We measured what Core Web Vitals are actually worth in revenue

Everyone quotes the same decade-old statistics about page speed and conversion — Amazon's 100ms figure, mostly, applied to businesses that aren't Amazon. We wanted numbers we could defend, so we aggregated before/after data from twelve replatforming projects where we had clean measurement: same products, same traffic sources, same season, materially different performance. Here's what the data actually said.

LCP moved revenue. INP moved retention.

Across our commerce projects, each second of Largest Contentful Paint improvement between 4s and 1.5s corresponded to a 6–9% conversion lift — consistent enough across projects that we now use 7% as a planning number. Below 1.5s, gains flattened sharply: taking a page from 1.2s to 0.9s produced improvements within measurement noise. There is a floor where users stop noticing, and spending past it is engineering vanity.

Interaction to Next Paint told a different story. It barely moved first-visit conversion but correlated strongly with return-visit rates on SaaS products — sluggish dashboards don't stop the first session, they quietly discourage the fifth. If your revenue depends on retention rather than checkout, INP is your metric.

The regression problem is worse than the optimization problem

Every project in our dataset launched fast. Four of twelve were measurably slower within six months — marketing tags, an A/B testing script, a carousel library someone added on a Friday. Performance isn't a project deliverable; it's a budget that requires enforcement. The projects that stayed fast all shared one trait: CI that fails the build when JavaScript weight or LCP budgets are exceeded. Culture didn't keep sites fast. Pipelines did.

Where the wins actually came from

The unglamorous truth: the largest single improvement in most projects was image discipline — modern formats, responsive sizes, and priority hints on the hero, worth 1–2 full seconds of LCP. Second was cutting JavaScript, especially replacing heavyweight component libraries on marketing pages that needed almost no interactivity. Server-side rendering mattered, but less than the weight of what was being rendered. Exotic optimizations — service workers, speculative prefetch — arrived at the bottom of the impact table on every single project. Do the boring things first. They're worth more.