Ask Alidade from Slack and Linear.
Research
Seasonality: the anomaly that isn’t
A dip that returns every year is not news. Compare each week to the same week last year before you raise an alarm.
Ruth Okonkwo
·
In the third week of December, one of our customers saw signups drop 31% against the week before. Three people asked in Slack whether something had broken. Nothing had. The same drop happened the year before.
Seasonality is the part of a metric that repeats on a calendar: weekdays, month ends, holidays, school terms. It is predictable, so it should never page anyone.
Compare like with like
A week-over-week comparison treats every week as the same. They are not. Compare each day with the same weekday over the last eight weeks, and each week with the same week a year ago when you have the history. Against that baseline, the December drop was 4%, well inside normal.
With less than a year of data you cannot see yearly patterns. Say so, and widen the range of what counts as normal until you have more history.
Keep the real anomalies
Removing seasonality also reveals moves the calendar was hiding. A 6% lift in a week that is usually 10% down is a bigger story than the raw number suggests.
In 400 flagged moves we reviewed, 112 were fully explained by the calendar once we adjusted for it. Those are the alerts a team learns to mute. Fewer, better alerts get read.
