The instinct is understandable: you run Lighthouse on your store, see a 54, and book a sprint to fix it. But Core Web Vitals scores and actual user experience have a gap that most optimization work never closes.
Here's what I keep seeing.
The wrong metrics get optimized
Performance budgets almost always target LCP and Total Blocking Time — the numbers that show up in audit reports. What they miss: perceived responsiveness on the first tap. On mobile, the time between a user touching a product image and seeing any visual response is often 400–800ms on stores that score fine in desktop Lighthouse. The culprit is almost always a third-party script bundle — review widgets, loyalty programs, live chat — that hasn't been deferred because "the vendor said it needs to be in the head."
It doesn't need to be in the head.
The image problem is almost never the images
Stores over-invest in compression and under-invest in layout shift. CLS is the Core Web Vital that most directly maps to the "this site feels janky" sensation people describe but can't name. When a product grid reflows as images load — or when a sticky header shunts content down on scroll — shoppers lose their place. That cognitive interruption costs conversions in ways that file size optimization never will.
The fix is usually boring: explicit width and height attributes, aspect-ratio placeholders for images, and font-display: swap with a system-font fallback that approximately matches the heading metrics. None of this requires a new build tool.
Speed is table stakes. Trust is the conversion lever.
After enough storefronts to have opinions: speed gets shoppers to the product page; trust closes the sale. Social proof placement, return policy visibility, and clear shipping timelines matter more than shaving 200ms off your TTFB once you're past a reasonable baseline.
Most stores aren't past a reasonable baseline — which is why speed work still matters. But it's a floor, not a ceiling. Once you're in the green on CWV, the next conversion gain almost always comes from copy, layout, or trust signals. That's where the interesting design problems live.