1. Need to remove all obvious confounding variables to arrive at a true model. We can automate this by running multiple times and then checking to see the feature importance 2. For precondition and conversion events need to check the first day that those events occurred so that we are not including datapoints where they were not even sending those events.