Core Web Vitals, translated into money
Core Web Vitals have an adoption problem, and it is not technical. It is that they are expressed in units nobody buys anything with. A finance director has never once approved a piece of work because a metric moved from amber to green.
So here is the translation, with the arithmetic shown, because the arithmetic is the part that gets the work done.
The three, in plain terms
Largest Contentful Paint is how long until the main thing has appeared. Under 2.5 seconds is fine. It answers: did they wait?
Interaction to Next Paint is how long the page takes to respond after someone taps something. Under 200 milliseconds is fine. It answers: did it feel broken?
Cumulative Layout Shift is how much the page jumps around while loading. Under 0.1 is fine. It answers: did they tap the wrong thing?
The conversion, in four steps
You need four numbers, all of which you already have.
- Monthly sessions on the pages in question.
- Current conversion rate on those pages.
- Average order value, or average lead value if you sell services.
- The current LCP at the 75th percentile — the figure any audit reports, and the one Google actually scores you on.
The published research clusters around roughly a 7% change in conversion for each additional second of load time in the two-to-five second range. Treat that as a planning figure rather than a law of physics; the point is to get an estimate that is defensible, not exact.
Worked example
A shop doing 40,000 sessions a month, converting at 1.8%, average order £62, LCP at 4.9 seconds.
Removing 2.1 seconds — which in that case was one oversized hero image — puts the estimate at about a 14% relative improvement in conversion. That is 1.8% becoming roughly 2.05%.
On 40,000 sessions that is 100 additional orders a month, or £6,200. Against about half a day of work, once.
Now the sentence is fundable. "Our LCP is 4.9 seconds" is not.
Three honest caveats
It is an estimate. Speed and conversion correlate, and the mechanism is real, but your traffic mix will move the number in both directions. Present it as a range and say so out loud — a range you can defend beats a point estimate you cannot.
Field data beats lab data. A score generated by a test tool on a simulated connection is a diagnostic, not a measurement. The Chrome UX Report shows what actually happened to real visitors on real devices. When the two disagree, the field data is right.
Averages hide the problem. Always use the 75th percentile, never the mean. Your median visitor is on a good phone on wifi and is having a lovely time. The visitor deciding whether your business is worth the wait is at the wrong end of the distribution, and they are the entire reason this work pays.
What this does to your roadmap
Once every finding carries a number, prioritisation stops being an argument about taste. Half a day for £6,200 a month goes to the top. A fortnight of framework migration for 300 milliseconds goes to the bottom, where it belongs, no matter how much anyone wants to do it.
That reordering is worth more than any individual fix on the list.