What is channel contribution?

A trained marketing mix model can take any week of sales and split it into parts: the baseline, the effect of each control variable, and the contribution of each marketing channel. Summed over a quarter or a year, the channel contributions say what each activity added to sales beyond what would have sold anyway. Divide a channel's contribution by its spend and you have its incremental ROAS for the period.

Drawn as a stacked chart over time, the decomposition shows the baseline and its seasonal shape, with each channel's contribution layered on top. TV and radio contributions spread over the weeks after a flight because of carryover; search sits close to its own spend. The marketing-driven share, all contributions over total sales, describes the business more than the marketing: a grocery chain can sit near 95% baseline, a business that sells mostly online far lower.

Contribution is a period figure, so it is an average reading. It says what a channel delivered, not what more spend would deliver; that is the marginal figure, read from the saturation curve.

The model estimates each channel's effect on sales, not the relationships between channels. When heavy TV lifts Meta sales in the following weeks, that lift is credited to TV, because the model sees TV spend followed by higher sales. Channels that often run at the same time are separated by the natural variation in the history, which is also why a large reallocation is best treated as next year's deliberate test rather than moved all at once.

In Odins the decomposition is shown week by week in the platform, next to the forecast the model made for those weeks, so contribution and accuracy can be read together.

• Baseline (base sales): The sales that would have happened without paid marketing. Everything the model credits to a channel is measured above it.
• Incremental sales: The sales a channel caused that would not have happened without it. The only kind of sales worth paying for.
• ROAS: Return on ad spend: revenue per unit of spend. Useful, as long as you know which ROAS you are looking at.
• Adstock: The carryover effect: the part of this week's advertising that lands in the weeks that follow.
• Control variable: The non-marketing factors the model explains sales with, so they are not credited to a channel. Fewer than you would think.
• Forecast verification: Forecast first, compare with actuals afterwards. The only honest way to know whether a model is right.

Go one level deeper: Reporting and insights. How the model works. Or go back to all terms in the glossary.