Training data quality has an enormous impact on the performance of an AI algorithm or machine learning model. Large and ambitious projects also need enormous amounts of data to achieve model performance. Is it better for a given project to opt for a fully hands-on data annotation process, resulting in high quality training data but at a large time and cost investment? Or do time and budget constraints necessitate a more generalized and scalable, but lower quality approach? Luckily, scale and quality are not always at odds. Strategic customizations at different points in the data preparation process can unlock quality at scale. This paper outlines how.