You've shipped a model and it is now making live decisions. After launch, you need a clear plan to keep performance stable as data, users, and operating conditions change.
What steps would you take to ensure a model remains reliable after deployment?