What are some best practices for managing and monitoring machine learning model drift in production?

What are some best practices for managing and monitoring machine learning model drift in production?