A company has a model that accurately predicts item prices on training data, but its performance drops substantially after deployment to production. What steps should the company take to address this issue?