Comment by const_cast

One hypothetical vector is that you can find out which women aren't menstruating normally, meaning they could be pregnant. If that information makes it into the hands of employers or even recruitment platforms that can manifest as discrimination. Typically, women have to be incredibly careful not to give employers the impression they might be pregnant or might be trying to be pregnant.

There's also angles of chronic illness, for example, you could maybe find out which women have PCOS or endometriosis and would therefore be more likely to request time off work.

In general, we've worked very hard to make sure employers cannot find out this data, going so far as to make it explicitly illegal to ask, even in a voluntary disclosure context. But data tracking complicates the situation, and if there's enough convolution you can get real-world results without technically breaking the law. For example, consider RealPage, which uses data analytics to facilitate collusion in the renting market. When we introduce algorithms and heuristics into systems like hiring, we might unintentionally be using biases in the data against applicants.