In this talk I report on experiences gained from more than five years of extensively designing static code analysis tools— in particular such ones with a focus on security—to scale to real-world projects within an industrial context. Within this time frame, my team and I were able to design static-analysis algorithms that yield both largely improved precision and performance compared to previous approaches. I will give a number of insights regarding important design decisions that made this possible. In particular, I argue that summary-based static-analysis techniques for distributive problems, such as IFDS, IDE and WPDS have been unduly under-appreciated. As my experience shows, those techniques can tremendously benefit both precision and performance, if one uses them in a well-informed way, using carefully designed abstract domains. As one example, I will explain how in previous work on Boomerang we were able to decompose pointer analysis, a static analysis problem that is actually not distributive, into sub-problems that are distributive. This yields an implementation that is both highly precise and efficient. This breakthrough, along with the use of a demand-driven program-analysis design, has recently allowed us to implement practical static analysis tools such as the crypto-misuse checker CogniCrypt, which can analyze the entire Maven-Central repository with more than 200.000 binaries in under five days, although its analysis is flow-sensitive, field- sensitive, and fully context-sensitive.