When a regional delivery fleet in Ohio received its renewal quote last year, the number was about 18 percent higher than the prior year. The fleet manager, a woman named Dana who had overseen the company's trucks for a decade, expected an increase. Fuel costs were up, parts were harder to source, and the insurer had already signaled that the book was underpriced. What she did not expect was the itemized line that explained a third of the jump: a "hard-braking surcharge" tied to the telematics devices installed in every vehicle two years earlier.

Dana had agreed to the telematics program partly because the insurer offered a discount tied to a driver behavior score. The devices tracked speed, acceleration, cornering, and braking, and the data streamed into a dashboard where she could see each driver's score. The first year, the fleet's scores were middling, and the discount was modest. The second year, scores improved, but the renewal quote included a surcharge that seemed to punish the fleet for events the dashboard had flagged as "coachable moments" rather than violations.

The surcharge was not about speeding. It was about braking. Specifically, it was about hard-braking events, defined as deceleration above a threshold roughly equivalent to a panic stop at city speeds. The rate filing that the insurer submitted to the state regulator justified the surcharge by showing a statistical correlation between hard-braking frequency and claim frequency. The filing was approved, and the fleet's premium went up. Dana's story, which she shared in an industry forum, became a small case study in how telematics can reshape insurance pricing in ways that are actuarially sound but operationally perverse.

The Surcharge That Wasn't About Speeding

Telematics-based insurance, often called usage-based insurance or UBI, has been marketed as a way to align premiums more closely with actual risk. Instead of relying on age, credit score, and accident history, insurers can observe how a vehicle is driven. The promise is that safe drivers pay less, and risky drivers pay more, with a precision that traditional rating factors cannot match. For fleets, the appeal is even stronger: telematics can identify specific drivers who need coaching, reduce fuel waste, and lower maintenance costs.

But the promise runs into a practical problem. Actuaries need to convert continuous streams of data into a single number that can be plugged into a rate filing. That number is usually a behavior score, and the score is built from a handful of variables: speed relative to the limit, acceleration events, cornering force, and braking. Each event is weighted, and the weights are derived from historical data linking those events to claims. The result is a score that predicts, with some accuracy, the likelihood of a future loss.

For Dana's fleet, the score was heavily influenced by braking. The insurer's model had found that hard-braking events were a strong predictor of rear-end collisions, which are among the most common and expensive claims for delivery trucks. The correlation made intuitive sense: braking hard often means the driver was following too closely or not anticipating traffic. The model did not care why the brake was applied, only that it was. A driver who braked hard to avoid a deer, a pedestrian, or a suddenly stopped car was treated the same as a driver who braked hard because they were distracted.

That lack of context is where the surcharge began to feel unfair. Dana's fleet operated mostly in urban areas with heavy traffic and frequent pedestrian crossings. Hard braking was sometimes the safest option. But the rate filing did not include a variable for pedestrian density or traffic congestion. It only saw the deceleration spikes. The actuary who built the model would say the correlation is real, and it is. The fleet manager would say the model is blind to the reasons, and she would be right.

How a Rate Filing Actually Works

To understand why the surcharge was not a random act of corporate greed, it helps to know how property and casualty insurance pricing works. Insurers set premiums so that, over time, the total collected from a group of policyholders covers the claims paid out, plus expenses, plus a margin for profit and risk. The key metric is the loss ratio: the ratio of incurred losses to earned premiums. If the loss ratio is too high, the insurer is losing money on that book of business. If it is too low, the insurer is overcharging and may lose customers to competitors.

Rate filings are the documents insurers submit to state regulators to justify changes in premiums. A filing typically includes a base rate, a set of rating factors, and an explanation of how those factors were derived. For a traditional auto policy, the factors might include driver age, vehicle type, and driving record. For a telematics policy, the factors include behavior scores derived from the data. The filing must show that each factor has a statistically significant relationship to claims, and that the factor does not unfairly discriminate against protected classes.

Regulators review filings to ensure rates are not excessive, inadequate, or unfairly discriminatory. "Unfairly discriminatory" is a legal term that means the rate does not treat similar risks differently without justification. A telematics factor that charges more for hard braking is justified if the data shows hard braking is associated with more claims. The problem is that the data also shows hard braking is associated with urban driving, and urban driving is associated with higher claim frequency for reasons unrelated to driver skill. The filing may not control for that distinction.

In Dana's case, the insurer's filing included a detailed actuarial memorandum. It described the data source, the modeling methodology, and the validation results. The model had been tested on a holdout sample, and the predictive lift was modest but real. The filing was approved by the state, which is typical; most filings are approved without modification, and regulators rarely have the staff to audit every assumption. The surcharge was thus not arbitrary, but it was also not the whole story.

The Mechanism That Rewrites Driving Into Numbers

The telematics device in Dana's trucks recorded a stream of data points: GPS coordinates, speed, acceleration, and deceleration, sampled several times per second. The raw data was processed into events, such as "hard brake" or "rapid acceleration," using thresholds defined by the insurer. Those events were then aggregated over a rating period, typically six months or a year, into a single score.

The score was not a simple count of events. It was a weighted sum, where each event type contributed a certain number of points based on its predicted impact on claims. The weights were derived from a Poisson regression model, which estimated how each additional event per 1,000 miles changed the expected claim frequency. The model controlled for mileage, vehicle type, and territory, but it did not control for route characteristics or time of day. A driver who delivered in a dense downtown during rush hour would accumulate more hard-braking events than a driver on rural highways, even if both drove equally carefully.

The score was then mapped to a premium tier. The insurer had five tiers, from A to E, with A being the best and E the worst. The mapping was based on the score's distribution across the entire book of business. A fleet with a score in the top 20 percent would get a discount; a fleet in the bottom 10 percent would get a surcharge. Dana's fleet started in tier B, which earned a modest discount. After a year of data, the fleet's score slipped to tier D, which triggered the surcharge.

The mechanism was transparent in the sense that the insurer published the thresholds and the weights in the rate filing. But the transparency was illusory. The score was a black box that aggregated thousands of events into a single number, and the number was then treated as a contract term. Once the score set the premium, the score became something to optimize, not just a signal of risk. That is a subtle but important shift.

Why the Surcharge Wasn't Arbitrary

Actuaries are trained to be skeptical of anecdote. A single fleet's experience is not evidence. The insurer's model was built on data from tens of thousands of vehicles, and the correlation between hard braking and claims was statistically significant. In a regression with controls for mileage and vehicle type, each additional hard-braking event per 1,000 miles was associated with a roughly 2 percent increase in claim frequency. That effect was consistent across segments and stable over time.

The validation process was rigorous. The model was tested on data from a different period, and the predictive accuracy was measured using a lift curve. The model did better than chance, but the improvement was modest. The area under the ROC curve, a common measure of predictive power, was around 0.62, which is better than a coin flip but far from perfect. The model could not distinguish between a driver who braked hard to avoid a collision and a driver who braked hard because they were tailgating. Both looked the same in the data.

The confounding factors are well known in the actuarial literature. Route density, traffic congestion, weather conditions, and vehicle load all affect braking behavior. A delivery truck carrying a full load brakes differently than an empty one. A driver in a city with frequent traffic lights brakes more often than a highway driver. The model did not include these variables because they were not available in the telematics data at the time the filing was built. The insurer argued that the score was a proxy for overall risk, and the correlation was enough.

Reasonable people can disagree about whether that argument is sufficient. The National Association of Insurance Commissioners has issued guidance on the use of big data and analytics, cautioning insurers to avoid unintended discrimination and to ensure that models are transparent and explainable. Some consumer advocates argue that telematics scores punish drivers for factors beyond their control, such as living in a congested area. The insurer's response is that the score is a better predictor than traditional factors, and that it gives drivers an opportunity to improve their score by changing their behavior.

What the Fleet Learned the Hard Way

After the surcharge appeared, Dana did what any fleet manager would do: she told her drivers to brake less. The dashboard made it easy to see which drivers had the most hard-braking events, and she coached them to anticipate stops earlier and to leave more following distance. Over the next three months, the fleet's hard-braking events dropped by nearly 40 percent. The scores improved, and the dashboard showed green icons instead of red.

But the accident rate did not drop. In fact, during the same period, the fleet had two rear-end collisions, one of which was serious. The drivers had learned to brake more gently, but they had not necessarily learned to drive more safely. They had learned to game the score. Some drivers discovered that the telematics device did not record hard braking when the truck was not moving, so they would coast through intersections without braking at all. Others found that braking in a curve was less likely to trigger the threshold than braking in a straight line, so they would steer around obstacles instead of stopping.

Dana disputed the surcharge with the insurer. She provided route maps showing that the fleet operated in high-congestion areas, and she argued that the hard-braking events were a reflection of the environment, not driver behavior. The insurer acknowledged the context but said the rate filing did not allow for route-specific adjustments. The only option was to file an appeal, which required a formal review process that could take months. The fleet paid the surcharge in the meantime.

The experience changed how Dana thought about telematics. The technology itself was valuable for safety, but the pricing application was flawed. The score was a proxy, and proxies can be gamed. The insurer's actuary would say that the score was still predictive at the portfolio level, and that the gaming behavior would eventually be reflected in claims data, leading to a model update. But for Dana, the immediate cost was real, and the long-term benefit was uncertain.

Where Telematics Pricing Goes Next

The industry is aware of these limitations. The next generation of telematics models is incorporating more contextual data: weather conditions, road type, traffic density, and even the time of day. Some insurers are experimenting with using smartphone sensors to detect driver distraction, such as phone use while driving. Others are integrating data from vehicle cameras to identify near-misses or to distinguish between emergency braking and aggressive braking.

Parametric triggers offer a different approach. Instead of setting a premium based on a behavior score, an insurer might offer a product that provides an immediate discount if a driver completes a safety course or installs a dashcam. The parametric flood trigger that paid a Kansas City bakery before its basement flooded is an example of how a different kind of trigger can work. In auto insurance, a parametric trigger might pay a benefit when a driver's telematics data shows a sudden deceleration, regardless of fault, to cover a deductible or a rental car.

Artificial intelligence is being applied to refine the scores. Machine learning models can identify patterns in the data that traditional regression models miss, such as interactions between braking and time of day or between acceleration and road curvature. But these models are harder to explain, and regulators are wary of black-box pricing. The trade-off between accuracy and transparency is a central tension in the industry.

The fundamental issue is not the technology but the contract. A telematics score is a summary statistic, and any summary statistic loses information. The question is whether the lost information matters for fairness. The answer depends on who you ask. An actuary will say that the score is calibrated to predict claims, and that is the definition of fairness. A fleet manager will say that the score penalizes drivers for circumstances they cannot control, and that is the definition of unfairness.

Policyholder Options in a Telematics World

For policyholders, the rise of telematics pricing brings both opportunities and challenges. On one hand, a behavior-based score can reward safe driving with lower premiums, potentially saving money for those who drive cautiously. On the other hand, the opacity of scoring models and the potential for unintended penalties mean that policyholders need to be informed and proactive.

One option is to ask the insurer about the specific variables that influence the rate. Insurers may not volunteer the weights and thresholds, but they are often willing to provide a summary of the scoring model upon request. Some state regulators require that rate filings be made public, and policyholders can access them through state insurance department websites. Reviewing the filing can reveal how the score is constructed and what factors are considered.

Another option is to request the telematics score itself. Many insurers provide scores on a dashboard or in monthly reports, but if not, policyholders can ask. Knowing one's score is the first step to understanding how it affects the premium. If the score seems unfair, policyholders can gather evidence to dispute it. This might include route maps, traffic reports, or logs that show the circumstances of a hard-braking event. Some insurers have formal appeals processes, though they can be time-consuming.

Policyholders also have the option to shop around. Telematics scores vary by insurer, and a score that triggers a surcharge with one company might be neutral or even favorable with another. The payroll audit that reclassified a restaurant's kitchen staff as managers and tripled its premium is a reminder that pricing structures can be opaque and arbitrary. Comparing quotes from multiple insurers can help policyholders find a more favorable treatment of their driving data.

Finally, policyholders can choose to opt out of telematics programs altogether. While this may mean forgoing potential discounts, it also avoids the risk of surcharges based on incomplete or misleading data. The decision depends on individual circumstances and risk tolerance. Some policyholders may find that the potential savings outweigh the risks, while others may prefer the predictability of traditional rating.

The insurance market is competitive, and telematics is still evolving. The stop-loss filing that trapped a carrier's whole book mid-year shows how pricing changes can have unintended consequences. For policyholders, the lesson is to be proactive, ask questions, and document everything.