Case study · 2024 → 2026
Speed & hosting
The site outgrew its shared hosting. We moved to our own server, then made it fast where it matters: on a phone.
- mobile load time for real users
- −73%mobile load time for real users
- desktop PageSpeed
- 43→99desktop PageSpeed
- mobile PageSpeed
- 16→63mobile PageSpeed
- real-user load time on phones, since Nov 2025
- Goodreal-user load time on phones, since Nov 2025
- Role
- Led the decision, the migration plan and the performance program
- Team
- With the second engineer on my team
- Timeline
- Sep 2024 → Jun 2026
- Stack
- AWS Lightsail, Ubuntu, Nginx, PHP, MySQL, Redis, FastCGI cache
01·The problem
September 2024.
The site had outgrown its shared managed hosting: timeouts were hurting SEO and crawling, and staying meant a much bigger plan.
Outgrown hosting
Staying meant moving to a dedicated plan at a much higher monthly cost, with the same limits on what we could control.
Slow for real users
Mobile PageSpeed sat between 16 and 35, Google's lab test took up to 21 seconds to show the page, and real visitors on phones waited 4.5 s. Core Web Vitals failed.
Weight everywhere
A December 2024 diagnosis found seven causes: heavy resources, unoptimized images, unminified code, too many plugins and pixels, too many requests, no CDN and an over-complex layout.
02·The decision
Not a bigger bill. Our own server.
Instead of paying more for the same shared hosting, we moved to infrastructure we control, then fixed the site itself.
- 1
Our own server
Dec 2024: a dedicated AWS Lightsail instance with Nginx, PHP, managed MySQL and daily backups. Subdomains consolidated.
- 2
Release discipline
Mar 2025: no more editing production live. Preprod, QA and a weekly release.
- 3
A lighter site
The rebuild: one design system, SVG and WebP, video instead of GIFs, and in-house tools replacing third-party plugins.
- 4
Three-stage tune-up
Jun 2026: Redis object cache, rate limits for AI crawlers, then front-end work (deferred scripts, CSS, lazy CAPTCHA, a lighter hero video).
03·Speed
From failing to passing. Measured by real users.
PageSpeed, homepage
DesktopMobileLighthouse lab scores from captures taken since week one. October 2026 is the median of three runs.
Feel the difference
How long a visitor waits for the page on a phone.
Chrome UX Report, real-user p75 LCP on the mobile homepage, played at real speed (the pages are an illustration). Desktop went from 4.8 s in 2022 to 1.3 s.
Today
Desktop
peak, June 2026
Mobile
from 16 in 2022
What's next
Desktop passes every Core Web Vital. On mobile, load time is good but interactivity (INP, 222 ms in October 2026) still fails: that's the next fix.
04·Timeline
Two years, one direction.
- Sep 2024Shared hosting reaches its limit: a much bigger plan, or a move.
- Nov 20, 2024Migration plan to AWS Lightsail.
- Dec 2024Live on our own server.
- Aug 2025A load balancer is proposed for traffic peaks. Optimization made it unnecessary.
- Nov 26, 2025Real-user mobile load time rated “Good” for the first time: 2.1 s, down from 4.5 s.
- Feb 19, 2026The rebuilt site launches on the same server.
- Jun 9, 2026Desktop PageSpeed 99. Blocking time −93%.
More case studies
Every figure on this page is public or my own measurement: Chrome UX Report, PageSpeed captures, git and the site's own content. The company's business data stays private.