In this talk, I’ll discuss rapid iteration in data science. Often our mission is to improve a certain metric. Nobody knows in advance how to do so. To be successful, we need to experiment, to test hypothesis and determine the causal relationships between factors. We’ll explore how rapid experiment iteration is similar to agile software engineering, and also a few ways in which it differs. What if there is no product owner who knows how to prioritize ideas? How do we manage infrastructure building tasks in parallel with experimentation? How do we write tests for pipelines whose data is constantly changing?