 In data science, data scientists usually explore complex databases in order to extract features which are then used as input to machine learning models. This process requires a lot of expertise, including in statistics, data mining, data visualization, machine learning, in addition to domain knowledge. It is known to be the most time-consuming task in data analytics projects. At IBM Research Ireland, we are working on a project called the One-Button Machine that allows people to use data science easily even without deep knowledge of data science. The team has participated in many data science competitions in order to learn how data scientists extract values from raw data. They have identified the key tasks that can be automated. The first version of the One-Button Machine system can automatically explore data by joining database tables and searching for transformations of the joined data that lead to features that are highly correlated with the prediction target. It is implemented on a big data platform that scales the databases with hundreds of gigabytes of data. The One-Button Machine has been validated in public data science competitions where it has outperformed 76 to 84 percent of participants and with only a small prediction accuracy gap to the winners. We are continually improving the One-Button Machine system and our next target is to start winning the competitions we enter.