Core Web Vitals and conversion: what speed fixes and what it does not
Performance discussions usually collapse into one number: “our score is 62, let’s get to 90”. But that score is a lab simulation and does not directly reflect what real users experience. Meaningful improvement comes from fixing the slow slice of users in field data.
Three metrics, three different problems
| Metric | Measures | Target |
|---|---|---|
| LCP | Time for main content to appear | Under 2.5 s |
| INP | Responsiveness to interaction | Under 200 ms |
| CLS | Unexpected layout shift | Under 0.1 |
These are not three views of one problem but three separate problems. LCP is usually images and server time; INP comes from heavy JavaScript; CLS comes from unsized images, late-loading ad slots and fonts. Optimising without knowing which one is bad is usually effort in the wrong place.
Lab data versus field data
A test run in the browser assumes one device and one connection. Real user data shows a 28-day distribution and speaks about the worst quartile. It is very common for the average to look fine while the top slice suffers. Decide on field data; use lab tests only to find the cause.
The four most common causes
- Unsized images: serving the hero image in a modern format at the right dimensions with
width/heightset fixes both LCP and CLS. - Third-party scripts: chat widgets, heatmaps, several analytics tools. Each occupies the main thread and hurts INP. Removing the ones nobody uses beats optimising code.
- Font loading: hiding text until the font arrives badly damages perceived speed. Show text in a fallback immediately.
- Server response time: uncached queries and a slow first byte put a ceiling on every other improvement.
What speed does not fix
Conversion does not automatically rise when a site gets faster. Speed removes an obstacle; it does not persuade. You win back visitors lost to slowness, but a visitor who does not understand the offer or trust the vendor will not convert on a fast page either. Put performance work before content work, not instead of it.
Measurement discipline
Ship changes one at a time and wait for field data to move; it lives in a 28-day window, so do not trust day-one jumps. Track mobile and desktop separately and, most importantly, look per page type. A fast homepage means nothing if the product or offer page where conversion happens is slow.