We're analyzing some creatives from past advertising campaigns and are looking for the best way to approach. Currently we've broken down each creative into 20+ objective characteristics, for example:
For each campaign we have two main data points: CTR and conversions.
What are some possible ways to approach analyzing this data? We're looking to find out how each feature affects CTR/conversions.
If you are looking to find effect of each feature on CTR then a correlation matrix can help you with it and you can do statistical test to prove it.
But if you want to find hidden causes then it a classical problem of prediction vs causation.
Suppose we find that giving discount help increase our campaign you may conclude that giving discount helps but it may actually happen that sales are usually higher that time of year and so giving more discounts necessarily doesn't help. Causal analysis can help us find hidden causes which will be hard to find.
For doing Causal Analysis you can refer Dowhy package.
