The tests people run first are the ones that matter least
Conversion work often starts with headline variants and button colors. Those tests are easy to set up and easy to report on, and on most sites they move almost nothing.
The losses that matter happen earlier: the visitor who leaves before the page renders, and the visitor who reads it and cannot tell what they are supposed to do next. Neither is a persuasion problem.
Speed is a conversion factor, not a technical metric
Treat Core Web Vitals as commercial numbers. A page that takes four seconds to become useful on a mid-range phone has already lost a meaningful share of the traffic you paid for, before any of your messaging has been read.
The usual culprits are unremarkable: uncompressed hero imagery, render-blocking third-party scripts, fonts that block text from appearing, and layout that shifts as elements load. All are fixable without a redesign.
Measure on field data at the seventy-fifth percentile, not in a lab on a fast connection. Your worst quartile of visitors is not a rounding error; on paid traffic it is a quarter of your spend.
One page, one decision
Clarity is mostly subtraction. A page with a single primary action, stated in the same words as the ad that brought the visitor there, will outperform a page offering four routes — even when each of the four is individually reasonable.
Check message match explicitly. Read the ad, then read the first screen of the landing page. If a stranger could not tell they belonged together, the ad account is not the problem.
Then count the steps between arriving and acting. Every field, every confirmation, every question that could be asked later instead of now is a place where people stop.
Test one decision at a time
Once the structural work is done, testing becomes worth doing — but only against one decision point at a time, and only with enough traffic for the result to mean something.
How long that takes is a function of your volume, not of anyone's schedule. At typical mid-market traffic it is two to six weeks per decision. Anyone promising faster is either testing something with a very large effect or reading noise.