When choosing a pre-rollout baseline, it is recommended to use data from at least 28 days before the update started, and to exclude abnormal factors such as holidays, promotional campaigns, website redesigns, and server failures.