Improving almost anything by George Box blew my mind. As children and even as university students we have been lied to. We have been told that experiments must be carried out one at a time to assign effects. This is patently false. In fact, we miss out on potential interaction effects (think temperature and time in cooking) by using one factor at a time experimentation. In this book he discusses many now old but fundamental techniques by which large improvements in models can be gained with minimal data. It has helped me in my career immensely and made me aware of many usual methods and fields of study.