How telematics scores are calculated: braking, speed, phone use

A telematics score is not a grade for how safe you are. It is a count of events that a statistical model has found to go with insurance claims, such as hard stops, time spent well over the speed limit, late-night miles and phone handling while the car moves, adjusted for how far you drove and blended across the drivers and trips on your policy. The insurer then converts that count into a discount, or in many programs a possible increase, at renewal.

That design explains most of the confusing results drivers report. A careful driver who commutes through stop-and-go traffic can be flagged for braking, and a reckless driver who never touches a phone can score well on that factor. This article explains what the sensors record, how thresholds are drawn, how short trips and city driving are treated, how trip data become one policy-level result, and what the research says about whether the scores predict crashes. Program details are as published by each insurer in November 2026 and vary by state.

What the sensors actually record

Three kinds of hardware feed these programs, and each sees a different slice of your driving. The companion article on tracking methods compares them for privacy and practicality; here the question is what each one can measure.

Liberty Mutual states that its RightTrack app needs a phone with GPS, an accelerometer, a gyroscope and proximity sensors, and that iPads, tablets and Windows phones are not compatible, according to its RightTrack page. Progressive's plug-in device terms (last updated March 2022) say the device records speed and location by GPS, time, the vehicle identification number, G-force on some devices, and when it is plugged and unplugged, and that miles driven and rates of acceleration and braking are calculated from speed and time. The same terms say where you drive is not used in determining your personalized rate.

The accelerometer is the workhorse. It measures how quickly the device is speeding up or slowing down in each direction. Because a phone can sit in a cupholder, a pocket or a mount at any angle, the software first works out which way the car is facing, then reads forward and backward acceleration relative to the car. Cambridge Mobile Telematics (CMT), a company that supplies scoring technology to insurers, describes its hard-braking measure in just those terms: longitudinal acceleration with respect to the car's frame of reference, in a 2024 report for the Washington Traffic Safety Commission.

The behaviors that get scored

Programs score some mix of five things: braking and acceleration, speed, time of day, phone handling and mileage. The mix differs by insurer and sometimes by state. State Farm's Drive Safe & Save page lists acceleration, braking, cornering, phone distraction, speed and annual mileage, with no mention of time of day. Allstate's Drivewise help page lists speed, braking events, time of day and phone activity. Liberty Mutual lists braking, phone use, speed and time of day in most states, but braking, acceleration and time of day in some. Nationwide's SmartRide page lists miles, hard braking and acceleration, idle time and nighttime driving.

The numbers behind each label are where programs differ, and few are published. These are the ones that are:

The CMT numbers are one vendor's published choices, not an industry standard, and an insurer that licenses CMT technology may configure it differently. Treat them as an illustration of the scale. A deceleration of 3.2 meters per second squared is about a third of the pull of gravity: firm enough that a passenger notices, well short of a collision.

Two details in the table matter for real drivers. First, the speeding lines include a buffer, so cruising 3 mph over the limit does not register under either published definition. Second, the phone definitions are about handling, not calls. State Farm's rule fires only when the screen is on and the phone is moving; CMT separates "phone motion" (picking the phone up) from "phone calls," which it reports as a different measure in its study. Whether a program penalizes a hands-free call is something to ask.

How a threshold gets chosen

Nobody decides that 3.2 meters per second squared is the "right" level of braking by looking at a safe driver. The line comes from a model. CMT says its threshold was selected through actuarial studies "to maximize the correlation between a pattern of hard-braking and the likelihood of a crash." In plain terms, analysts tried many cutoffs against historical trips and claims and kept the one that best separated higher-risk patterns from lower-risk ones.

That has three consequences:

1. The line can move. If a newer model fits better with a different cutoff, the program can change it. Programs also change by state, because rating factors must be filed with and accepted by each state's regulator.
2. Events are rare by design. A threshold set so high that most drivers never cross it is useful to a model, because it isolates the unusual stop. The CMT report notes that events at or above its braking line are "not a typical part of safe driving aside from isolated incidents."
3. Context is partly invisible. A sharp stop to avoid a child running into the road is exactly what a safe driver should do, and it counts the same as a stop caused by tailgating. The score reflects the pattern across hundreds of trips, not the story behind one.

Why your score is relative, not absolute

Insurers do not publish a formula that turns events into dollars, and the public pages describe the results in terms of outcomes rather than arithmetic. What they do reveal is that the result is judged against other drivers and against the insurer's own pricing model.

Allstate's help page says Drivewise "considers the driving info from all qualifying drivers on the policy" and distinguishes the participation discount from the performance rating. Liberty Mutual says the impact on your premium is "calculated using a combination of your policy driving score and participation at each policy renewal." In other words, there are two moving parts: what you earn for taking part, and what your behavior adds or subtracts.

The usual way to build such a score is to take a baseline model that predicts claim frequency from ordinary rating factors such as age, vehicle and location, then ask how much better the prediction gets when the telematics measures are added. A driver whose events sit below what the model expects for similar drivers earns a credit; one whose events sit above gets a smaller credit or, where the state and program allow, a charge. That is why the same driving can yield different outcomes at different insurers and why a 90 percent "score" in an app does not map to 90 percent of anything. State Farm's own help page says the app's 14-day scorecard is not directly tied to the premium adjustment for each vehicle.

Whether the result can raise your price is a separate question from how it is calculated, and it depends on the insurer and the state. See telematics programs from the major insurers compared for the insurer-by-insurer picture, and how a telematics discount works: sign-up credit and renewal for the money side.

Short trips, stop-and-go traffic and highway miles

The treatment of trip length is the most common source of complaints, and the public detail is thin. The one measure we can describe from a published source is CMT's. It counts hard-braking events and divides by distance, expressing the result as events per 100 kilometers. Speeding and phone handling are measured as time: minutes of speeding or distraction per hour of driving. Trips with no events still count in the denominators. Under that approach, a long highway trip with two hard stops produces a low rate, and a short city trip with two hard stops produces a high one.

Cities are where that matters. The same CMT report found that in the Washington cities it analyzed, hard-braking rates were higher than in the surrounding counties, while speeding, phone-motion and phone-call rates were lower. Dense traffic really does produce more abrupt stops. Whether your insurer adjusts for road type, and how, is not published for any program we reviewed, so the safe assumption is that city driving puts you at a mild disadvantage on braking and an advantage on speed.

A worked example with invented numbers

Say two drivers each record exactly five hard-braking events in a week. This is a hypothetical, not program data.

• Driver A drives 60 miles a day on a suburban interstate, about 420 miles in the week. Five events over 420 miles is roughly 1.2 events per 100 miles.
• Driver B drives 10 miles a day in a downtown grid, about 70 miles. Five events over 70 miles is roughly 7.1 events per 100 miles.

If the program measures events per distance, Driver B looks six times worse, even though both had the same count. If the program scores events per trip, the result flips depending on how many trips each driver took. This is why a program's description of its measure matters more than its marketing, and why asking "per mile or per trip?" is a fair question to put to an agent.

From trips to one policy-level result

Insurers price policies, not trips, so trip data are rolled up in a few ways.

• A minimum amount of driving. Allstate says at least one enrolled driver must take 50 trips in the driving period to keep the participation discount, and that renewal reviews the trips taken "over the past six months." Nationwide says the driving discount starts only if all drivers activate the app within 30 days of the policy's effective date and it stays active for 80 days. State Farm does not give a minimum, saying only that the adjustment is updated at renewal "after you complete setup and we have enough data," using up to the prior 12 months of driving.
• All drivers together. Liberty Mutual says it calculates the policy score "based on the driving behavior and risk factors of all participating drivers." Allstate says only drivers who complete 50 or more trips are counted as qualifying. For households with several drivers, see telematics in households with several drivers: who is scored?.
• A fixed window or a rolling one. Liberty Mutual's 90-day program determines a final score after 90 days and applies it "for the life of your policy," while its continuous program applies a new score at each renewal. State Farm's program recalculates at each renewal, typically every six months.

The participation discount sits on top of the behavior result. Liberty Mutual's page describes an immediate discount of 15 percent if all drivers participate and 10 percent if only some do, with total savings of up to 30 percent for safe driving. Allstate describes a 10 percent participation discount for the initial policy period in most states for drivers who choose a personalized rate.

Do these scores predict crashes?

The evidence is real but narrower than the marketing suggests. Three kinds of studies matter.

Pricing research. A paper by Roel Verbelen, Katrien Antonio and Gerda Claeskens in the Journal of the Royal Statistical Society, Series C (2018) used data from a Belgian telematics product for young drivers and found that telematics variables improved the prediction of claim frequency, enough to make gender redundant as a rating variable. The abstract is on IDEAS/RePEc. It is a single European product aimed at young drivers, so it does not prove the same result for every U.S. program.

Behavior-link research. The Insurance Institute for Highway Safety published a study in April 2026 using roughly 600,000 trips from CMT-supplied safe-driving apps. It found that phone handling rose 12 percent for every 5 mph over the local limit on limited-access roads, and 3 percent on other roads. The IIHS write-up measures phone use and speeding together; it does not measure crashes. It shows that the behaviors cluster, not that a given score predicts a claim.

Behavior-change research. A study in Marketing Science by Miremad Soleymanian, Charles Weinberg and Ting Zhu, summarized by ScienceDaily, used data from a large U.S. insurer and found that daily hard braking fell by about 21 percent after six months of monitoring. That means a score can change what it measures, which is a benefit to the driver and a reason an early score is not a permanent label.

Taken together, the research supports using telematics measures as rating variables. It does not show that any one app's score is a fair or complete safety grade, and much of the crash-prediction evidence comes from companies that sell the scoring technology.

Steps to see how you are being scored

1. Open the in-app trip list before you commit. Allstate lets drivers try the app without affecting their rate in some states; Liberty Mutual's dashboard shows trip details and what changed.
2. Find the measure's unit. Ask whether braking and speeding are counted per mile, per hour or per trip.
3. Check how the app decides you were driving. Allstate says its app can detect passengers and vehicle type (automobile, train, bus, plane or boat) and that only driver trips count; drivers can correct a mislabeled trip in the app. State Farm lets a listed driver reclassify a trip, which removes that trip's phone distraction from the score.
4. Look at your worst hour. Late-night windows are the cheapest event to avoid if your schedule allows.
5. Treat the phone as a sensor. Mount it, and leave it alone while the car moves.

For disputes over a trip that was wrong, see when a driving app scores a trip wrong: how to dispute it; to model whether a trial is worth it, see how to run your own numbers on a telematics trial before you commit.

Common questions

Does a hard brake always hurt my score?

No. A single event is one data point among hundreds of trips, and the models are built on patterns. Frequent events per mile are what move a result.

Why did my score drop on a day I drove carefully?

Common causes are a phone that was picked up at a red light (screen on and moving), a trip taken in the high-risk late-night window, or a passenger's trip recorded as yours. Check the trip list and reclassify any trip you were not driving.

Is every program scoring me against the same standard?

No. Thresholds, time windows and weights differ by insurer, state and program version, and most are not published. The CMT figures above are one vendor's published choices.

Can a good score be wasted if I drive very little?

Possibly. Programs that require a minimum number of trips, like Allstate's 50, may not give full credit to a driver who takes few trips, and mileage-based programs price on distance rather than behavior. pay-per-mile, low-mileage discount or telematics: which fits? compares the options.

Key takeaways

• A telematics score is a model's estimate of claim risk built from counted events, not a measure of how safe you are.
• Braking, speed and phone thresholds are chosen statistically and differ by program; few insurers publish them.
• City driving produces more hard-braking events, and whether a program normalizes by distance is something to ask before enrolling.
• Policy-level results combine trips, often from every driver on the policy, over a fixed window or at each renewal.
• Research supports telematics as a rating tool, but most evidence is narrower than the marketing claims.