Abstract: A conversation control system and associated methods adapt or modify currently active or future conversations and/or interactions based on machine-generated models that continuously evaluate client satisfaction over the entirety of the active conversations and/or interactions. The system receives a conversation with utterances by two or more participants, and generates a vector with features comprised of words, word sequences, and category classifications associated with at least one of the utterances. The system provides the features as inputs to a prediction model that assigns weights to a set of features based on a modeling of different impacts that each feature of the set of features has on one or more aspects of customer satisfaction. The system generates predictions for those aspects of customer satisfaction without input from the participants based on the assigned weights, and performs an automated action based on the one or more predictions.