How do you manage model drift and performance monitoring in production-grade AI systems for The Boston Consulting Group clients?