As data scientists we usually like to apply fancy machine learningmodels to well-groomed datasets. Everyone working on industrial problemswill eventually learn, that this does not reflect reality. The amount oftime spent on modeling is small compared to data gathering, -warehousingand -cleaning. Even after training and deployment of the model, the workis not done. Continuous monitoring of the performance and input data isstill necessary.In this talk I discuss how important data handling is for successfuldata science projects. Each milestone, from finding the business case tocontinuously monitoring the performance of the solution, is addressed.This is exemplary shown on a project, with the goal of improving aproductive system.