The averaging problem
Report conversion as one number and mobile failure becomes invisible. A site converting well on desktop and badly on mobile produces a middling aggregate that looks like a site with no particular problem. Segment by device before doing anything else.
This is the most common reason a mobile regression survives a launch. Nobody was looking at the number that would have shown it.
What actually causes it
- Forms designed at desktop width. Six fields side by side become a scrolling ordeal, and every field is a place to abandon.
- Tap targets too small or too close. If a thumb cannot reliably hit the primary action, nothing else matters.
- The primary action below several screens of scroll, because the desktop hero collapsed into a tall stack.
- Load time on cellular. Testing on office wifi tells you almost nothing about the connection your buyers have.
- Interstitials and cookie banners covering the content someone arrived for.
- Phone numbers that are not tap-to-call, on sites where calling is the primary conversion.
Test on the device, not the emulator
Browser device emulation shows you layout. It does not show you the CPU throttling, the network latency or the thumb reach that determine whether a real person completes the action. A mid-range Android phone on cellular is the honest test, and it is frequently unflattering.
The order to fix things in
Fix load and the primary action first — they gate everything else. Then forms, then the content order. Redesigning the visual treatment before any of that is spending on the least consequential layer.