I took on a slow website and fixed it in this order

A slow website, fixed in the order that matters: images, fonts, third-party scripts, hosting. Plus what we couldn't fix.
A client's site was loading in about nine seconds on a phone, on an average 4G connection. The client was paying for ads that directed people to it.
We didn't start with a 20-page audit. We took the page that received the most paid traffic and fixed things in order of their difficulty.
Images, first thing
Six images, all JPG, the largest over 4MB, all displayed at a width of a few hundred pixels. Someone had uploaded them directly from their phone via the admin panel.
We resized them to their actual display size, converted them to WebP, put the dimensions in the code to prevent layout shifts, and everything below the first fold was left to load later. Just this step cut out most of the time.
There's nothing clever here. It's maintenance. And it's the primary cause in most slow sites I've seen.
Fonts
Three font families, with seven weights, loaded from an external server. I kept two weights, moved them to our own server, preloaded the ones used in the first fold, and set a system font as a fallback so the text would appear immediately.
Pleasant side effect: the design didn't visibly change. It's not worth keeping seven font weights for a company website. Nobody on the client's team noticed that five weights were missing.
Third-party scripts
The most unpleasant part, because here you have to say "no" to some things someone asked for.
Found: two analytics systems, a chat, a review widget, a poorly implemented cookie banner, and a gallery plugin that loaded an entire library on every page. The chat had, in the last three months, four conversations.
I removed the chat and the review widget, replaced the gallery, and deferred the loading of the analytics system. The cookie banner remained, but was rewritten.
Hosting
The last step, not the first. Moving to better hosting helps, but it doesn't fix 4MB images – it just delivers them a bit faster.
What we changed: server closer to users, correctly enabled cache, compression. And I removed a security module that was adding delay to every request.
What we didn't fix
An interactive calculator on the services page, made by someone else, which brought in good enquiries but loaded slowly. We decided with the client that it was worth keeping as it was until there was a budget to rewrite it. Not every technical problem gets solved right away, and how much it slows you down depends on how much traffic comes from mobile.
And an honest observation: scores from measurement tools fluctuate between runs. We use them as a direction, not as a grade. What interests us more is whether mobile users reach the form and whether the bounce rate decreases.
If you want to know why your site is slow, we'll look at a page and send you the list in the order we would work.
What we fix first, and how much it costs in working time
The order isn't based on how spectacular it looks in the report, but on how much you gain per unit of effort.
| Intervention | Effort | What usually changes |
|---|---|---|
| Resized and converted images | Small | Most of the gain, for most sites |
| Fonts limited to two variants | Small | Text appears faster |
| Third-party scripts deferred | Medium | Page becomes usable sooner |
| Fixed dimensions for images and banners | Small | Content shifting disappears |
| Unused modules cleaned up | Medium | Less code loaded for nothing |
| Moved to better hosting | Variable | Only matters if the server was truly the problem |
The last row is what many sell as a universal solution. For the sites we've fixed, hosting was the main problem less often than generally believed.
What we measure, and why we don't just look at the score
The score from testing tools varies between runs, especially on mobile. We look at it, but we make the decision after opening the site on an average phone, on a mobile connection, and counting the seconds until we can read something.
A score of ninety on a site that has content jumping under the user's thumb doesn't mean it's fixed.
What we don't touch, even if it would give a few points
We don't remove conversion tracking to gain score points, because then the client wouldn't know where enquiries come from. Nor do we remove brand fonts just for speed, if the company's identity depends on them. We discuss these compromises with the client; we don't make them on their behalf.
What we couldn't fix
On a site with a booking module from an external provider, we were left with a heavy script that we couldn't completely defer. We gained what we could around it and clearly stated the limitation to the client, instead of promising a perfect score.
If you have a slow-loading site on mobile, we'll test it and send you the list in the order we would work.
Want a website that brings in customers?
We'll give you a clear proposal, with price and delivery time, no obligations.
Related service
Business website design
See details and pricing
Written by
Andrei MunteanuHead of Development, NCHDesign
Builds the websites and online stores: speed, stability and an admin you can use without us.
Updated: 3 September 2026
