Telematics promises to make each person pay the fair price of their own risk. But as pricing individualizes, risk pooling erodes. Behind actuarial fairness lies a choice of society, between knowledge of risk and the solidarity of the group.
Traditional motor insurance rests on a trade-off that few policyholders question, that of paying a premium calculated from group averages rather than from one's actual behavior. A careful driver pays substantially the same rate as a reckless one of the same age, sex and place of residence, because the insurer has, to tell them apart, only indirect variables. Behavioral pricing promises to break this trade-off. By directly observing how each person drives, it claims to make every policyholder pay the fair price of their own risk, and not that of the statistical category to which they belong. The argument is powerful, for it appeals to an intuition of fairness, why should the attentive driver subsidize the dangerous one.
To this fairness is added a second, subtler argument, drawn from the economic theory of insurance itself. Classical insurance suffers from a known flaw, moral hazard, by which the fact of being insured relaxes the policyholder's vigilance, since the consequences of their imprudence are borne by the group. By linking the premium directly to observed behavior, behavioral pricing claims to correct this flaw, by re-establishing an immediate link between conduct and its cost. The driver who knows they are observed and billed accordingly would thus be prompted to caution, which flat-rate insurance, by diluting responsibility into the average, discouraged. This argument is theoretically solid and forms the most defensible foundation of telematics. It says nothing, however, about the effects of this individualization on the solidarity of the group, which form the reverse side of the promise.
This promise meets a fast-growing market. Usage-based insurance, which gathers the formulas of pay-as-you-drive, pay-how-you-drive and manage-how-you-drive, is expected to rise from around 43 billion dollars in 2024 to more than 70 billion in 2030 according to a MarketsandMarkets estimate1. Europe is among the most advanced regions, driven by a confidence backed by the GDPR and by a series of regulatory mandates that standardize in-vehicle data, including the automatic emergency call and the event-data recorders made compulsory on new vehicles2. According to a survey by the European authority EIOPA, around 17% of insurers already market telematics products, and the forthcoming European data regulation should further accelerate this diffusion by formalizing data-sharing rights4.
The technology that makes this promise possible lies essentially in the phone everyone carries. A smartphone's sensors, accelerometer, gyroscope and geolocation, suffice to reconstruct a detailed portrait of driving without any additional hardware. Pay-how-you-drive programs thus measure acceleration, braking, cornering, speed, the times of travel and even the distraction caused by handling the phone at the wheel3. These parameters are aggregated into a driving score that modulates the premium, sometimes in real time, and which the policyholder can consult in an app. The rollout of such programs has accelerated, whether the YouDrive scheme of Direct Assurance, the largest in France, or the partnership between Kia and LexisNexis deployed in twenty-eight countries to integrate behavioral analysis directly into the manufacturer's app4.
The promoters of telematics put forward a benefit that goes beyond pricing alone, that of prevention. By coaching the driver, by flagging their harsh braking or their speeding, the device claims to change behavior and reduce actual claims. Faced with around 1.19 million road deaths each year worldwide according to the World Health Organization, the argument carries, and some studies suggest a reduction in at-fault claims of the order of 20 to 30% thanks to behavioral incentives5. Telematics thus presents itself not as a mere pricing tool, but as an instrument of safety policy, which confers on it a social legitimacy that mere tariff segmentation would not have. It is precisely this dual nature, at once price and prevention, that makes criticism delicate.
It is in the insurance of professional fleets that behavioral pricing meets its most natural acceptance, and the examination of this segment illuminates the whole debate. A company that equips its vehicles with telematics sees a double gain, a premium cut of the order of 20% and a reduction in accident costs of around 19%, without the question arising of an employee's privacy in the exercise of their duties8. The management of driving is moreover the most dynamic segment of the market. This ease in the professional world contrasts with the difficulties of the personal-lines market, and the reason for this contrast is illuminating, for what distinguishes the two cases is not the technology, which is identical, but the legitimacy of the surveillance. To monitor a driver at work falls within the employer's power of direction, to monitor a private individual in their private life engages a far more problematic relationship with consent and intimacy.
The dark side of this mechanics came to light in the United States. A New York Times investigation revealed in early 2024 that the manufacturer General Motors collected, through its connected OnStar service, very fine driving data, acceleration, hard braking, cornering speed and even night-time trips, and sold them to data brokers such as LexisNexis and Verisk, which in turn resold them to insurers6. Many drivers thus saw their premium rise without understanding why, and without having knowingly consented to this collection, enrolment in the program having been obtained through misleading means. General Motors discontinued the service in April 2024 under media and political pressure, several senators having referred the matter to the competition authority.
The American federal authority reached a settlement with the manufacturer, finalized in early 2026, banning it for five years from selling geolocation and behavior data to consumer reporting agencies, and requiring it to offer an opt-out7. Beyond the particular case, this episode reveals a structural flaw in behavioral pricing. The data collected to adjust a premium does not remain confined to the insurance relationship, it becomes a commodity liable to circulate, to be resold and to be exploited unbeknown to the person concerned. The surveillance consented to in order to obtain a discount turns into a surveillance suffered whose perimeter escapes the policyholder. The asymmetry of information, which insurance was meant to correct, reconstitutes itself to the detriment of the one who hands over their data.
This episode brings to light a player long kept in the shadows, the data broker. Between the producer of the data, the driver, and its end user, the insurer, an intermediation industry has interposed itself that aggregates, enriches and resells behaviors on a large scale. These brokers build risk profiles that the policyholder never sees, generally cannot contest and discovers only when their premium rises. The driver most often does not even know of the existence of the file these companies hold on them, even though that file determines their rate. Behavioral pricing therefore creates not only a relationship between the policyholder and their insurer, it feeds a secondary market in personal data whose opacity is the rule, and where the value extracted from behavior escapes the one who produces it.
To grasp the deep stake, one must return to what insurance is in principle. To insure is to pool the risks of a group of whom no one knows in advance who, within it, will suffer the loss. This shared ignorance is not a defect of the system, it is its foundation, for it is what justifies the lucky paying for the unlucky, in the reciprocal expectation of one day being the beneficiaries of the group's solidarity. Insurance is, in this sense, a technology of solidarity built on a veil of ignorance. Yet behavioral data has precisely the effect of lifting this veil. The more finely the insurer knows the risk of each individual, the less ignorance remains to be mutualized, and the less reason the group's solidarity has to operate.
From this results a tension that no technology can dissolve, for it is logical before being technical. As pricing individualizes, risk pooling shrinks by the same amount. At the theoretical limit of a perfect prediction of risk, each would pay exactly the cost of their own expected loss, and insurance would dissolve into a mere personalized savings account, emptied of all solidarity. This movement strikes first the most vulnerable, those whose behavior or life constraints generate a high risk, who would find themselves excluded or priced out of reach. Actuarial fairness, presented as a virtue, therefore has a price, the progressive disappearance of collective protection in favor of an integral individual responsibility. It is the same tension, already encountered in relation to climate risk, between the accuracy of the price and the solidarity of the group9.
This erosion does not proceed only from a choice by the insurer, it can set itself in motion through a market mechanism. As soon as an insurer offers good risks an advantageous behavioral pricing, it draws them out of the mutualized pool, where only the less favorable risks remain, whose average premium must then rise. This rise in turn pushes the best of the remaining risks to join the individualized offers, and so on, until the classical pool contains only the heaviest profiles, become uninsurable at a sustainable price. This phenomenon, the inverse of traditional adverse selection in which the insurer feared attracting bad risks, here sees the good risks voluntarily desert risk pooling. Behavioral pricing therefore does not merely reduce solidarity by a pricing decision, it can undo it through a competitive dynamic that no player entirely controls.
To the erosion of pooling is added a more insidious danger, that of indirect discrimination. Behavioral variables are never purely neutral, they often encode characteristics that the law protects. To penalize night-time driving disadvantages shift workers, who do not choose their hours. To surcharge certain routes or certain areas may overlap with the social or ethnic composition of the neighbourhoods crossed. To price on phone-usage data may indirectly capture a state of health or a disability. The risk is then that a pricing presented as based on behavioral merit alone reconstitutes, by the detour of correlated variables, the discriminations that the law forbids operating directly. The apparent neutrality of the algorithm is no guarantee of real neutrality.
The law attempts to frame these excesses, but with a lag. The GDPR imposes explicit consent, data minimization and algorithmic accountability, at the price of a high compliance burden10. Beyond it, the question of behavioral discrimination extends debates that this series addresses elsewhere, whether the prohibition of gender-based differentiation since the Test-Achats ruling, the non-discrimination requirements of the European regulation on artificial intelligence, or the right to be forgotten applied to algorithmic scores11. The common thread of these undertakings is that they seek to preserve, against the predictive power of data, a space of protective opacity, that is, a right not to be entirely transparent to one's insurer. Behavioral pricing places this right under tension as never before.
At the end of this analysis, behavioral pricing appears for what it really is, not a mere technical advance, but the shifting of a political frontier, the one that separates fairness from solidarity. Technology does not dictate where to place this cursor, it merely makes its setting possible, and finer than ever. A society may legitimately wish to reward virtuous behaviors and make each person responsible, just as it may legitimately want to preserve a broad pooling that protects the most exposed. These two objectives are partly irreconcilable, and the choice between them falls not to the actuary but to the legislator and the citizen. The real question is therefore not whether one can individualize risk, one can technically, but how far a society agrees to do so without renouncing the very idea of insurance.
Several safeguards are conceivable to hold this cursor. The legislator can impose floors of pooling, by forbidding the premium gap between the best and the worst behavior from exceeding a certain ratio, in the manner of the pricing bands already practiced in other lines. It can strictly frame the admissible variables, setting aside those that indirectly capture a protected attribute. It can grant the policyholder a right of inspection and contestation over the data that scores them, and a right to erasure that prevents an old conduct from pursuing them indefinitely. None of these measures removes the tension between fairness and solidarity, but each sets its bounds, and it is in the choice of these bounds that the physiognomy of tomorrow's insurance will be decided. To let the market alone set this cursor would amount to letting the logic of knowledge prevail by default over that of solidarity.
This tension goes beyond motor insurance and foreshadows a vaster dilemma. Insurance was historically a technology of solidarity, which drew its strength from the shared ignorance of the future. The data economy is, conversely, a machine for producing knowledge, which dissolves this ignorance as it advances. The two logics are in fundamental opposition, and behavioral pricing is only the first theater of their confrontation, before health, protection and every domain where individual data will become measurable.
One must measure how irreversible this shift would be. Once risk pooling is undone and the pools dissolved into individual pricings, no market mechanism reconstitutes them spontaneously, for no good risk agrees to join a group where it would pay for others as soon as an individualized offer is available to it. Insurance solidarity, slow to build and inscribed in regimes that are sometimes a century old, is of the kind one restores only by making it compulsory. This is why the present moment, when behavioral data is deploying but the framework is still under debate, is decisive, for it traces a frontier that it will be very hard to move afterwards.
The limiting case where each would be perfectly priced according to their own risk would not be the culmination of insurance, it would be its end, for there would be nothing left to mutualize. To preserve insurance as collective protection will therefore suppose, paradoxically, defending a right not to know everything, and assuming politically that a measure of ignorance is the price of solidarity. The predictive capacity of artificial intelligence, which will be the subject of a forthcoming article in this series, will only make this choice more pressing12.
Underground economy, ransom markets and insurer pricing: how ransomware became a structured industry whose growth is partly financed by cyber insurance payouts.
Insurtech promised to disrupt insurance as fintech had shaken up banking. But one does not disrupt the bearing of risk as one disrupts an interface. The promise was neither kept nor betrayed, it was transformed, from conquest toward infrastructure.
The AI Act is a safety code, not a liability regime. Yet the withdrawal of the dedicated directive has made it, by default, the keystone of liability for artificial intelligence, and compliance with the regulation has become the new frontier of insurability.
The regulator asks a simple question, how much money must you hold to survive your worst year in two centuries. For a book of hurricanes, the actuary answers without flinching. For a book of ransomware, the room falls silent.
The fifty-two coverage one risk at a time. This special issue crosses two. The AI buildout and the sustainability constraint are not two subjects; they are a single point of friction, and that point has a name, the data center.
One analysis with every new publication, on insurance, reinsurance, cyber and AI. No account, no noise.
No spam, one-click unsubscribe, no data ever sold.