> Because the world is so complex, you’ll often have an answer faster by just building some prototype of the hypothesis you want to test.
> If it catches on, you can always scale later.
I think one of the coolest real-world examples of this in practice is how Facebook runs thousands of mini-experiments with different users. Facebook is known for having one of the best data science departments in the world and when they want to change or add a feature of the website they do it for a small subset of users and see what the response is like. If the users seem to like it they can repeat the experiment with a larger subset of users. If the users don't like the change there's no big risk because only a few users were exposed.
Contrast this to how Snapchat completely changed their UI for nearly every user all at once and got a ton of negative feedback for it. That was a complete UI overhaul which is different than FB and their hundreds of mini-changes but the idea is similar.