When many factors must be considered for decision-making, people dynamically change their emphasizing points, along with their understanding of these factors and the relationships between them. In previous work, we proposed a method to dynamically estimate emphasizing points (DEEP) based on utterances, physiological indices, and proposal selections. To evaluate this method in actual interactions, we conducted controlled WoZ (Wizard of Oz) experiments using Embodied Conversational Agents (ECAs), which interactively provide controlled information for decision-making. Using ECAs, we compare our method to an existing method, which estimates emphasizing factors through the “gradual method”. We confirm that our method can accurately estimate dynamic changes of emphasizing points, and that participants were more satisfied with the final proposal from the ECA that used DEEP.