Problem with insurance companies measuring risk this way is that local government externalises costs of bad road design to the people who are unfortunate to have to drive there.

Makes you wonder why there aren't more insurance companies out there using their data to lobby local governments to fix their road design. They have all the data to find hotspots, and reducing accidents would increase their margin.

Maybe this would require an insurance company to have outsized market share in a specific area so they are the main beneficiary of the improvement

It should also be generalize to when (for example a specific corner during dusk or dawn), and for insurers what would also be an important factor would be what other cars are nearby at the hard-braking event, it's not exactly productive to flag the chicken in chicken-or-dare scenario's.