Maintaining data quality is a major challenge when handling millions of product offers in an online shop. Given that our suppliers have their own product type hierarchies and choose to name brands and properties differently, a lot of manual effort goes into mapping and curating this data to our shop’s data model. Machine Learning can support this process and significantly speed up product offer generation. In this talk, we demonstrate a convolutional neural network model for categorizing new products into one of ~2500 product types, using images, descriptions and other attributes. Furthermore, we show how we run and retrain this model in production alongside other machine-learning assistants, and how the model affected the business workflow since its Go-Live.