I've both worked on systems like that professionally and built some myself just for fun to play with GPS/geospatial data/streaming etc (fun tech). I'd _never_ want something like that installed in my vehicle.<p>First off, you're usually relying on ODB-2 + GPS + usually some type of streaming backend, which means you're looking at 2 unreliable (or, at least, hard to _interpret_) sources of data delivered and presumably analyzed by a notoriously tricky delivery mechanism (near-real time streaming), which has a tendency to be too much to handle for the notoriously non-"tech" insurance world. The chances that data is being interpreted quite differently to what's happening on the road, but yet is treated as a source of truth for your premium model!, is certainly not zero.<p>Even if you assume these companies handle all that properly (or outsource it), these data sets are some of the creepiest by their very nature - not only can you trivially determine somebody's home address by analyzing frequency of the reverse-geocoded data points (or any other geospatial features you derive), you can also determine any other patterns within a person's life - schedules, work location & employer, health habits (do they go to the gym or a bar? do they only ever leave to go to the grocery store?) etc., not to mention all the data ODB-2 gives you about actual driving behavior. And guess what, you can't use synthetic data for testing for much of this (I tried), so we would up testing these systems with real data, albeit with employees who volunteered.<p>A home address is arguably not that sensitive, given that that is public record if you own your home and your insurance company has that data anyways, but the patterns + all the other metadata + all the "public" data that is out there ready for purchase about the average American, now you have something close to a personality profile that would make your average marketing exec (or, potentially, bad government actor) squeal in delight.