What is seasonality? Seasonal patterns in traffic and how to adjust for them

What is seasonality?

Seasonality is a pattern that repeats on a fixed calendar cycle. Traffic dips every weekend, sales peak every December, sign-ups rise every January. It is predictable, which is what makes it useful: once you know the pattern, you can stop mistaking it for a change.

Time-series data is often described as trend + seasonality + noise. The trend is the long-term direction; seasonality is the repeating swing around it; noise is everything else.

Four cycles in web data

B2B vs B2C: a sample weekly profile

Average daily visitors by weekday, as an index (1.00 = weekly average, sample numbers):

For the B2B site, a Saturday at half of Tuesday is completely normal. For the store, a quiet Saturday is a real warning.

Measuring a seasonal index

seasonal index = value in a period ÷ average across the full cycle

Average several cycles (for weekly patterns, at least 4–8 weeks; for annual, two or more years) so one unusual week does not set the pattern. Divide a raw value by its index to get a seasonally adjusted value.

Adjusting for seasonality

• Compare like with like. Same weekday last week, same month last year (year over year once enough history exists).
• Smooth with the cycle length. A 7-day moving average removes the weekly cycle completely.
• Use seasonal baselines in alerts. A rule that compares Sunday with recent Sundays will not fire every weekend.
• Handle moving holidays as events. Compare the event window, not calendar dates.
• Annotate. Mark campaigns and holidays so next year's comparison has context.

Seasonality in TapCub

Set the comparison to same period last week to compare each day with the same weekday a week earlier; the previous-period option is better for multi-week ranges. Monitors are built for weekly cycles: the forecast band compares with the last 7 days and the same weekday, and the anomaly condition uses the mean and spread of the same weekday over 28 days. Add annotations for holidays and campaigns, and switch the chart grain to weekly to remove the weekday pattern. See also anomaly detection.