Third Party Count

What a third party actually costs you

Every script your page loads from another company is a decision you have handed over. Their server being slow this morning is your page being slow this morning. Their outage is your outage. Their choice to ship twice as much code next quarter is your page getting heavier without anybody on your side touching anything.

None of that appears on your hosting bill, and none of it appears in the dashboards those same companies give you. The chat widget reports how many conversations it started. It does not report how many people left before it finished loading.

The individual decisions are almost always reasonable. Somebody wanted to see heatmaps for a month. Somebody added a review widget for a campaign. Somebody put in a second analytics tool to check the first one. Each was defensible on the day. Nobody ever removes them, because removing things is not anybody's job, and three years later the page opens connections to eleven companies and the business has forgotten what four of them do.

The number to compare yourself against

Benchmarks in this industry are usually somebody's opinion dressed up. This one is not. In the Florida aviation speed study every site of 1,042 Florida aviation businesses was measured three times with Google's own tooling, and the median site loaded five third party entities.

The study found something more useful than the median, though. Businesses marketing across many Florida markets at once loaded seven, against five for operators serving one airport, and they scored measurably worse on performance for it. That difference survived a permutation test at p equals 0.026, which in plain terms means it is very unlikely to be noise.

The related finding is blunter. Sites built with a page builder made 111 requests against 57 for sites built without one, and took 10.1 seconds to paint their largest element against 6.8. That gap held at better than one in ten thousand. Third parties are not the only reason for it, but they are a large part of how a page gets from 57 requests to 111.

Why the count here is a floor

This tool reads your HTML. That is an honest thing to measure and a limited one, and the limit runs in exactly one direction: it undercounts, and on some sites it undercounts badly.

A tag manager is the clearest case. In your HTML it is one script tag from one company. At runtime it reads a container somebody has been adding to for years and loads whatever is in there, which might be nine more companies. None of them are written in your markup, so nothing that reads markup can see them, including this.

The same applies to anything that loads something else. An advertising pixel that calls three more pixels, a chat widget that pulls in its own analytics, a consent banner that loads a font. The study's median of five was measured by a tool watching the page actually run, which is why your number here being under five is not a clean bill of health.

If this page reports a tag manager, the real work is opening that container and reading the list. Most people have never looked, and most people find at least one thing in there they cannot account for.

The ones in the head are the expensive ones

Position matters more than count. A third party script in the head of your document with neither async nor defer on it stops the browser. Nothing further down the page is drawn until that file has been fetched, from somebody else's server, over a connection that has to be opened first.

A third party stylesheet in the head does the same thing, which surprises people, and font stylesheets are the usual culprit. The browser will not paint text it does not yet know how to style.

This is the cheapest thing on the list to fix, because in many cases it is one attribute. The script that has been holding up your page for two years does not need to run before your page is visible, and nobody ever checked.

A reasonable way to use the result

Do not try to get to zero. Some of these earn their place and a payment processor is not optional.

Go in this order. Anything blocking in the head gets moved or deferred first, because that is the biggest effect for the least work. Then anything duplicated, where two companies are measuring the same visits. Then anything nobody can name a current use for, which is where the abandoned experiments live. Then fonts, which can usually be served from your own domain in an afternoon.

What is left is your actual stack, and it should be a short list you could justify one item at a time. If you want the reasoning behind building it that way from the start, the web design and development page makes the case, and count the third parties is the short version.