Well they do say they calculate propensity scores, but the whole language of the blog post is very hand-wavy. Warning over and over again that something is not causal proof, and then dramaticizing it as if it were proof is just unprofessional.
If you have a fraud model, just show the model and the data and the validation - everything else is marketing fluff.
Given they have enough data, at some point it's perfectly reasonable to have cars with 5+ crashes per 12 months - just because of chance.
This is exactly why statistics was invented, damnit!