What is the baseline?

Not every sale is caused by this month's marketing. Existing customers come back, the brand is known, the stores are where they are, and demand has a season and a trend of its own. The baseline is the model's estimate of that part: the sales that would have happened with no paid marketing at all. What marketing adds is measured above it, and that increment is what the model calls the marketing-driven share.

The model decomposes each week's sales into a baseline (an intercept with a trend and a yearly seasonal pattern), the effect of control variables such as price and major events, and the contribution of each marketing channel. The baseline is what is left once the channels and the controls have been explained, which is why a model that over-credits a channel will under-state the baseline, and the other way around. Well-set priors and the forecast-versus-actual check are what keep the split honest.

Always-on activity is the hard case. Paid search and PR run every week, so a data-only model cannot easily tell them from the baseline. A prior on how the team believes those channels work is what gives the model a starting point to refine.

The marketing-driven share depends on how much of the business would happen anyway, not on how good the marketing is. A grocery chain can sit around 95% baseline because people buy food regardless; a business that sells mostly online, with demand created by its own marketing, has a far larger marketing-driven share. A customer who sees a third of sales attributed to marketing is not seeing a verdict, just a description of the business.

In Odins the decomposition is shown week by week, so you can see the baseline, the season, each channel's contribution and what the forecast expected, side by side.

• Channel contribution: What each channel added to sales above the baseline over a period. The decomposition behind every ROAS figure.
• Incremental sales: The sales a channel caused that would not have happened without it. The only kind of sales worth paying for.
• Control variable: The non-marketing factors the model explains sales with, so they are not credited to a channel. Fewer than you would think.
• Seasonality: The yearly pattern in demand that the model separates from marketing, so a December spike is not credited to December's ads.
• Long-term brand effect: The slow lift in baseline demand from sustained brand investment, beyond the weeks adstock covers.

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