A retail company develops a machine learning model to predict product demand. The model shows 98% accuracy on the training dataset, but performs poorly when tested on real customer data from the following month. What is the most likely reason for this behavior?