Some optimization yes, but this has been already achieved decades ago by simply using driving/accident history that insurances shared among themselves probably

Some optimization yes, but this has been already achieved decades ago by simply using driving/accident history that insurances shared among themselves probably since they went digital. All the consistently bad drivers will easily fall into this. What I talk about is anticipation/prediction, possibly wrong conclusions from data (since we all know data can be pretty bad or incorrectly analyzed), also no way to correct any incorrectly derived bad rating. There are no consumers winning in this scenario, even if it may feel intuitively as such if you are a stellar safe driver. Also there are many second-order effects, ie poor risky people pushed out of insurances, still driving since in US you can't do anything without a car in rural places, still causing accidents but no way in hell to pay back, ever. So we move the losses from private corporations to random citizens caught in some bad luck. I'd say keep the risk at those corporations, they anyhow still manage to earn billions annually, no need to make their life even easier.