AI use case

AI telematics driving scores for motor insurance pricing

AI telematics scoring turns data from a phone app or a plug in device (braking, speed, cornering, phone use, time of day) into a driving behavior score for each policyholder. The insurer uses the score to price motor insurance, to reward safer driving and to give the driver feedback.

Last verified 18 November 2026 · 3 public deployments

USD 2.2
Customer savings
Progressive Corporation (organization claim).
USD 300,000 to USD 3.6 million
Indicative value per year
A motor insurer with 500,000 policies and a telematics program for new customers. Worked example, see how it is calculated.

What problem does it solve?

Motor insurance has long been priced on age, address, vehicle and claims history. These are proxies for how a person drives, and they charge careful drivers in a risky group the same as careless ones. Telematics measures the driving itself: sensors in the phone or a device record braking, acceleration, cornering, speed, time of day and phone use on every trip.

The raw trip data is noisy, since a phone in a passenger's pocket or a bus ride looks like driving. Machine learning cleans the trips and statistical models turn them into a score that predicts claims. The insurer uses the score in the price, as an upfront discount or a rate at renewal, and shows it to the driver. Some insurers add coaching that points to the habit that matters most. The evidence on this page shows the scale of the programs and, from one insurer, a measured effect of coaching on driving behavior. It shows less about claims results, because insurers publish few figures that isolate the effect of the score.

How does it work?

  1. Collect trips with consent. A phone app or a device records each trip. A test drive of a few weeks can serve as the basis for the first price.
  2. Clean the data. A model detects whether the person was driving, whether the phone user was the driver and which events were real, such as a hard brake or phone use.
  3. Score the driver. A model turns the cleaned trips into a behavior score that predicts claims, and a pricing model combines it with the usual rating factors.
  4. Price and reward. The insurer applies the score as an upfront discount, a renewal rate or a reward, within the limits of the rates filed with the regulator.
  5. Give feedback. The app shows the score and the trips behind it, and may suggest a goal for the habit that most affects the score.
Audience
Customer facing
Autonomy
Assist
Adoption
Mainstream
Channels
Mobile app

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

Value benchmarks for AI telematics driving scores for motor insurance pricing
KPIMedianReported rangeData pointsClaimed by
Customer savingsNot pooled
USD 2.2
11 organization

Value drivers: Risk and loss reduction, Revenue growth, Customer experience.

Indicative value

A motor insurer with 500,000 policies and a telematics program for new customers

USD 300,000 to USD 3.6 million

Claims cost avoided through better pricing per year

How this is calculated

Formula: policies * premium * lossRatioGain. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Policies in the telematics program policies, policies50,000100,000Editorial assumption for 10 to 20 percent of the reference insurer's book, replace with your own enrolment.
Average annual premium premium, USD per policy600900Editorial assumption, replace with your own average premium.
Loss ratio improvement on enrolled policies from better selection and pricing lossRatioGain, fraction of premium0.010.04Editorial assumption, replace with your own measured difference. No insurer on this page publishes a figure that isolates the effect of the score.

What it leaves out: Gross claims effect only. It leaves out discounts given to customers (Progressive reports $2.2 billion since 2009), device and app cost, drop outs from the program and the effect of self selection, since safer drivers join voluntarily.

Which organizations already do this?

3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

This library is not a list of Blits.ai customers. It documents public AI deployments from any vendor, each with its source.

Discovery Insure

South Africa · Insurance · 2026

PilotGrade B

Discovery Insure, a division of Discovery Limited in South Africa, rewards clients for driving well through Vitality Drive, using telematics from a phone or a device. It reports that in ten years of working with Cambridge Mobile Telematics its crash related claims fell by 26%. In January 2026 it launched an AI powered driving coach as a proof of concept, and in October 2026 it published a study with Johns Hopkins University of 5,000 drivers, randomly assigned to coaching strategies, in which drivers receiving personalized AI coaching improved on cellphone use, harsh acceleration, braking, cornering and speeding against a matched comparison group.

No outcome disclosed.

Progressive Corporation

United States · Insurance · 2025

ScaledGrade B

Progressive's Snapshot program measures driving behavior with a plug in device or a mobile app and lets personal auto customers improve their rates, in every US state except California. Its annual report calls Snapshot a key element in improving the accuracy of personal auto rating and says that its latest segmentation model was available in states representing 81% of countrywide personal auto net premiums written, excluding California, at the end of 2025. The company also runs a commercial version for fleets. It states that Snapshot has provided $2.2 billion in discounts to drivers since 2009.

  • Customer savings: USD 2.2, since 2009, in billions
    "Since 2009, Snapshot has provided $2.2 billion in discounts to drivers."
    Claimed by: organization

Root, Inc.

United States · Insurance · 2024

ScaledGrade B

Root, a US car insurer, prices policies mainly on driving behavior measured by telematics. In some sales channels an applicant first completes a test drive of two to four weeks, during which the phone sensors record braking, turning, time of day and attention. The company says it uses machine learning to clean the phone data and statistical methods to turn it into behavior scores, and that its pricing models use those scores. Its annual report says that a California law restricting the use of such data has made cost competitive pricing there hard. It publishes loss ratios for the whole book but no figure that isolates the effect of the driving score.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Trip data with sensor readings from phones or devices, with consent
  • Policy and claims history to train and validate the score against claims
  • Rate filings and pricing rules that say how a score may change a price

Systems to integrate

  • Mobile app or device platform for trip capture
  • Rating and policy administration system
  • Customer app or portal for the score and feedback

Complexity: High

The scoring needs a telematics platform or an in house app with trip detection, a data science team or a vendor that can build and validate a score on claims, a rating engine that accepts the score and rate filings approved by the regulator in each state or country. Consent for continuous location data and a legal review of what the score may be used for are part of the build.

  1. 1

    Start with a score that predicts claims

    Build the score on trips matched to later claims and test that it adds to the usual rating factors before it moves any price.

  2. 2

    Use a test period before pricing

    A trial of a few weeks gives the first score, lets the driver see what is measured and lets you remove bad trips before a price is set.

  3. 3

    File and explain the pricing use

    Agree with the regulator how the score enters the price, and write plain language explanations of what raises and what lowers a score.

  4. 4

    Add coaching and test it

    Run a randomized test of feedback or coaching against a comparison group before rolling it out, and measure behavior and claims.

Guardrails

  • Explicit consent for location and sensor data, with a way to leave the program
  • A driver can see the trips behind the score and dispute a wrong trip
  • A score cannot raise a premium above the limits in the filed rating plan
  • Checks that the score does not act as a proxy for protected characteristics

KPIs to instrument

  • Lift of the score over the standard rating factors on claims frequency and cost
  • Enrolment, completion of the test period and drop out rates
  • Share of trips corrected or disputed by drivers
  • Change in risky behaviors among coached and comparison drivers

Human in the loop

Actuaries and pricing managers validate the score and own how it enters the price. Customer service staff handle disputes about trips and scores, and a person reviews the fairness testing before each rate change.

Common failure modes

Self selection looks like a model effect
Careful drivers join voluntarily, so low claims in the program may reflect who joined. Compare against matched drivers and measure the score on customers who did not choose it.
Legal limits on the data
Root's annual report says that a California law restricting use of driving behavior data has made cost competitive pricing there hard, and Progressive offers Snapshot in every state except California. Check each market's rules on telematics pricing before launch.

What are the risks and rules?

EU AI Act

Minimal risk

Annex III point 5(c) makes AI for risk assessment and pricing of life and health insurance high risk. Motor insurance pricing is not listed, so a driving score used for car insurance falls outside that point. GDPR applies in full, because continuous location and sensor data is personal data, and insurers must also meet national rules on rating factors and consent.

Rules that apply

Guidance

Controls to put in place

  • Consent records and a documented way to withdraw
  • Validation of the score against claims on a holdout sample
  • Fairness testing across age, gender and area before each model change
  • Audit trail of the score version used in every price

Frequently asked questions

How much do telematics programs save drivers?

Progressive states that Snapshot has provided $2.2 billion in discounts to drivers since 2009. That is a total for the program, not an average per driver, and it is the company's own figure.

Does telematics coaching change how people drive?

In an October 2026 study with Johns Hopkins University, Discovery Insure assigned 5,000 drivers at random to coaching approaches and reports that drivers receiving personalized AI coaching improved on five risky behaviors compared with a matched group, most on cellphone use. The release comes from the insurer and the full paper was not reviewed for this page.

Is telematics pricing allowed everywhere?

No. Root says in its annual report that a California law restricting the use of driving behavior data has hindered its ability to offer cost competitive policies there, and Progressive offers Snapshot in all states except California. Rules on rating factors and consent differ by market.

How to cite this page

Blits.ai AI Use Case Library, "AI telematics driving scores for motor insurance pricing", last verified 18 November 2026, https://www.blits.ai/ai-use-cases/telematics-driving-score-insurance. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 18 November 2026: First published

Related use cases

Insurance

AI for insurance renewal processing and customer retention

Renewal and retention AI prepares and runs the insurance renewal cycle: it digitizes renewal submissions and changes in risk for underwriters, flags policies at risk of lapsing or leaving, and runs renewal and premium reminders and the renewal conversation with the customer, while renewal prices stay governed by the insurer's pricing rules and fair value obligations. Routine mid term changes and coverage questions belong to the insurance policy servicing agent, and telecom churn to churn prediction and retention offers.

Deployments
4 public, best grade B
Reported automation rate
100%
JAGADIRI, vendor claim
Insurance

AI aerial imagery risk scoring for property insurance underwriting

AI property risk scoring reads aerial imagery to rate roof condition, wildfire exposure and other building features at quote, so underwriters inspect fewer homes. It turns an address into a score and a set of flags that feed the insurer's eligibility, inspection and pricing rules, and the underwriter or the rules engine keeps the decision.

Deployments
4 public, best grade C
Reported accuracy
at least 95%
Florida Peninsula Insurance Company, organization claim
Insurance

AI copilot for insurance pricing and actuarial analysis

An insurance pricing and actuarial copilot is AI that speeds up the work of pricing and actuarial teams, from risk and demand model building to natural language analysis of data. It builds automated, transparent risk and demand models and analyzes rate filings, experience data and reserving diagnostics, while actuaries select the models, sign off the rates and own the professional judgment.

Deployments
9 public, best grade B
Reported productivity gain
5x
Generali France, organization claim
Insurance

AI for commercial underwriting submission intake and triage

AI that reads incoming broker submissions for commercial insurance so underwriters open a complete, prioritized file instead of an inbox. It reads emails, applications, schedules of values, loss runs and supplements, extracts the risk data into a structured record, checks clearance and appetite, and enriches and ranks the risk with internal and third party data. Its job ends when a cleared, in appetite and ranked submission reaches an underwriter's queue.

Deployments
8 public, best grade B
Reported accuracy
about 98%
Paragon Insurance Group, organization claim