember this for later" — a bookmarking action that doesn't require, or necessarily lead to, immediate deeper engagement. These are different points in a user's decision process, not interchangeable proxies for the same underlying interest. A carousel that produces fewer clicks and more likes might mean users are browsing faster and saving more for later — a real, if different, behavior change — or it might mean the carousel format nudges people toward a low-effort tap (liking) instead of the higher-effort action (clicking through) they'd have taken on the old page, without actually serving their original intent any better. Those are very different stories, and the raw counts alone don't distinguish between them.
What This Adds to the Earlier Lesson
The point isn't that the earlier retrospective was wrong to accept a CTR drop as acceptable — it wasn't, because in that case the redesign's actual purpose was never to maximize engagement with that one surface, and the displaced attention was traceable to real value elsewhere. The point is that the same logic isn't a universal alibi. Before crediting a click-through drop to "healthy exploration," two questions have to be answered honestly: was this surface's own job actually to redirect users elsewhere, or was its job to be engaged with directly? And if engagement shifted from one action to another instead of disappearing, do you actually know what each of those actions means to the user, or are you assuming they're interchangeable because they both count as "activity"? Skipping either question turns a legitimate, case-specific argument into a blanket excuse that would defend almost any regression a redesign could produce.