Find Your Real Seasonal Pattern And How Big It Is

How it works

The analysis parses your dates, aggregates duplicates by sum, infers the cadence (daily, weekly, monthly) from the median date gap, and fills small gaps by linear interpolation, every adjustment stated in the report. It then runs an STL decomposition (seasonal-trend decomposition using loess with a periodic seasonal window), splitting the series into trend, seasonal, and remainder. Seasonal strength is max(0, 1 minus Var(remainder) divided by Var(seasonal plus remainder)), banded weak below 0.30, moderate below 0.60, strong at 0.60 or above. Each season gets an index, its percentage above or below the trend level, reported in calendar order, with the peak and trough named.

Use it when you have a dated history of one metric and want to know whether a seasonal cycle is real, how big it is, and which seasons peak and dip: before setting targets, staffing, or stock levels around a calendar.

Not for projecting future values (use the forecast tool), for series shorter than two full cycles, or for irregular event-driven spikes that do not repeat on a calendar.

What data do you need?

A dated history of one metric. For example, monthly revenue:

Minimum 10 rows · Best with 2-6 years of monthly or weekly history, or 3+ months of daily history

What's in the report?

Find out whether a number such as sales or visits has a real seasonal pattern and how big it is. Map a date and the number, and the report splits it into trend, season and noise, names the peak and low seasons, and says how much is just noise. You need a bit more than two full cycles of history, and it describes the pattern rather than forecasting it.