A percentile locates a value in an ordered distribution. P90 describes its upper region but depends on cohort size and definition.
01STORY BRIEFThe idea in three screens.3 SCREENS
P90 is not a new word for good.
A percentile locates a value in an ordered distribution.
A percentile across ten posts is less stable than it looks on a dashboard. State the calculation method if tools disagree. Compare equally mature series and keep counts visible. A percentile describes observations; it does not guarantee future performance.
When reading change, show several points in the distribution. A stable median with a stronger upper region tells a different story from broad improvement. The first may suggest more positive extremes, but their cause remains to be understood. These summaries should open the content analysis rather than replace it. Show the same observation window and calculation method so apparent movement is not a reporting artifact.
A percentile needs a population
The phrase top ten percent sounds precise until you ask: of what? Your own posts this month, all posts in a format, or a hand-picked list of successful accounts? A percentile is a position inside a defined population. Change the population and the same post can move without receiving a single additional view. Keep the inclusion rules and observation age beside the chart. For small samples, avoid decorating unstable cut points with impressive decimals. A ranked list of actual posts may communicate more than a polished percentile table.
Read the percentile label completely
- 01Population
Name every rule for inclusion.
- 02Age
Measure posts after a comparable elapsed period.
- 03Position
Describe rank without promising the next result.
Editorial model: a way to reason about the process, not measured platform results.
Use percentiles to discuss the range an operation experiences. The lower end helps expose failures that an average can conceal; the middle describes ordinary output; the upper end shows upside. Do not turn those descriptions into guaranteed future bands. If a new cohort contains different countries, account ages or creative formats, it is not automatically comparable with the previous one. Annotate the change and, where possible, show matched groups. The practical benefit is better questions about the shape of results, not a new way to label every post a winner.
“P90 is not a new word for good.”
02FIELD KITFrom concept to practice.2 SCREENS
Document the calculation convention.
Adapt this to your audience, budget and test scope.
Put it to work.
- 01Document the calculation convention.
- 02Show each cohort’s size.
Notes & sources.
The limit. A high percentile in a small series is not a reliable forecast.
Further reading is linked throughout this article. Worked examples explain a process; they are not campaign results.
THE ORGANIC CLUB is an editorial project initiated by TokPortal. That relationship is separate from the evidence behind claims. Editorial signatures and image credits are explained in our policy. Our editorial policy →
