Start with the visitor’s task
Performance matters when a visitor must wait to understand the offer, interact with navigation or complete an enquiry. A fast homepage is not enough if the service page or form stalls.
Measure representative pages and devices. Keep laboratory diagnostics separate from observations of real users. A local desktop test cannot establish how a mobile audience experiences the deployed site.
Use metrics to locate the problem
Google’s Web Vitals documentation identifies LCP for loading, INP for responsiveness and CLS for visual stability. Current good-experience thresholds are LCP within 2.5 seconds, INP at most 200 milliseconds and CLS at most 0.1, assessed at the 75th percentile. These are diagnostic targets, not promises of sales or ranking.
| Symptom | Investigation |
|---|---|
| Hero appears late | Image size, request priority, font and server delivery |
| Menu responds slowly | Long main-thread work and unnecessary JavaScript |
| Layout jumps | Missing media dimensions and late inserted content |
| Form feels frozen | Submission state, network timeout and retry feedback |
Preserve the visual identity deliberately
Animation is not automatically the cause of poor performance. Inspect how it is implemented. Competing animation engines writing the same transform can cause jitter; large blurred layers and continuous work can increase rendering cost. Pause work when it is not visible and respect reduced-motion preferences.
Load private workspaces and article bodies only when needed. Resize images for their displayed role and avoid shipping an entire content archive in the homepage bundle. Give images dimensions so space can be reserved before download.
Evaluate conversion separately
A performance improvement can remove friction, but a before-and-after revenue claim needs actual data and a suitable comparison. Do not apply an unrelated industry percentage to your business forecast.
After a change, repeat the same diagnostic scenario and observe the deployed site. Record the device, connection, page and metric source with the result. Discuss performance work using measured bottlenecks and the critical enquiry journey.
A practical question
Is one Lighthouse run evidence of every visitor’s experience?
No. Record it as a laboratory observation with its page, strategy and time. Field metrics, when available, describe a different population and need their own reporting context.