Mobile speed work for the pages your ads point at

The number worth arguing about is p75 LCP in field data for the exact URLs your ads point at, not a Lighthouse score for your homepage. Paid landing pages routinely measure one to two seconds worse in the field than in the lab, because the lab never simulates a cold cache on a mid-range Android handset arriving through two or three redirect hops from a click tracker. We start at the field number and work backwards. What we measure first: A real handset on a throttled connection, a Chrome trace, and RUM from the web-vitals library reporting to an endpoint you own, so the pages carrying your spend are read individually rather than averaged into a site-wide CrUX bucket that low-traffic landing URLs never populate on their own. INP replaced FID in March 2024 and is the metric most paid pages now fail; the two usual causes are the consent banner and a tag manager container that has accumulated tags nobody removed. The click chain before your page: Between the ad click and your first byte there is often a tracker redirect, an ad platform redirect and a decoration parameter that defeats the cache. We time that chain and report it, even where the fix belongs to your media agency rather than to us. It is common to find several hundred milliseconds sitting outside the page entirely, and no amount of image work recovers it. What we change: The LCP element gets fetchpriority high, loses any lazy attribute it should never have carried, and is served as AVIF at breakpoints matching the devices in your own analytics rather than a default set. Fonts get font-display and size-adjust so the swap stops moving text under the reader's thumb. Third-party tags are cut, deferred, or moved to a server-side container on a first-party subdomain where they stop competing for the main thread during the first three seconds. Where the origin is the problem we say so and fix cache rules, rather than adding a plugin on top of one. Not included: Application rewrites, theme replacement or design changes made in the name of speed. Media buying and ad account changes. CDN or hosting contracts, which you sign directly. Ongoing tag governance unless you take the retainer, because a page we tuned in March and a team that added four tags in April are not the same page. Who this is not for: Anyone unwilling to change a hero. If the autoplay video at the top is not negotiable then the LCP number is already decided, and we would be taking a fee to confirm it. Also not for teams where anyone in marketing can publish a tag without review; we will not hold a number that a Tuesday afternoon container publish can undo. What we refuse: We refuse to chase a Lighthouse 100. It is a lab score on a simulated device and it has cost more real engineering hours than any other vanity metric in this discipline. We refuse to remove or delay a consent banner to improve a metric, because the metric is not the thing you are exposed on. We refuse to install an all-in-one optimisation plugin that inlines, defers and rewrites indiscriminately; when it breaks a checkout at eleven at night, nobody can tell which of its forty switches did it. And we will not sign up to a number for a page whose content and tag stack pass out of our hands at handover.