This project builds a count regression model to predict the number of wine cases purchased by distributors based on chemical and marketing properties of roughly 12,000 commercially available wines. We walk through exploratory data analysis, data preparation including missing value flags and median imputation, and model building across Poisson, Negative Binomial, and Linear regression families. The key finding is that expert star ratings and label appeal dominate the prediction, while chemical variables add little beyond those two. The parsimonious Poisson model is selected as the final model based on AIC and interpretability.