What Core Web Vitals Actually Cost You
Core Web Vitals are usually sold as an SEO problem. They are a real ranking input, but a modest one, and framing them that way understates the cost. The expensive part is what slow pages do to behaviour before Google is involved at all.
What each metric is actually telling you
LCP — how long until the page looks usable
Largest Contentful Paint measures when the main content element renders. On a category or product page that is almost always an image. A slow LCP does not read to a shopper as “slow site”; it reads as a blank or half-built page, and the decision to leave happens in that window.
INP — whether the page responds when touched
Interaction to Next Paint measures the delay between an interaction and the visual response. This is the one that hurts most on mobile commerce, because the interactions it covers are the funnel: variant selection, filter toggles, add to cart, quantity changes. A page that looks fast but responds at 400ms feels broken in a way people struggle to describe.
CLS — whether the page moves under you
Cumulative Layout Shift is the cheapest to fix and the most directly punishing. A shopper who taps the wrong thing because a banner loaded late does not file a bug report. Mis-taps near the add-to-cart button convert into abandonment.
Why the funnel cost is larger than the ranking cost
Ranking effects are aggregate, gradual, and mixed with dozens of other signals. Behavioural effects are immediate and compound: every slow session costs you across the funnel at once — fewer category pages viewed, fewer products opened, more abandonment at the point of interaction.
This is also why lab scores mislead. A synthetic test on a fast connection tells you the page can be fast. Field data tells you what your actual customers experienced on the devices and networks they actually have, which is where the money is.
Where to look first
Segment field data by device before anything else. Desktop numbers routinely hide a mobile problem, and mobile is where most storefront traffic and most of the performance gap live.
Then segment by template, not by page. Category, product, cart and checkout have different failure modes, and an average across them points nowhere. A single failing template is a fixable project; a site-wide average is a vague anxiety.
Look at the worst quartile rather than the median. Vitals are assessed at the 75th percentile, and more importantly, the slow quartile is disproportionately made up of people on constrained devices — who are not a rounding error in most catalogs.
The practical order of work
Fix CLS first: it is usually cheap, usually caused by a handful of un-reserved spaces, and immediately improves how the page feels. Then LCP, which on most storefronts is an image delivery problem before it is a server problem. Then INP, which tends to require harder decisions about third-party scripts and is the right place to spend real engineering time.