Hi everyone,
In my current project, after careful evaluation, we’ve decided to train our MMM (Marketing Mix Modeling) using impressions data instead of spend data. However, I’ve noticed that the PyMC Marketing framework—particularly the response curves and budget optimization features—is primarily built around spend-based inputs.
Could anyone share guidance on how to best adapt the PyMC Marketing framework to work with impressions data? Any examples or references would be greatly appreciated.
Thanks in advance!
Hi @Shagun_Kala. The docs of the MMM class say that the media drivers could be “impressions, clicks or costs”.
But yes the optimisation is set up around monetary budgets. Depending on your situation maybe you know the $/impression?
1 Like
@Shagun_Kala really good question here. There are two potential options;
-
Convert your impressions into spend before the optimization step. Impressions and spend are typically correlated together so this shouldn’t be too messy in terms using the parameters estimated from the impressions data. However, this is a more complicated step than option 2.
-
Optimize toward impressions then convert into spend after. This is a more cleaner solution, in my opinion.
Typically you can convert impressions into spend like so,
Spend = (CPM * Impressions) / 1000
Where CPM represents the cost an advertiser pays for one thousand impressions.
Both options will require you to have all of the CPMs per channel available. Like @drbenvincent mentioned, do you have access to that information?
2 Likes