Sentiment analysis captures consumer sentiment, providing valuable insights for more accurate product sales forecasting. However, it represents an aggregated value that does not distinguish the specific features influencing the evaluation. Topic sentiment is an advanced sentiment analysis approach that goes beyond a general assessment of the emotional tone of a text (positive, negative, or neutral). Instead, it links sentiment to specific topics or aspects within the analyzed content, enabling a more nuanced understanding. This paper applies a sequential strategy that leverages an embedding-based approach for topic classification and sentiment quantification. We integrate topic sentiment analysis to refine the accuracy of sales forecasts and adopt a state-space approach to model and predict sales volumes. Our model is applied to the Fiat 500L, covering August 2012 to December 2018. During the same period, we analyzed a corpus composed of 20,387 tweets.