I started doing data science because it seemed super useful and highly general. It had applications everywhere and could solve all sorts of problems. What could be more awesome than basically predicting the future?
But in reality you only get to predict the future in the limited domains where you already have a lot of data about previous instances of the thing you’re trying to solve. That means that it’s helpful for making incremental improvements to existing systems—targeting ads or pricing plane tickets or dispatching drivers or whatever—but it’s not likely to enable lots of breakthrough new ideas. Instead, it seems likely to be useful mostly once organizations get enough scale that it’s worth it to invest in even the kind of incremental improvements you can get from having better predictive models.