What is seasonality in MMM?

Most businesses have a shape to their year. Retail peaks before Christmas, travel in the summer, a bank around pay days and rate decisions. Seasonality is that recurring pattern, and a marketing mix model has to estimate it on its own, because the marketing calendar usually follows the same shape. A model that cannot tell the two apart credits the November sales spike to the November campaigns.

Seasonality lives in the baseline, as a yearly pattern the model estimates from the history. That is the main reason two years of data is the comfortable floor: the model needs to see each season at least twice to separate it from whatever marketing ran alongside. With six months of history the model can be built, but its view of the season is a prior more than an estimate.

Special days such as Black Friday are handled as control variables on top of the seasonal pattern, because they are sharper than a smooth yearly curve can capture.

Seasonality is not only something to remove. In peak weeks the same spend often converts more easily, which means a channel's response curve is steeper and saturates later than in a quiet month. Where the data supports it, the Odins model lets each channel's curve follow the year, so the plan can say which weeks to lean into and which to hold back rather than splitting the budget evenly by month.

The result for a scenario is a monthly split that follows where the next unit works hardest, with the seasonal pattern and the marketing effect kept apart in the forecast.

• Baseline (base sales): The sales that would have happened without paid marketing. Everything the model credits to a channel is measured above it.
• Control variable: The non-marketing factors the model explains sales with, so they are not credited to a channel. Fewer than you would think.
• Time-varying saturation: Letting a channel's response curve follow the year, so November and July are not forced to share one curve.
• Scenario planning: Forecasting what a budget would return before committing it, and comparing several futures side by side.

Go one level deeper: Time-varying saturation. Data for marketing mix modelling. Or go back to all terms in the glossary.