> Even on a simple calculator.
Why not? If they're paying people to build this software, they presumably want to make it good. With this data, they can learn all kinds of things to help improve future versions: when/why is it crashing, what features are people using most/least, are there patterns of clicks that suggest people are mis-understanding or mis-clicking certain features, etc, etc.
edit: Another one that's often useful is looking at different usage patterns across countries. For example, if nobody in Israel is using COSINE, then there's likely a problem with the right-to-left internationalization that you should fix.