Brzina · Studija slučaja

How We Sped Up Our Own Website: Lighthouse from 69 to 96–99 (Concrete Steps)

2026-10-23 · ~8 min read

Website speed is a topic everyone discusses and few show concrete steps for. So here we openly show what we did on our own site astralogic.rs, with measurements before and after.

Starting point

Measured with Lighthouse (mobile, simulated slower network):

  • Home page performance score: 69
  • Accessibility: 93
  • Layout shift (CLS): 0.12 to 0.20
  • Simulated LCP (largest element): about 7 seconds
  • Home page weight: about 864 KB

An important note: Lighthouse is a lab test. The real first-paint time on a fast connection was already good (about half a second), but the slow-network simulation exposed real problems that would hit visitors on a weak signal.

What we changed

1. A 4.3 MB logo

One client logo in the "who we worked with" strip was 4.3 MB although it is displayed 40 pixels tall. We converted it to a correctly sized WebP (about 1.4 KB), did the same for the other logos and gave them dimensions so the layout does not shift.

Lesson: the biggest wins are often in one image nobody thinks to check.

2. Icons: from 258 KB to 9 KB

We used an icon library from an external server, but we only need about 70 icons. We built a font subset with just those icons and host it ourselves. Font weight fell from about 258 KB to about 9 KB, with fewer external requests.

3. Fonts hosted on our own server

We no longer load fonts from an external server but from our own, with preload for the ones needed immediately. For Serbian letters (č, ć, š, ž, đ) we extracted only the needed range, so one file fell from about 85 KB to about 15 KB. We also added a fallback font with matching metrics so text does not "jump" when the real font loads. The result is layout shift (CLS) equal to zero.

4. The above-the-fold image and animations

The main home image was a CSS background, so the browser discovered it late. We added a high-priority preload. On phones we turned off the costly background animation and the title entrance animations, because they delayed the first paint.

5. Caching and file versions

Static files got long caching, and CSS and JavaScript carry a version marker in the URL, so visitors always get the new version as soon as we change it. Without that, old cached scripts could break a new feature.

6. Accessibility as a side win

We fixed heading order, slider button sizes, link descriptions and screen-reader labels. The accessibility score rose from 93 to 100.

Results after

  • Mobile performance: 96–99 on most pages (the home page ranges 86–99 across runs)
  • Accessibility, best practices and SEO: 100
  • CLS: 0
  • Simulated LCP: about 2 seconds
  • Home page weight: about 600 KB

What this does NOT mean

  • Lighthouse is not the same as real users. For the real picture, look at the Core Web Vitals report in Search Console, which uses data from real visits. Our site does not yet have enough traffic for that data, so we do not claim it is confirmed.
  • Lab results vary from run to run. The same code on the home page gave us 86 and 99, so measure several times.
  • Speed is one factor. It does not replace useful content and good links.

How you can apply this

  1. Measure the page in PageSpeed Insights and read the "Opportunities" list.
  2. Find the largest images and convert them to correctly sized WebP.
  3. Reduce external requests (fonts, libraries, scripts).
  4. Prioritise the above-the-fold image and give all images dimensions.
  5. Set up caching and file versions.

For background on what Google measures, read Core Web Vitals on web.dev. If you want someone to do this on your site, it is part of our website development, and you can check the earlier steps in our SEO audit checklist. Why speed affects sales is also covered in why your website isn't bringing in clients.

Want a speed review of your site? Get in touch.

Related articles

Useful sources

← Blog