This work considers improving the heat recovery of an existing heat exchanger network (HEN) while minimizing the total annualized cost of the retrofit design. The focus is on stepwise approaches that generate, starting from the base case network, a candidate retrofit HEN at each step in order to sequentially retrofit the HEN topology while permitting user intervention during the resolution procedure. Stepwise approaches which consider at most one new heat exchanger addition at each step sometimes fail to find a retrofit option that improves heat recovery. The present study addresses this problem by using the notion of bridge, a thermodynamic concept related to the creation of new heat paths in the network. A HEN superstructure is associated to each bridge and a MINLP model minimizing the total utility consumption is solved to obtain an explicit network topology corresponding to a chosen bridge. This new approach is illustrated on a milk processing case study where the network Pinch method fails to find a retrofit solution. For this example, the diagnosis stage of the network Pinch method cannot yield a retrofit project that improves the base case utility consumption. Using the concept of bridge and the proposed superstructure, an increase in heat recovery is achieved by adding two new process-process heat exchangers in a single step.