Let's take a hypothetical example of where you assign a given user a random ID, based on something (machine ID). You then track what this user is doing, in general, in the operating system.
You can find data points such as what % of users discover/use X feature, how many pin applications to the taskbar, use X app with Y app, have night mode enabled, etc.
This allows you to expend resources where they are actually needed, to improve the features that the majority of people would find benefit in.
And at the end of the day, none of the above features require knowing anything about WHO that person is. Sure, they can be classified into groups based on how they use the software, such as "business user", "power user", or "gamer" ... but not "John Doe at 123 Main Street, Hollywood, California".
This is likely all for product improvement and not for $$$. Why do they subject end-users and not opt-in testers for this? To get real world data at scale.