Michael Kaminsky explains why most Marketing Mix Models assume a constant ROI across all time periods, which is wrong. Sunglasses ads work better in summer than winter, so ROI varies seasonally. Recast models these temporal patterns using Gaussian processes to capture how sales on a given day resemble the same weekday or date in previous years.
Chapters
Most MMMs assume marketing ROI is frozen in time. That's one of the main reasons their forecasts are wrong. Things change over time. Today's ROI is not going to be yesterday's ROI. And to think about it, let's use a simple example. Let's say you sell sunglasses. You're going to sell more sunglasses in the summer than you are in the winter, and your ROI on advertising spend is going to be higher in the summer than it is in the winter. So your MMM that makes an assumption that you just have a single ROI across all of time for your Facebook spend is absolutely 100%, without a doubt, incorrect. At Recast, we model these relationships over time, and the way that we're able to do it is because time is not totally unpredictable. Thursday sales look like Thursday sales. August 15th sales, 2025, are going to probably look something like August 15th sales, 2024 and 2023. Today's sales are going to be like yesterday's sales. These relationships across time, today's like yesterday, today's like a year ago, today's like a week ago, can be encoded in these mathematical functions called Gaussian processes and other structural ways on coding relationships across time. That's what we use at Recast to capture these structures in time that allow us to make forecasts into the future that match reality.
