Offering differentiated services in the electric power grid is a powerful approach to reveal as well as learn the customer flexibility in response to economic incentives.  
To manage this task a load serving entity needs: 1) a decision model for
measuring ex-ante the expected additional revenues that can be obtained by relaxing service constraints; 2) a model  for computing the profit maximizing prices for these services, 3) a model for presenting its offers to customers and, 4) a model for scheduling the service. 

In this talk we discuss a scalable architecture for the load serving entity that is based on clustering flexible loads and combining them in a hybrid stochastic model that retains the individual customer constrains.