You are building a machine learning system that produces predictions used in a user-facing product. Some features change continuously, while others are updated on a schedule, and the team needs to decide how predictions should be computed and delivered.
How would you choose between online and batch serving for a model?