Bayesian Marketing Mix Model (MMM) Contribution Chart = Frequentist Marketing Mix Model (MMM) Contribution Chart
Bayesian MMM vendors rely on point estimates too
Bayesian MMM vendors often put forth the argument against Frequentist MMM vendors that Frequentist MMM vendors provide point estimates while they don't.
Is this true?
Well not really
Most Bayesian MMM vendors showcase to clients a contribution chart that is identical to what Frequentist MMM vendors show !!
Bayesians Straying From Core Philosophy
Bayesian MMM vendors stray from one of their core philosophy of 'quantifying uncertainty' when presenting MMM contribution charts.
Ideally, their MMM contribution charts should say something like this for each contribution - "There is 80% probability of Instagram ads having a contribution between 2-4%"
But you see this is a very complex statement and this simply does not fly with C-level people.
The C-level people would retort - "what is it? is it 2%, 2.5%, 3.2% or 3.5%? And what does 80% probability even mean? We can't allocate marketing budgets based on a fluctuating range"
To avoid all these hassles, Bayesian MMM people report the expected value (mean) of the posterior distribution. Hence now the report reads - the Instagram ads contribution is 2.9% (the posterior mean value).
So you no longer have a traditional Bayesian range but just a point estimate.
Killing Mosquito with an Anvil
At this point, both Bayesian and Frequentist MMM give you point estimates.
But the amount of extensive compute, complexity and not to mention politics that one had to bear as part of Bayesian MMM process
is simply not worth it.
Bayesian MMM is like killing a mosquito with an Anvil.
MMM at its core is a parameter estimation exercise. Bayesian MMM parameter estimation exercise is costly, time consuming and most of the time inaccurate.
This is why at Aryma Labs we have always followed Frequentist approaches to build our MMM models. If you are a first time MMM adopter, it would be prudent to go with Frequentist MMM.
Thanks for reading.
For consulting and help with MMM implementation, Click here
Stay tuned for more articles on MMM.