Site speed is not one score: improve Core Web Vitals users can actually feel

A manager sends a screenshot showing a score of 92. On the same day, three customers say the mobile buy button feels unresponsive. Both accounts can be true: the tool saw a controlled run, while customers experienced modest phones, busy networks and production scripts.

A useful performance review follows the commercial journey before opening a scorecard. The delay that blocks a product choice or payment matters more than a weak number on a page almost nobody visits.

Three signals, three different problems

LCP describes when the main content becomes visible; a heavy hero image, blocking font or slow server response often drives it. INP reflects interaction delay and suffers when long JavaScript tasks occupy the main thread. CLS is the unexpected jump that moves a button as someone tries to tap it.

There is no single fix for all three. Compressing the hero may improve LCP while a heavy chat widget continues to damage INP.

Separate lab and field evidence

Lab tools are excellent for reproducing a problem and comparing a controlled before and after. Field data captures the devices, networks and behaviour of real visitors. Sound decisions use both.

Group evidence by commercial template—home, category, product, cart and checkout. A site-wide average can hide a serious problem on the page that produces most revenue.

Prioritise by impact and risk

Optimising the LCP asset, reserving media dimensions, limiting third-party scripts and splitting long tasks are common starting points. Give the site a performance budget for JavaScript, images and external requests.

Measure the same journey after release. If a technical score rises but engagement falls, the intervention has not met the business goal.

A professional performance deliverable

• Baselines exist for key templates.
• Each poor metric has a root-cause hypothesis.
• Fixes are ranked by impact, effort and risk.
• Before and after tests use comparable conditions.
• Monitoring detects regressions after launch.

Official reference: Web Vitals guidance on web.dev