Hello, @drbenvincent !
Your comment is very relevant. Indeed, we have been and are using Causal Impact (we are looking at CausalPy with much interest, but Causal Impact is just more established for now).
Such methodologies indeed can give accurate insights about impact of ads. But they require that a set of assumptions holds, so you have to set up proper conditions for the experiment. And this is a limitation to scale. It gets very complicated if you have many cities and assume that each of them can have different impact throughout a year.
So we are thinking of a solution that could extract impacts probably less accurately, but could give insight about:
- varying impact of ads across seasons
- diminishing value of ads as market gets saturated.
and confirm the results of causalimpact-like tests
Basically, it is a custom Marketing Mix model (we are also watching pymc-marketing).
Actually, I came to thinking about the model in the question when was thinking about extending CausalImpact-like tests.