Can Telematics Improve Driving Style? The Use of Behavioural Data in Motor Insurance

TL;DR

This paper investigates whether telematics-based coaching can improve driving style by integrating two behavioural data streams: driving behavior and policyholder engagement with a coaching app. Using a PHYD dataset of 498 drivers over 35 weeks, it defines a weekly driving score and four driving-subscores, and measures engagement via app usage; it then analyzes improvement both week-by-week and across the full period, stratifying by initial driving ability. The findings show a correlation between engagement and driving improvement for some groups, but the relationship is heterogeneous and time-dependent, with strong engagement being relatively rare and short app sessions limiting observed effects. The study highlights methodological challenges in defining and measuring engagement and coaching, discusses the risk of second-order data use, and proposes directions to enhance engagement and further explore second-order coaching in proactive insurance models.

Abstract

Motor insurance can use telematics data not only to understand the individual driving style, but also to implement innovative coaching strategies that feed back to the drivers, through an app, the aggregated information extracted from the data. The purpose is to encourage an improvement in their driving style. Precondition for this improvement is that drivers are digitally engaged, that is, they interact with the app. Our hypothesis is that the effectiveness of current experimentations depends on the integration of two distinct types of behavioural data: behavioural data on driving style and behavioural data on users' interaction with the app. Based on the empirical investigation of the dataset of a company selling a telematics motor insurance policy, our research shows that there is a correlation between engagement with the app and improvement of driving style, but the analysis must distinguish different groups of users with different driving abilities, and take into account time differences. Our findings contribute to clarify the methodological challenges that must be addressed when exploring engagement and coaching effectiveness in proactive insurance policies. We conclude by discussing the possibility and difficulties of tracking and using second-order behavioural data related to policyholder engagement with the app.

Figures (8)

• Figure 1: On the left, the critical events displayed on each trip map, such as phone distractions (blue icons), risky manoeuvres (purple and orange icons), and speeding (highlighted as red road segments). On the right, the app homepage where the weekly overall driver score and the four sub-scores are updated after each trip.
• Figure 2: Histogram of all weekly scores.
• Figure 3: On the top, visualization of the number of active users on the telematics application (i.e., with at least one app session during the week) and inactive users (with no app sessions), as a function of the week index. On the bottom, histogram of the computed weekly durations (in seconds).
• Figure 4: Example of simple linear regression (blue line) computed on the weekly scores (in red) as function of the week index $i$, for a user with 25 tracked weeks.
• Figure 5: Scatterplots of data points given by $s_0^{(k)}$ initial scores, on the horizontal axis, and metrics for improvement quantification, on the vertical directions. The blue vertical lines denote the merit-based classes. On the top, the values of $\delta_i^{(k)}$ are given by difference as in Equation \ref{['eq:delta_ik']}; on the bottom, the values of $\delta_i^{(k)}$ are given by ratio.