Why insurers want your driving data: usage-based economics

Insurers want driving data because it is the closest thing to a direct measurement of the thing they are pricing. Traditional rating works by grouping drivers on stand-ins for risk, such as age, ZIP code and claims history, and charging each group an average. Telematics replaces some of that guessing with how a particular person brakes, accelerates, drives at night and uses a phone. That has three payoffs for the insurer: more accurate prices, a way to attract safer drivers before competitors do, and some evidence that drivers behave better when watched. The sign-up discount is the price the insurer pays to obtain the data, and it only works if the data is worth more than the discount.

The case is real but narrower than the advertising suggests, as the evidence below shows.

The pricing problem: averages and proxies

An insurer cannot see how carefully you drive when you apply. So rating relies on characteristics that correlate with claims. A 2015 study by the NAIC's Center for Insurance Policy and Research, written with regulators and an actuarial consultant, puts the history plainly: for most of the time auto liability insurance has been sold, premiums were set, "in the absence of true causal data," by "a variety of group behavior-based demographic proxy factors affecting loss costs, such as driver record, age, gender, marital status and residence geographic location known as territory." Credit scores, education and occupation were added later, and the study notes their use "is controversial and restricted in a number of jurisdictions." The NAIC CIPR study, dated March 2015, is the backbone of this section and of several below.

The trouble with a proxy is that it describes a group, and groups are mixed. As the same study explains, "risk characteristics grouped together in the process of risk classification are priced on an average, so some individual risks are above the average and some are below." Those below the average subsidize those above it.

Here is a made-up illustration. Two 30-year-old drivers live in the same ZIP code, drive the same model of car and have clean records. One drives 6,000 miles a year, mostly in daytime and in light traffic, and brakes gently. The other drives 18,000 miles, often late at night, and brakes hard. A rating plan built on proxies puts them in the same cell and charges them about the same. An insurer with good driving data can separate them. It can charge the first driver less and the second more, or at least keep the first driver from leaving for a competitor that will.

The same study says "traditional rating factors tend to be proxies for risk. The idea of telematics is to actually measure risk on an individual level." A section by Robin Harbage of Towers Watson adds that "almost no insurers base the entire premium on just the driving behavior, and most still largely rely on the common proxy variables approved for use in their jurisdiction." That is still the design of programs: driving data adjusts a premium that was already built from the traditional factors. See how car insurance rates are calculated for the usual factors and base rates and relativities: how a premium is multiplied together for how they multiply.

Adverse selection: why safer drivers join first

Adverse selection is what happens when the buyers who choose a product are systematically different from the average buyer, in a way the seller cannot see. In usage-based insurance it runs in the insurer's favor if the insurer moves first, and against it if it does not.

The mechanism is simple. Anyone who drives carefully and does not mind being watched has a reason to enroll, because the program rewards what they already do. A heavy night driver with frequent hard stops has a reason to skip it. So enrollment sorts drivers by risk without the insurer having to identify them in advance. The NAIC study says it directly: "It makes sense that someone who drives a lot, at unusual times and unsafely probably will not sign up for these programs. The early adopters will bring in good drivers and can rate them at fairly cheap prices." A Belgian study of young drivers makes the same point, noting that insurers reason that telematics products "are only purchased by drivers who consider themselves to be either safe or low-kilometer drivers," a form of positive selection that justifies "an upfront discount."

Economists Yizhou Jin and Shoshana Vasserman measured this in U.S. data. They studied a voluntary monitoring program at a major auto insurer using data on more than one million customers from 2012 to 2016, matched with competitors' price menus. Their NBER working paper, Buying Data from Consumers, reports that "safer drivers self-select into monitoring" and that "those who opt-in become 30% safer while monitored." The NBER Digest summary adds that opt-in drivers made 30 percent fewer claims during monitoring than in later periods, and that the monitoring scores strongly predict later claims.

Why the competitor has to follow

Selection makes the market unforgiving, and a small example shows why. The numbers are invented.

Say 1,000 drivers split evenly between 500 careful drivers who will cost an insurer $800 a year each in claims and 500 risky drivers who will cost $1,200 each. Two insurers share the market equally and, unable to tell the groups apart, both charge about $1,000.

Insurer A launches a monitoring program and offers a discount. Careful drivers can earn it and many sign up. Suppose 300 of the careful drivers at Insurer B switch to A for a price near $850. Insurer A has 300 drivers costing $800, so it profits. Insurer B is left with 200 careful drivers ($800) and 500 risky drivers ($1,200), a pool that costs $1,086 on average, and its $1,000 price now loses money. To fix that, B must raise its price, which pushes out more of its careful drivers.

This is the logic behind the NAIC study's warning that "late adopters may end up competing with each other for a shallower pool of riskier drivers," and its observation that "there will be adverse selection where riskier drivers may be more likely to use insurers not using a telematics system." It also explains why insurers advertise telematics heavily even when a minority of customers use it: the threat of losing good risks matters more than the share of customers enrolled. Real markets have frictions, such as loyalty, agents and bundling, so the spiral is slower in practice, but it points in the right direction.

How much better is behavior than the proxies?

There is good evidence that driving behavior carries information the traditional factors do not, though the strongest claims come from vendors and one-company studies.

• Vendor claim, reported in the NAIC study. Harbage writes that Towers Watson's DriveAbility score "has been demonstrated to be at least three times more predictive than any rating variable previously employed when comparing the difference in loss costs between the riskiest decile of insured vehicles and the safest decile." That is a claim by the vendor, reported inside a regulator-sponsored document, and I could not check the underlying data. Treat it as a direction, not a measurement.
• Peer-reviewed work. Roel Verbelen, Katrien Antonio and Gerda Claeskens used Belgian data on young drivers to test whether telematics variables improve a claims-frequency model. They found that "such variables increase the predictive power and render the use of gender as a discriminating rating variable redundant." In the classic model without telematics, women appeared 16% less risky. Once they added kilometres driven by time and road type, gender dropped out, because the women in the sample drove fewer kilometres. Their working paper is dated January 2017 and is about Belgium and a young-driver product, so it shows the principle and not a U.S. number.
• Model design matters. The NAIC study describes two approaches: "event counter" scores, which count harsh braking, acceleration and cornering, and richer models built on second-by-second data. It says event counters are limited because they assume "a few harsh braking, acceleration or cornering events constitute the universe of variables."

For a consumer, the lesson is that better prediction does not automatically mean a bigger discount. A more accurate model can also move more drivers up.

What it costs, and why the sign-up discount exists

Collecting the data is not free for the insurer, and neither is getting a customer to agree.

The hardware and software cost has fallen. The Belgian paper notes that early telematics insurance had "high implementation costs and its complexity limited its success," and that "advances in technology and telecommunication have however reduced the cost substantially." The NAIC study describes the smartphone as an attractive route because insurers "would save the cost of purchasing monitoring devices for each user," but adds a quality problem: smartphone accelerometers are not calibrated and phone position changes. I found no public figure for what a device or app costs an insurer per customer, and I will not guess one. Progressive's FY2025 annual report says that "the majority of new enrollments" now choose mobile devices over the plug-in.

The larger cost is the customer's reluctance. Jin and Vasserman's model finds sizable demand frictions in monitoring, which push the insurer to offer big discounts to get people to opt in. According to the NBER Digest, they estimate that customers implicitly pay about $93 to avoid monitoring. That figure is a modeled estimate from one insurer's data, not a price anyone quotes, but it explains the economics. The enrollment discount is the insurer's purchase price for the data. Progressive's plug-in terms describe a participation discount for up to the first two policy terms, "until sufficient Snapshot Data has been collected." Our guide how a telematics discount works: sign-up credit and renewal explains the mechanics.

What does the insurer get for the money? The same paper puts a dollar value on the whole arrangement: combined gains to consumers and insurers of about $13 per driver per year, about two-thirds from safer driving and the rest from more accurate pricing, rising to about $21 under profit-maximizing prices, according to the digest. These are model results for one program and period. They suggest the economics are positive but modest per driver, which is why scale and selection matter.

The NAIC study lists further benefits for insurers, including "improving brand recognition and loyalty" and "additional evidence to consider when investigating claims," which cuts both ways for the policyholder, as can your telematics data be used in a claim or a lawsuit? explains.

Competition and market share

Telematics is also a competitive weapon. The NAIC study notes that private-passenger auto premiums grew only from $158 billion to $175 billion over the ten years to 2013 and "the market has not even kept up with inflation," so "for some large insurers showing strong growth, most of the growth is primarily a result of increasing their market share." In a stagnant market, a tool that attracts better drivers is valuable.

Three features of the competition stand out:

• Lock-in. The study says "insurers are keenly aware that the first telematics device consumers install will most likely be their last." A customer who has built a discount history, an app and a score has a reason to stay.
• Data scale. Insurers with large datasets of driving and claims can build better models. "Competing insurers entering the market later are placed at a serious disadvantage because they lack the valuable large and statistically credible UBI data sets," the study says.
• Ownership of the data. Jin and Vasserman find that monitoring can strengthen the sponsoring firm's market power when the data stays proprietary, because competitors cannot see a customer's score. Their analysis of a data-portability policy concludes it would reduce the firm's incentive to collect monitoring data in the first place and lower consumer welfare in equilibrium. That is a debated policy question, not a settled one.

Progressive's FY2025 annual report shows how central this has become for one company. It says, "We believe a key element in improving the accuracy of our personal auto rating is Snapshot," that Snapshot "is available in all states, other than California," and that the latest segmentation model was available in states representing 81% of countrywide personal auto net premiums written, excluding California, at year-end 2025. It also reports that adoption rates for consumers enrolling in 2025 fell 2% in the agency channel and rose 6% in the direct channel, compared with the prior year. You can read the passage in the FY2025 Form 10-K.

From pilots to national programs

The idea is older than the apps.

The 2015 study also estimated that Progressive's program had about 2 million customers and $2 billion in premiums, enough to rank among the top 15 private-passenger writers if it stood alone. Launch dates differ by state and product, so treat the table as milestones, not a complete history.

The LexisNexis release, dated July 24, 2020, is revealing about the economics. Its executive said that "previously many consumers agreed to a UBI policy without historical driving behavior data, before knowing its true benefit," and that "U.S. insurance carriers offered enrollment discounts meant to excite consumers without knowing the full risk to their books of business." The product exists to let insurers see existing data, with consent, at the point of quote instead of paying for a trial period. In other words, the industry sees the enrollment discount as a cost to be reduced.

Limits and risks

The economics are not a free lunch for anyone, and several cautions deserve to be said plainly.

Observation changes behavior, and the change may fade. The 30% figure comes from claims during the monitored period, and the NBER Digest notes it may partly reflect higher driving effort while monitored, not only lasting change. Selection also means the group that enrolls is not a random sample, so comparisons between enrollees and non-enrollees overstate what the program would do for the average driver.

Many drivers do not share the gain. Progressive's Snapshot page says "only about 2 out of 10 drivers actually get an increase," which implies most enrollees see a discount or no change, and that some do not. Nationwide's page says its enrollment discount applies "until a driving behavior discount can be calculated, which could be zero." Whether the program saves you money depends on your driving and your starting price.

Proxies can sneak back in. The NAIC study asks whether a model that finds late-night urban driving risky might penalize evening-shift workers, and whether combining driving data with territory models could "produce rates that are unfairly discriminatory." Regulators review these questions in the rating plan, as how regulators review usage-based insurance rating plans explains.

Privacy has a price. The Consumer Federation of America's January 2025 letter to the NAIC says telematics can benefit consumers "by more accurately matching insurance rates to risk, promoting safer driving behavior, and reducing insurers' use of unfair socioeconomic factors," but that there are "substantial concerns about accountability and transparency," and that most consumers do not participate. The same letter describes a March 2024 New York Times report that automakers shared driving data with data brokers, which our guide connected-car data, data brokers and your premium covers.

Measurement errors cost real money. A phone that cannot tell a driver from a passenger, or a device that records a trip nobody drove, can move a price. See when a driving app scores a trip wrong: how to dispute it.

What this means when you are deciding

Knowing the insurer's motive helps you bargain. The company is paying you to reveal something it cannot see. If you are a lower-mileage, cautious driver whose current price is built on averages, you may be worth more to the insurer than the discount it offers. If you drive a lot at night or brake hard, you are the person the program is designed to filter out, and a discount-only program may still give you something while a program that can raise rates may not.

A few questions help:

1. Is this a discount-only program or can it raise my rate? Programs differ.
2. How large is the sign-up discount, and when does it expire?
3. What data do I give up, and who else gets it?
4. What would I pay without it? Compare quotes from insurers with and without telematics, using the usage-based insurance glossary entry for terms and our guide how to run your own numbers on a telematics trial before you commit for a worked method.

Common questions

Do insurers make money from the data itself?

The sources I read describe the money as coming from better pricing, safer drivers and lower claims, not from selling the data. Some insurers state they do not sell it. Progressive's terms, for example, say location may be used for underwriting, accident detection and claims, and research and development. Check each program's privacy statement.

Why give a discount just for signing up?

Because the enrollment discount is the cost of getting the data, and of getting a customer to try something they may fear. Researchers estimate consumers need to be paid to accept monitoring. The discount is often temporary, and the real price depends on the score.

Does telematics make car insurance cheaper overall?

Not by itself. It redistributes premiums toward safer drivers. The NAIC study notes that discounts do not replace sound classification: "While premium discounts are welcomed, they are not a substitute for the establishment of appropriate classifications of risk and actuarially sound rates."

Is the approach new?

Pay-as-you-drive ideas were in development in the 1990s, and the first national programs from major U.S. insurers date from around 2011, according to the NAIC study. What changed was the cost of data and the arrival of the smartphone.

Key takeaways

• Insurers want driving data because age, ZIP code and similar factors are proxies that price groups on averages, while behavior measures the individual.
• Safer drivers join voluntarily, which is adverse selection working in the insurer's favor; a competitor that does not offer a program risks being left with riskier drivers.
• Research on one large insurer found safer drivers self-select and become about 30% safer while monitored; the modeled gains were about $13 per driver per year, and customers implicitly valued avoiding monitoring at about $93.
• The sign-up discount is the price of the data, which is why it is often temporary and why insurers want to reduce it with data they can obtain at quote.
• The case has limits: monitored behavior may not last, many drivers see small or no gains, proxies can return, and privacy and measurement errors are real costs.