The Role of Monitoring Effect in Risk Classification: Evidence from Telematics Adoption

The adoption of the Internet of Things (IoT) empowers firms to collect and analyze individual consumers’ activities to classify customer types based on their actions and behaviors that are traditionally unobservable. Although behavioral tracking has become more prevalent in practice, its effectiveness in classifying customer types has hardly been discussed. We utilize detailed individual-level data from a field experiment conducted by an insurance company and its car rental partner to show that tracking using IoT devices such as telematics induces a monitoring effect that sways consumers to behave differently than usual when they are monitored, which significantly undermines the effectiveness of using temporal behavioral tracking data to classify individual types. More importantly, the magnitude of the monitoring effect is correlated with drivers’ unobserved inherent behavior, the very behavior that the tracking devices target to uncover. This correlation is fundamentally different from the typical heterogeneity in effects driven by observable characteristics. Our exercise demonstrates that even if firms recognize the existence of the monitoring effect, overlooking this correlation can more than double the misclassification rate, rendering the classification strategy less effective compared with the traditional classification approach based on customer profiles and sensitive to customer self-selection. Furthermore, we show that the monitoring effect spills over into the postmonitoring period and manifests through both habit formation and crowd-out effects. The direction and the magnitude of the postmonitoring effect vary across individuals. These findings alert practitioners about the caveats in utilizing temporal behavioral tracking data to classify customer types. Suggestive solutions for firms to properly adjust for the influence of the monitoring effect are also provided.

This paper was accepted by D. J. Wu, information systems.

Funding: This work was supported by the Fundamental Research Funds for the Central Universities [Grant CXTD202403].

Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2022.00286.