Core recommendation of the talk: make an informed choice between row-level, schema-level and database-level partitioning as early as possible. The decision locks you in significantly — reworking later is very expensive, while a well-informed early choice minimises losses and rework.