Should I build a model in-house, or work with a traditional agency or a modern software vendor for MMM?
We’re biased at Recast being a software vendor but we readily admit we’re not the right solution for everybody. Typically if your ad spend is below $100k a month or you spend only across 1-2 channels, our solution isn’t a good fit. Below this level, you can get started with a manually built MMM with a traditional agency or consultant, and start to learn as a company how to make decisions using MMM as an input.
If you’re looking to do MMM in-house know that building a regularly updated model takes a lot of engineering and statistics resources, so you shouldn’t attempt this unless you already have a team in place. In particular, we’ve found machine learning teams underestimate statistical inference problems and you need someone with deep domain knowledge in marketing attribution.