For decades, auto insurance was priced on proxy variables: age, gender, postal code, years licensed, claims history. They are statistically correlated with risk, but only indirectly: they describe who the driver is, not how they drive. Telematics breaks with that logic by measuring actual driving behavior, and in doing so opens the door to usage-based insurance, known as Usage-Based Insurance (UBI).
What telematics is and what data it captures
Vehicle telematics means collecting and transmitting driving data through connected devices. There are three main sources: OBD-II dongles plugged into the vehicle's diagnostic port, factory-installed systems in connected cars, and smartphone apps that use the accelerometer, gyroscope and GPS. Each strikes a different balance between cost, accuracy and friction for the policyholder.
Typical data include mileage, speed, harsh acceleration and braking, aggressive cornering, time of day (for example, night driving), road type and phone use behind the wheel. These signals are used to build a driving score that summarizes the policyholder's behavioral risk profile.
From pay-as-you-drive to pay-how-you-drive
There are two main approaches. Pay-as-you-drive (PAYD) prices mainly on how much a person drives: fewer kilometers mean lower exposure and therefore a lower premium. Pay-how-you-drive (PHYD) goes further and adjusts the price for the quality of driving, not just its quantity. A driver who brakes smoothly, respects speed limits and avoids night driving can earn discounts that the traditional model would never have recognized.
The distinction matters conceptually for the actuary. PAYD refines exposure, a variable that was always in the model. PHYD brings in causal behavioral information, closer to the mechanism that actually generates claims, which in principle improves predictive power and actuarial fairness.
The role of machine learning
The volume and granularity of telematics data make purely manual analysis impractical. This is where machine learning comes in. Raw data streams, sampled at high frequency, are turned into aggregate variables through feature engineering: the distribution of accelerations, the share of kilometers driven in high-risk hours and the frequency of severe events, among many others.
Predictive models are trained on those variables. Generalized Linear Models remain the regulatory and interpretability standard in pricing, but they increasingly coexist with techniques such as gradient boosting or neural networks that capture nonlinear interactions. A common approach is to use complex models to uncover patterns and then carry that knowledge over into more transparent rating structures. Pricing stops being static and becomes dynamic: the premium can be recalculated period after period as new driving information arrives.
Benefits: better risk selection and behavioral incentives
The benefits of UBI operate on several levels. For the insurer, segmentation improves: observing actual behavior reduces adverse selection and sharpens pricing. For the good driver, the promise is a fairer price, aligned with individual risk rather than the average of a demographic group.
- More accurate risk selection: behavioral variables complement, and sometimes outperform, the predictive power of traditional rating variables.
- Behavioral incentives: by tying price to conduct, UBI rewards good driving and can reduce loss ratios, a virtuous effect for both policyholder and insurer.
- Driver feedback: many apps provide reports that help correct risky habits.
- Prevention, not just indemnity: the model shifts the focus toward active risk management.
The challenges: privacy, consent and fairness
UBI is not free of serious tensions. The first is privacy. Collecting location, timing and movement patterns amounts to building a detailed record of the policyholder's life. Handling these data requires a robust governance framework, which in Mexico falls under the Ley Federal de Protección de Datos Personales en Posesión de los Particulares (Federal Law on the Protection of Personal Data Held by Private Parties).
Closely related is informed consent. The policyholder must understand what is measured, for what purpose, how long it is retained and with whom it is shared. Genuine consent, not a checkbox buried in a contract, is a condition for the model's legitimacy.
The third challenge is fairness. Although UBI promises more individualized prices, behavioral variables can act as a proxy for protected characteristics. For example, penalizing night driving could disproportionately affect shift workers. The actuary must make sure the model does not reproduce biases or discriminate indirectly, an issue that organizations such as the Society of Actuaries (SOA) and the Casualty Actuarial Society (CAS) have addressed in their literature on fairness and the responsible use of data in predictive analytics.
Implications for the actuary
UBI is redefining actuarial practice in property and casualty. Mastering claims statistics is no longer enough: actuaries now need to understand data engineering, model validation, interpretability and ethical considerations. Telematics also raises questions about reserving, the stability of variables over time and the comparability of portfolios with different levels of data maturity.
Telematics and UBI are not a passing fad but a concrete expression of insurtech: the convergence of data, technology and actuarial technique. For Mexican practitioners, the opportunity is clear. Adopting these tools with statistical rigor, ethical sensitivity and respect for privacy will make it possible to build products that are fairer, more preventive and more sustainable. For the first time, the price of insurance can learn to drive with you.