Percentile Analysis in Web Performance
You'll learn why p75 and p95 reveal the real user experience that averages hide.
The Problem with Averages
Imagine measuring page load time across 100 visits. 90 users experience 1-second loads, but 10 suffer through 10-second loads. The average is 1.9 seconds—not terrible, right? But that average hides the fact that 10% of your users are having a miserable experience.
This is where percentiles become critical. They show what percentage of your users experience a given level of performance or better.
Understanding p75 and p95
- p75 (75th percentile): 75% of users experience this speed or faster; 25% are slower
- p95 (95th percentile): 95% of users get this speed or better; only 5% are slower
Why these matter more than averages:
Averages can be skewed by extremes (very fast or very slow loads). Percentiles show you what most real users actually experience. The p75 and p95 tell you about your "typical worst-case" and "edge-case" performance.
Real-World Application
When analyzing Core Web Vitals in the field (like in Chrome User Experience Report), Google uses the 75th percentile to classify your site. If your p75 LCP is under 2.5 seconds, you're in the "good" range. This means 75% of real users get good performance, which is a much healthier benchmark than optimizing for an average that most users never experience.
The p95 is valuable for catching problems affecting your slowest users—often those on poor networks or older devices who need the most help.
Key Takeaway: Use p75 and p95 percentiles instead of averages to understand what the majority and nearly all users actually experience, ensuring you're not ignoring significant groups of frustrated visitors.