In an approach for conversational interchange optimization, a processor monitors conversations in a communication system. A processor generates a list of topics in the conversations. A processor determines a flow for each topic based on length of time, number of participants, and sentiment associated to each topic. The flow is the progress of interactions and shifts for each topic. A processor builds a statistical model for each topic based on the flow to predict success or failure of a topic progression for an incoming conversation. A processor updates the topic progression for the incoming conversation based on the predicted success or failure.