Core Web Vitals: what actually moves the needle

Performance work attracts busywork. Teams spend weeks shaving kilobytes off a bundle while the largest element on the page is an unoptimised hero image loading from a slow origin. The fixes that move real-user metrics are usually unglamorous, and there are not many of them.
Fix the largest element first
Largest Contentful Paint is dominated by one element on most pages — typically a hero image, a headline blocked by a web font, or a banner rendered client-side. Identify that element per template, then remove everything standing between the browser and painting it: preload it, serve it in a modern format at the size it is displayed, and make sure it is not waiting on JavaScript to appear.
Stop layout from moving
Cumulative Layout Shift is the cheapest of the three metrics to fix well, and the most visible to users — nothing erodes trust faster than a button that moves as it is tapped.
Treat third-party scripts as a budget
Tag managers accumulate. Audit what is loading, who asked for it, and whether it still serves a purpose; then defer everything that is not required for the first interaction. A single poorly behaved analytics or chat script can account for most of the interaction delay on mobile.
Optimise for mid-range mobile, not your laptop
Field data is segmented by device for a reason. Test on a throttled mid-range Android over a 4G profile — that is where interaction latency shows up, and it is where a meaningful share of your audience actually is. Improvements that only appear on desktop rarely change the aggregate.
What usually is not worth it
Micro-optimising CSS selectors, chasing a perfect lab score, or rewriting a framework to save a few kilobytes generally will not shift field metrics. Neither will lazy-loading the hero image — a common own goal that delays the exact element the metric is measuring.
Make it stick
Performance regresses the moment nobody is watching. Set budgets in CI, monitor field data monthly by template, and treat a regression as a bug rather than a backlog item. If your site is slipping and you are not sure why, that is the kind of question we enjoy answering.