Lazy loading is a trade: you move work out of the initial load, but you add a request later. It works for below-the-fold images, heavy widgets and third-party embeds. It backfires when the deferred element is the hero or a core interaction.
Key takeaways
- Target: Defer non-critical work without delaying the first meaningful view.
- Measure field data before choosing a fix.
- Change one bottleneck at a time and re-measure on a mid-range phone.
- A fast lab score is not a substitute for real-user data.
What Lazy loading measures
Use the network waterfall and the LCP element to see what loads early. Check whether a lazy component causes a visible delay or layout shift when it appears.
The fixes that usually move the number
- Lazy-load images below the fold with
loading="lazy"ornext/imagedefaults. - Dynamically import heavy client components that are not needed on first paint.
- Load third-party embeds only after consent or interaction.
- Keep the hero image, primary heading and critical CSS eager.
- Provide a stable fallback to avoid layout shift.
Symptom, cause and fix
| Symptom | Likely cause | What to do |
|---|---|---|
| Hero image late | Lazy-loaded hero | Remove lazy loading and mark it priority |
| Widget appears with a jump | No reserved space | Use a sized skeleton |
| Chat loads on every page | Global dynamic import | Load after interaction or on the support route |
| Video blocks the page | Autoplay iframe in the first viewport | Use a poster and load the iframe on click |
How to measure it properly
- Find the initial viewport. List the elements needed for the first screen.
- Check the waterfall. Identify lazy requests that start too early or too late.
- Test interaction. Open the deferred component and measure the wait.
- Check CLS. Confirm the fallback reserves the same space.
- Compare field data. A local win that does not move LCP or INP is not a priority.
A practical example
import dynamic from "next/dynamic";
const SupportChat = dynamic(() => import("./SupportChat"), {
loading: () => <div style={{ height: 64 }} />,
});
export function SupportButton() {
const [open, setOpen] = useState(false);
return open ? <SupportChat /> : <button onClick={() => setOpen(true)}>Chat</button>;
}
Common mistakes
- Lazy-loading the LCP image or the font used above the fold.
- Loading a heavy widget on every page when it is used once.
- Forgetting a fallback and causing a layout shift.
- Deferring a component the user needs immediately.
A 30-day improvement plan
- Week 1. Collect field data and pick the worst route. Write down the baseline.
- Week 2. Fix the largest bottleneck and deploy behind a normal release.
- Week 3. Test on a mid-range phone and a throttled connection; fix regressions.
- Week 4. Compare field data, document the change and add a monitor or budget.
Field data vs lab data
| Source | What it tells you | How to use it |
|---|---|---|
| Field data | What real users experienced | The source of truth for the metric |
| Lab test | A reproducible diagnosis | Find the cause and test a fix |
| Synthetic monitoring | Availability and regressions over time | Alert on a budget breach |
| RUM segmentation | Device, country and route differences | Target the worst segment |
| Server timing | Database, cache and render time | Separate backend from frontend |
Set a performance budget
| Budget | Example starting point | Why |
|---|---|---|
| LCP | 2.5s or less at p75 | The main content must appear quickly |
| INP | 200ms or less at p75 | Interactions must feel responsive |
| CLS | 0.1 or less at p75 | Layout movement erodes trust |
| JavaScript | A route-level KB budget | Prevents bundle creep |
| Images | Responsive sizes and modern formats | The largest common payload |
| Third-party scripts | A named owner and a cap | Uncontrolled tags are a common regression |
Third-party script policy
- Every script needs an owner, a purpose and a review date.
- Load analytics and marketing tags after consent and after the main content.
- Prefer server-side events where they reduce client JavaScript.
- Test the checkout and sign-up with the full tag set enabled.
- Remove tags that no longer answer a business question.
Frequently asked questions
The target is Defer non-critical work without delaying the first meaningful view. A single fast test is not enough; use the 75th percentile of real page views and segment by mobile and desktop.
No. Start with the routes that matter to revenue or support: checkout, pricing, sign-up, the main product page and the most-read blog template. A fast homepage with a slow checkout is not a fast site.
It depends on traffic. Low-traffic sites can take weeks to accumulate enough data. Use lab tools for immediate feedback, but keep the field metric as the source of truth.
Often yes. Image dimensions, priority, bundle splits, caching and third-party scripts usually deliver more than a visual refresh. Redesign only after the measurement shows the layout itself is the bottleneck.
Sources and further reading
- Lazy loading (web.dev)
- Next.js lazy loading guide (nextjs.org)
- Web Vitals (web.dev)
Related reading
Need a performance audit?
Send the URL and the metric that matters. I will identify the bottleneck, the likely fix and the cheapest way to prove it worked.
How to cite this guide: Nur Hossain Repon, “Lazy Loading Optimization Guide for Next.js and React Apps”, Nhrepon Portfolio, reviewed October 2026, https://nhrepon.com/posts/lazy-loading-optimization-guide-for-next-js-and-react-apps.





Comments
No comments yet.