Seasonality

Seasonality is the part of a metric's movement that comes from the calendar rather than from anything you did. Business software has a strong weekly cycle because most of its users are at work on weekdays, so a chart of daily activity is mostly a picture of the working week with a product underneath it. On top of that sit slower cycles: end-of-quarter procurement pushes, the summer and end-of-year periods when whole regions stop buying, budget years that reset on different months in different markets, and religious and national holidays that fall on different dates each year and therefore move the shape of the comparison as well as its size. None of this is noise. It is a real and repeating pattern, and the failure is not that it exists but that it gets explained as something else. Three habits make it manageable. Compare like with like: a month against the same month a year earlier rather than against the month before, and full weeks against full weeks, since a four-week month next to a five-week month produces a twenty percent swing that means nothing. Measure experiments in whole business cycles for the same reason, and be sceptical of any test that ran from Wednesday to Wednesday. And separate the trend from the cycle explicitly before drawing a conclusion, because a business can be growing steadily and still post its worst quarter of the year on schedule. The harder judgement is how much history you need before a pattern is a pattern. Two years of data means two observations of any annual effect, which is enough to notice and not enough to be confident, so an annual claim made from one prior year is usually a story about a single event that happened to land in that month. Two practical consequences follow. Hiring and capacity plans built on a peak month will look wrong for most of the year, and targets set as a flat monthly number quietly ask a team to beat the calendar in some months and coast in others.