Under The Hood Part 3: Adstock and Halo impact/Spillover effect
Christopher Engman, CMO/CRO at Proof, explains two common regression analysis terms: Adstock and Halo impact/Spillover effect. What are they telling you and how are they working? Why should these be included in your marketing strategy?
What is Adstock?
Adstock is how the cumulated brand value and the accumulated campaign value is carried over into a following period. So you’re promoting, let’s say a BMW x5. How is that having a spillover effect on the BMW x3? So Adstock is a way to quantify how much of your previous campaigning and marketing can be carried over into the next period.
Two versions of Adstock: a simplified version and a complex version
There are two different versions. There’s a simplified version and there is a complex version. The complex version is obviously better in this case. In the simple one you take the quantified sales for the previous period, and carry that over to the up-coming period. In reality this has a lower number than one. Less than 100% of the previous month is pulled over to the next month, this term is called lambda. Lambda can be everything between 0 and 1, which corresponds to the range between 0-100 percent. That is the amount that can be carried over into the following period. To quantify and to run real marketing mix modeling lambda has to be introduced.
So that’s the simplified version, the more real version is similar. The sales in period two equals the full previous function. So all of the variables because the spend on retargeting might have a smaller carryover effect into the next period, than great videos. Lambda for this specific variable can be 0.1 so 10% of the spend is carried over into into the next period whereas for great videos 90% is carried over to the next period. You have individual lambdas for each variable, because they are different and this gives you more accurate modeling. Again it’s between zero and hundred percent of the effect from the previous month that can be carried over into the next month and that is being taken into account when doing the modeling to make it more accurate and more precise with less error.
The Halo Impact or Spillover Effect
Then there is halo impact or spillover effect. In large companies you have a big portfolio of various services and products. And typically you’re not promoting only your company all the time. You’re promoting the BMWs x5 for example. The promotion of the x5 can have a spillover effect into the sales of x3. We introduce the halo impact or spillover effect into the modeling to make the modeling more robust.