Largely through hypothesis which you then test. Eg you might see a lot of people abandoning your cart. You might guess (perhaps based on some or other evidence,

Largely through hypothesis which you then test. Eg you might see a lot of people abandoning your cart. You might guess (perhaps based on some or other evidence, but it can be a straight guess) that this is because the final price isn't prominent enough on the page. So you to test that you make the price bigger. And if the abandonment goes down then that's ONE data point in support of your hypothesis. You should then re-rest the idea in a different way to make sure, for example by moving the price to a different place on the screen such that it achieves more prominence. And it if abandonment goes down further then you have TWO supporting tests. But abandonment might go up, so you need to then re-evaluate your hypothesis (eg maybe it's some notion of apparent "honesty" you need to confer, not visual prominence). This is the work of the UX designer. It takes time an effort. PMs usually don't have the time or understanding to do it properly, so products and companies often suffer. This is why, personally, I would be more inclined to invest in companies that have actual UX designers working for them than mere jack-of-all PMs. But hey.