With the increasing enrichment of high-involvement products, such as automobiles and home appliances, designing a recommendation system to assist consumers in choosing high-involvement products has become an important research issue. Focusing on the attributes of high-volume and high value products, as well as multistage of consumer consulting, this paper proposes a modular multistage conversational recommendation method for high-involvement products. The proposed method adopts the paradigm of “system query-user answer” to obtain user preferences through questions and generate recommendation results based on user answers. For the issue of multistage consulting, the proposed method introduces the state variable of the stage to the reinforcement learning algorithm. To conquer the contradiction problem between the tasks of preference acquisition and product recommendation, this paper constructs a modular conversational recommendation system.The system includes three components:A dialogue strategy based on reinforcement learning, an attribute selection method based on reinforcement learning, and a product selection method based on knowledge graph and ideal point method. Experiments based on a real purchase dataset on a well-known Chinese auto forum and on simulated user interaction data indicate that, compared with the benchmark method, the proposed method can achieve higher recommendation accuracy with fewer interactions.