At a certain point companies employ "experimentation" (mainly A/B tests) to do exactly this. They measure dropoff and other things, and have targets, e.g., "acc

At a certain point companies employ "experimentation" (mainly A/B tests) to do exactly this. They measure dropoff and other things, and have targets, e.g., "acceptable dropoff" or whatever they might term it internally. That small amount of user loss (fraction of a percent to a few percent depending on feature/daily usage/other metric) due to a design is considered acceptable: not that they don't care, it's just not considered impactful enough to the business to matter.