Why people do guerrilla usability testing if the sampling is always non-representative?

As far as I know guerrilla usability testing is very widely used by design teams. But why? In guerrilla usability testings the sampling method is called convenience method and it is known as a non-probability sampling (or non-representative). So you can't really say something like "our data shows us that 8% of the population will have problems with the navigation", for instance. And if you can't make such claims, is this kind of usability testing still useful?

Statically speaking the data collected by such surveys should be virtually worthless and could never be treated with the mathematical models usually employed because they don't hold true if the sample isn't perfectly random. And as far as I know, there are no mathematical models for such scenarios.

I get guerrilla usability testings are not expensive and fit agile teams. But if the data isn't good at all and if you can't really used it conclude a thing about the population, why people keep doing it at all? I mean, the data gathered could even be harmful and misleading.