On a storefront, performance is not a technical concern

Every section a marketing team adds makes the first paint later. At some point the page arrives after the visitor has already decided, and the funnel leaks at the top.

Nukes AI

Commerce team

A storefront that takes too long to become useful loses the visitor before anything else about it matters — before the photography, the copy or the price. That loss does not appear in any report as a loss. It appears as traffic that arrived and did not convert, which is read as a marketing problem.

They leave before it lands

Nothing on a storefront happens until the page is usable. A visitor still looking at a blank rectangle has no opinion about the product, and the ones who leave at that point are counted as traffic that did not convert rather than as a page that did not arrive.

That is why performance sits with the commercial numbers rather than the engineering ones. It is the first filter every visitor passes through, and it is applied before any of the expensive work on the page has been seen.

What the thresholds describe

The Core Web Vitals are three questions rather than three numbers: how long until the main content is painted, how quickly the page responds when it is touched, and how much it moves about while it loads. Published thresholds exist for each and are worth reading, but the questions are the useful part.

Images are usually the answer

On a commerce page the largest element is almost always a photograph, so the largest paint is a picture arriving. Correct sizes, a modern format, real width and height attributes, and the hero image not being lazily loaded will move that number further than most rewrites.

Third-party scripts compound

Every tag added to a storefront is paid for by every visitor, on every page, for as long as nobody takes it off again.

Analytics, chat, reviews, heat maps, two testing tools and a consent banner each look small on their own. They arrive on the main thread, they arrive everywhere, and none of them were in the estimate. An audit of what is installed is usually the cheapest performance work available to a team.

Layout shift is a trust problem

A page that reflows under the cursor makes a person press the wrong thing, and a person who has pressed the wrong thing once becomes careful. Reserving space for anything that loads late — banners, badges, embedded reviews — costs nothing and removes it.

Measure the field, not the lab

A score from a local run describes a fast machine on a fast connection. What matters is what real visitors experienced, on the devices they actually own, which is what field data reports and a lab score cannot.

test on a mid-range phone
test on a throttled connection
watch the slowest tenth, not the median

The median customer is not the one being lost.

Performance needs an owner

Speed is not a project that gets completed. It regresses every time a section is added, and it regresses quietly. A budget enforced in the build — a size limit that fails the deploy — is the only version of this that survives a busy quarter.

What this actually buys

Not a badge, and not a score to put in a report. A page that is present when the visitor is, which is the condition every other thing on it depends on.