A driver safety score only works if the driver believes it. Score someone badly for a night shift in the rain through a rough part of town and you haven't measured risk, you've measured their route. They know it. They tell the other drivers. The programme is dead inside a month.
The system we filed in 2016 scores each event twice. First the magnitude - how hard the braking, how far above the limit, how long the phone call. Then a second factor for circumstances.
That second factor is the spatio-temporal index. It combines time of day banded by historic crash rates, weather at that moment, the crash history of the zip code, whether anyone else was in the vehicle, and seatbelt use. Multiply magnitude by index and you get the event score.
So the same pedal input produces different numbers, which is the whole point. Hard braking at one in the morning in the wet, in a zip code with a bad crash record, isn't the same event as hard braking on a clear Tuesday afternoon.
Vehicle weight comes in separately, further up the chain. A vehicle a thousand pounds above average carries a multiplier, because the same manoeuvre in something heavier is more dangerous to everyone outside it.
Scores are normalised against the fleet rather than an absolute scale - a typical driver lands near 70 - and adjusted for distance, so nobody is penalised for covering 400 km instead of 40.
Braking 0.3. Distracted driving 0.3. Speeding 0.2. Acceleration and cornering 0.1 each.
Most people expect speeding to dominate. It doesn't, and that was deliberate: in our fleet data braking was the stronger crash predictor. My read is that a hard brake is usually the trace of something that already went wrong a few seconds earlier - following distance, or attention.
Safety 0.5, productivity 0.3, fuel economy 0.2.
Including the other two mattered. A score built only from things you lose points for is a punishment system, and drivers treat it as one. Give them dimensions they can win on and the number means something different. The design also set behavioural targets mined from each driver's own history rather than generic ones, and timed rewards to land when they'd register.
US10430745B2 claims priority to May 2016, was filed in May 2017, and granted in October 2019, assigned to Azuga. Nineteen later patents cite it. Seventeen come from BlueOwl and Quanata - the same company, renamed in 2023, a State Farm company building behaviour-based insurance - across telematics insurance pricing, gamified risk interfaces and driver risk badging. CalAmp and Vinli account for the other two.
Inventors: Ananth Rani, Ashwin Sabapathy, Mahesh Kumar, Akash Sharma.