Analytics is supposed to tell you how your site performs, and it is one of the things that makes it perform worse. Every script you add is downloaded, parsed and run on the same device, during the same seconds, as the page your visitor came for. On a fast laptop nobody notices. On a mid-range phone on a train, it is the difference between a page that feels instant and one that does not. This post covers what a measurement script costs, how to see the cost of the one you already have, and how to keep it small, with the figures we hold ourselves to.
What does a third-party script cost a page?
Three costs, in order. The network: the file has to travel, and if it comes from another domain, the browser may open a new connection to that domain first, which means another DNS lookup and TLS handshake. The main thread: once it arrives, the script is parsed, compiled and executed on the same thread that lays out and paints your page and responds to taps. And the runtime: listeners the script attaches to scroll, clicks or page visibility run again and again for as long as the page is open.
Core Web Vitals measure the effects. Largest Contentful Paint (LCP) is how long the main content takes to appear, and a script that loads early competes with it for bandwidth and for the main thread. Interaction to Next Paint (INP) is how quickly the page responds to input, and a long task from a script delays the response. Cumulative Layout Shift (CLS) is how much the layout jumps, and anything injected into the page after it has painted, a banner or a widget, can push the content down. Google calls a page good at the 75th percentile of visits when LCP is under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1.
How do you measure the scripts you already ship?
Open your site in Chrome, open DevTools, go to Network, reload, and filter by the domains your analytics and tags load from. Read two numbers per file: the transferred size, which is what crossed the wire compressed, and the resource size, which is what the browser has to parse. Then count the requests. A tag manager is a small file that loads other files, and the real cost is the sum of everything it pulls in, not the container.
Then switch to the Performance panel and record a page load. Long tasks, anything over 50 milliseconds on the main thread, are marked in red, and the bottom-up view shows which script each one belongs to. Lighthouse and PageSpeed Insights summarise the same thing in an audit about reducing the impact of third-party code. If you want a quicker first look, the free site teardown reads your homepage, detects your stack and shows the exact size of the cookie-free tracker you would use instead, with no signup.
Bytes are not the whole bill
A small script loaded badly can cost more than a larger one loaded well. Load analytics with defer or async so it never blocks the HTML parser. Serve it from a host the browser already has a connection to when you can. And look at what it does after it loads: a script that reads the layout on every scroll event costs INP for the whole visit, whatever it weighed.
The consent banner deserves its own line. Its script is a script like any other, and the banner itself is painted on top of your content in the first viewport. On some pages that large block of text becomes the LCP element, and if it is inserted late, it shifts what is beneath it. A measurement setup that writes nothing to the device does not depend on a banner to run. Whether your site needs one for other reasons is your legal team’s call, and cookie-free analytics covers what Vero leaves out.
How Vero keeps its share small
Vero’s core is 1.2 KB gzipped and counts pageviews, referrers, UTM tags and custom events on its own. Everything else is one of 9 optional modules, and a site loads only the ones it turns on. All 9 together weigh 2.7 KB. Gzip compresses a combination better than its parts, so the size of every one of the 320 valid combinations is measured rather than added up, and the tracker builder shows the measured size of the combination you pick before you install it.
The script is assembled per site at the edge, so the file a visitor downloads contains only that site’s modules, and it is loaded with defer. Our own site runs the core with no modules, which measures 1,232 bytes gzipped. PageSpeed Insights scored its home page 100 for performance on mobile on 3 October 2026, in lab conditions. A lab score is one run on one simulated device, not a promise about yours, which is why the next section matters more.
Measure the effect on real visitors
Lab tests tell you what could happen. Field data tells you what does. The vitals module, 587 B on its own, records LCP, INP, CLS, FCP and TTFB from your real visitors, by page and by device, on every plan. Read them at the 75th percentile, the same way Google does.
The Vero demo shows why per page matters. Across the site, LCP at the 75th percentile is 812 milliseconds, INP 94 milliseconds and CLS 0.02, all comfortably good. But one blog post has an LCP of 3,410 milliseconds while the home page sits at 760. A sitewide average would have hidden it. The likely culprit on a page like that is a heavy image or an embed, not the analytics, and per-page field data is how you find out which.
Set a budget and keep it
Write down how many kilobytes of third-party script a page may load, and check it whenever someone adds a tag. A budget turns a vague worry into a yes or no. Ours is in the code: the tracker’s full bundle has a ceiling of 2.8 KB, and the tracker’s audit fails if it goes over. The next module has to fit, or the budget has to be changed on purpose and written down.
For your own site, start with the question each script answers. If nobody reads scroll depth, do not ship the module that measures it. If a tag was added for a campaign that ended, remove it. And if blocked visits matter more to you than scroll or Web Vitals, count pageviews where no script is needed at all, at the edge or on your server, as ad blockers skip your analytics explains. The install guide has the snippet for every framework.
Unsure of a term? Read the glossary

Vero is built by one person in Spain. I read every waitlist entry and pick the beta accounts myself.