A second is not a vibe

People treat speed as a design opinion until they watch a form. Deloitte’s milliseconds-make-millions paper, written with Google, looked at mobile sites in retail and travel and reported that a 0.1 second improvement lined up with about a 10% lift in retail conversion in that dataset. The study is from 2020. It is not a law, and it is not your analytics. It is the cleanest public evidence that delay and money move together.

web.dev’s Renault case is the other number worth keeping. Across more than 10 million page views, a one-second improvement in Largest Contentful Paint went with a 14-point bounce-rate drop and a 13% conversion increase. Again: one company, one period, one metric. Anyone who turns that into ‘we will add 13% to your enquiries’ is selling a study they did not run.

What you can specify is the work that usually produces the second. The largest element is an image with a width, a height, a modern format, and fetchpriority high. It is not lazy-loaded. The font is one file, subset, with a fallback that does not shove the headline when it arrives. Third-party chat, heatmaps and tag managers are the usual INP tax. If a tool does not change a decision you make every week, it does not ship.

Measure the field, not the demo. PageSpeed Insights will give you a lab score on a fast connection that your buyers do not have. Search Console’s Core Web Vitals report is the 28-day window of real phones. Improvements show up there on a delay. A before-and-after on the day of launch is theatre.

On our builds the speed budget is part of the scope: what the first screen may weigh, which scripts are forbidden, and which report we will look at a month later. The enquiry is a mailto. There is no widget waiting to paint over the button.