A South African short term insurer wanted to apply analysis of telematics data to generate insurance risk insights based on vehicle movement patterns.
We processed over a billion records from several months of telematics data, for a portion of a telematics company fleet, and performed fine-scale spatial-temporal trajectory and cluster analysis, as well as some novel applications of latent Dirichlet allocation statistics. We generated a risk cube that supported assessment of insurance risk for individual vehicles and cohorts of vehicles.
Our analysis produced a set of useful variables that allowed the insurer to gain fine-grained insight into their customers behaviours that could introduce insurance risk.
Some examples included persistent late night driving, persistent speeding in different speed zones and frequent movement into and through high crime areas.