One Platform, One Population

532 sites on a single dealer-program platform carry a 17.2s median mobile LCP and a 0.2% mobile pass rate, against 8.4s and 3.4% across the other 7,175 verified dealer sites.

Those are not two samples of the dealer layer. They are one measured population cut along a line that has nothing to do with performance, which is whether the dealer’s website was built by the program the dealer belongs to. The two halves barely overlap. Three quarters of the program sites are slower than the point where the rest of the cohort has already finished loading, and the gap holds in every month in which both groups were audited.

The obvious reading is that franchised dealers run worse websites than independent ones. The corpus does not support it. Three other franchise and buying-group programs sit inside the same measured population, and all three perform at or above the cohort. Whatever is happening here is a property of one shared build rather than of the arrangement that produced it.

That distinction is the whole finding. A dealer on this platform did not choose a slow website. They joined a program, took the site the program provided, and inherited its delivery characteristics along with everything else the program gave them. Five hundred owners made one decision each, and one vendor decision produced the result.

What it means for a manufacturer or a program

A manufacturer looking at the dealer layer sees several thousand independent businesses and reasonably concludes that the website problem is diffuse, which makes it somebody else’s to solve. That is true of most of the layer and not true of this part of it. Where a program supplies the platform, the websites of hundreds of dealers move together, and they can be changed together.

That cuts both ways, and the direction it cuts is decided years earlier by whoever selects the platform. A program that picks well gives every member a floor they would not have reached individually. A program that picks badly distributes the same defect to every member, and none of them can see it, because a dealer comparing their site to the one down the street is comparing it to another site on the same build.

For anyone choosing what a dealer network will run on, this is the measurement that argues the choice is worth more attention than it usually gets. The platform decision is not a back-office procurement item. It is the delivery characteristic of the entire network’s public face, set once, for everyone, for as long as the contract lasts.

One limitation

The cluster is identified from a shared signature in the sites themselves, which means the count is a floor rather than a census. Any member whose site does not carry that signature is sitting in the comparison group, which makes the contrast conservative and not the reverse. The three counter-example programs are each too small a group to carry a published figure of their own, so they are stated here as direction and shape and are doing one job, which is to stop this from being read as a claim about franchising. And the measurement describes page delivery on the day it was taken. It says what this platform produces, and it says nothing about why, because nothing in the audit reaches inside the build to find out.