What are the main pitfalls when running experiments on insurance claims data, such as novelty effects, network interference, or sample ratio mismatch?
