What is the 95th percentile (P95)? How to calculate and read percentiles

What is a percentile?

A percentile tells you where a value sits in a sorted list. The 95th percentile (P95) is the value that 95% of measurements are at or below. The other common ones:

• P50 (median): the middle; half the values are faster, half slower.
• P75: three quarters are at or below it. Google judges Core Web Vitals at P75.
• P95 / P99: the slow tail, the experience of one visitor in twenty or one in a hundred.

A worked example

Twenty page loads from one page (sample data, in seconds), already sorted:

1.1 · 1.2 · 1.3 · 1.4 · 1.4 · 1.5 · 1.6 · 1.6 · 1.7 · 1.8 · 1.9 · 2.0 · 2.1 · 2.3 · 2.4 · 2.6 · 2.9 · 3.4 · 4.8 · 7.2

With the nearest-rank method, the P-th percentile is the value at rank ⌈P ÷ 100 × n⌉:

The mean (2.31 s) is higher than what 13 of the 20 loads actually took, because two very slow loads pull it up. P75 says most people waited 2.4 s or less, and P95 shows that one load in twenty takes almost 5 seconds. That is a different problem to fix than the typical case.

Spreadsheets and tools may interpolate between ranks, so small samples can give slightly different answers. With few samples, percentiles also jump around; read them only with enough data.

Where percentiles are used

In 95th percentile billing, usage is sampled every 5 minutes. A 30-day month has 8,640 samples; the top 5% (432 samples, about 36 hours) are discarded and you pay for the highest remaining one, so short bursts are free.

Percentiles in TapCub

In Web analytics › Monitoring › Performance, the metric cards switch between P95, P75, median and average, and the trend and breakdown can use P50, P75, P90 or P99. Start with P75 to see whether a page passes, then switch to P95 to find the slow tail by device or country. See Web Vitals.