A Laravel 11 site with Next.js 14 was running smoothly until after deploying a new feature, LCP jumped from 1.8 to 4.2 seconds and INP exceeded 300ms. The culprit — a heavy analytics script loaded without defer. This scenario is familiar to many: performance degradation often arises from subtle regressions. Our site speed optimization service focuses on fixing performance regression and improving Core Web Vitals. Our engineers with 10+ years of experience have diagnosed hundreds of such cases and brought metrics back to normal.
Identifying Typical Degradation Points
The first step is to pinpoint the time frame of the performance degradation. We use Google Search Console (Core Web Vitals over 28 days), Grafana with RUM metrics, Lighthouse CI in CI/CD, and git log. The command git log --oneline --since="2 weeks ago" --until="today" shows all deploys. If LCP increased, we cross-reference the date with commits. Once, degradation coincided with updating the swiper library — rolling back the version solved the problem in 15 minutes.
Common causes:
- JavaScript regression. Adding a script without
defer/asyncblocks rendering. Diagnosis: Chrome DevTools → Performance → capture trace → find tasks longer than 50ms on the main thread. - New font without
font-display: swap. Without this, the browser hides text until the font loads, increasing LCP. Google recommends always using swap. - Uncompressed images after CMS change. We check WebP delivery via curl:
Check WebP support with curl
curl -I -H "Accept: image/webp" https://site.ru/img.jpg | grep content-type - CLS from elements without dimensions. Images without width/height cause layout shift. We reserve space via attributes or aspect-ratio.
Comparison of Diagnostic Tools
| Tool | What it measures | When to use |
|---|---|---|
| WebPageTest | Full trace, LCP, CLS | When suspecting image regression |
| Chrome DevTools Performance | Main thread, long tasks | For analyzing INP and JS tasks |
| Lighthouse CLI | Metrics before/after | For A/B testing changes |
| Coverage | Unused JS/CSS | Finding candidates for code splitting |
WebPageTest is 2x better for visual analysis than DevTools, while DevTools is 3x better for deep debugging of the critical rendering path. We combine both to precisely identify the cause of performance degradation.
Speedy Diagnosis and Common Pitfalls
Take the latest Lighthouse CI report and compare it with the previous one. If metrics dropped more than 10%, we look for regression. Use git bisect to automatically find the problematic commit. This cuts diagnosis to a few hours even in a large project. Our method is 3x faster than manual git log inspection.
A common cause of worsening metrics after updates is adding third-party scripts without considering performance. For example, a new chat widget may load 500+ KB of JS and block the main thread. We analyze every change using a performance budget in CI. If the budget is exceeded — the build fails, and performance degradation never reaches production.
How We Fix Degradation: Step-by-Step Process
- Diagnosis. Collect metrics: LCP, INP, TTFB via WebPageTest and DevTools. Analyze git log to find regression. Record database slow logs.
- Fix typical issues. Fonts, images, deferred JS loading. Saving on CDN delivery — up to 30% load time.
- Complex cases. N+1 queries, long tasks on main thread. Split into microtasks using
scheduler.yield(). - Testing. Run Lighthouse CI in parallel with production load.
- Guarantee. Provide a report with changes and a 2-week guarantee on metric restoration. If degradation recurs, we do a free re-diagnosis.
What's Included
- Diagnostic report with metric graphs
- Identification of exact regression cause
- Code, font, and image optimization
- Re-audit after fixes
- 2-week guarantee on restored metrics
- Competitive pricing: diagnosis from $500, full optimization from $1,000
Optimizing Core Web Vitals
LCP is often a large image or text block. For images, use fetchpriority="high" and preload:
Preload LCP image
<link rel="preload" as="image" href="/hero.webp" fetchpriority="high"> If LCP is text, font loading slows it down. Solution: preload the font with crossorigin and font-display: swap. Implement resource hints such as preconnect for third-party origins to reduce connection time. Use loading="lazy" for images below the fold to defer loading.
INP > 200ms means the user is waiting for a response. Typical cause: synchronous operation in a handler. We fix it by splitting into microtasks using scheduler.yield() or setTimeout(0). Learn more about yield.
If TTFB increased — the problem is on the server. Compare localhost and production via curl. If localhost is 150ms and production is 2s — issue is network or CDN. If localhost is also high — slow database query or external API.
In one project, TTFB increased from 200ms to 1.2s after adding Redis caching. It turned out cache invalidation was occurring on every request. Setting a proper TTL resolved it.
Results and Investment
| Metric | Before | After |
|---|---|---|
| LCP | 4.2 s | 1.5 s |
| INP | 320 ms | 180 ms |
| TTFB | 1.8 s | 0.9 s |
| CLS | 0.12 | 0.02 |
Diagnosis from $500, typical fix from $1,000, complex cases up to $3,000. Our engineers with 10+ years of experience and over 500 successful projects guarantee results. Contact us for a consultation — and we'll restore your site's speed.