Commercial auto posted a $4.9 billion underwriting loss in 2024, a 14th consecutive year in the red (AM Best). AI in-cab cameras across more than 2,600 fleets show 73% crash rate reductions over 30 months. Filed credits run 5% to 15%. The gap between those two numbers is not regulator skepticism about the technology; it is the credibility weight the vendor data can carry in a rate filing.

$4.9B
2024 Underwriting Loss (14th Straight)
73%
Crash Rate Reduction (Samsara, 30 Months)
48%
Harsh Event Drop in First 6 Months
5–15%
Current Carrier Telematics Credits

Key Takeaways

  • 73% crash rate reduction over 30 months across more than 2,600 fleets, with harsh driving events down 48% in the first six months and mobile phone use down 84%.
  • 1,082 expected claims is the classical full credibility standard for frequency at 90% probability within 10%. A 100-vehicle fleet generating 12 bodily injury claims a year needs roughly 90 years to reach it alone.
  • Filed credits sit at 5% to 15%, including The Hartford's FleetAhead at up to 5% per vehicle, because the Z-weight on two to three years of equipped-fleet data is well below 1.0.
  • Severity has compounded at roughly 8% a year for nine consecutive years against about 3% general inflation, with total commercial auto claim severity up 64% since 2015.
  • 49 thermonuclear verdicts above $100 million in 2024, an 81.5% increase, and auto cases were 22.8% of all verdicts exceeding $10 million.

What the 30-Month Study Measured

The Samsara 2025 Physical Safety Report covered more than 2,600 fleets worldwide over a 30-month window. For fleets of 175 or more vehicles running the full solution, dual-facing cameras plus in-cab alerts plus structured coaching, harsh driving events fell 48% in the first six months and mobile phone use 84%. By month 30, harsh events were down 69% and mobile use 96%. Crash rates fell 73% across the population, and CSA Unsafe Driving scores improved 43%.

The intervention model is what makes those curves steeper than first-generation telematics. GPS and engine diagnostics platforms measured behavior after the event and priced credits off retrospective scores. An in-cab alert fires within seconds and a coaching session follows within hours, interrupting the sequence before a claim exists.

Capability First-Generation Telematics AI Fleet Cameras (2022+)
Intervention model Outcome monitoring after the event Real-time behavioral interruption
Alert speed Retrospective weekly score review In-cab alert within seconds
Coaching trigger Aggregate score thresholds Event-specific session within hours
Insurance pricing model Retrospective behavioral credit Prospective prevention credit (actuarial gap)
Industry loss experience 10-plus years at scale 2 to 3 years at fleet scale
Filing credibility status Established (ISO Telematics Classification Plan) Emerging; below full credibility thresholds

Two features of the study population complicate the magnitude. Fleets buying AI camera programs at early-adopter scale tend to be larger operators with professional driver programs, dedicated safety managers and existing coaching infrastructure. The operations most likely to generate catastrophic claims, small regional carriers and high-turnover fleets without safety coordinators, are underrepresented. A matched-control comparison against non-equipped fleets with similar safety cultures would produce a different number. The direction is well supported; the size is not yet separable from selection.

The Credibility Weight Is Doing the Work

The classical full credibility standard for claim frequency requires about 1,082 expected claims for 90% probability that observed experience falls within 10% of the true value. Run that against a fleet. A 100-vehicle operation generating 12 bodily injury claims a year needs roughly 90 years. A 500-vehicle fleet at 60 claims a year reaches it in about 18. No individual fleet arrives at full credibility on its own experience.

Buhlmann blending handles the shortfall by weighting fleet experience against an industry complement at Z = n / (n + k). When n is small the complement dominates, and in 2026 n is at most two to three policy years of equipped-fleet data inside any one carrier's book, while the industry complement carries no meaningful AI camera penetration at all. The frequency credit that survives a filing is therefore necessarily far below the 73% vendor figure, because the Z-weight on that figure is well below 1.0.

The development clock compounds it. Accident frequency in commercial auto develops over 12 to 24 months; bodily injury claims develop over three to seven years. A fleet installing cameras in 2023 with visible behavioral change by mid-2024 does not have mature loss experience for the 2023 through 2025 accident period until 2027 through 2029.

That is what the filed numbers reflect. The Hartford's FleetAhead offers up to a 5% per-vehicle discount for qualifying telematics in selected states, and fleets sharing verified driving data across the broader market qualify for 10% to 15%. Those are credibility-weighted behavioral credits documented from carrier loss experience, not prospective technology credits taken from a vendor publication. ISO's Telematics Classification Plan asks for exactly that: demonstrated correlation between behavioral metrics and loss cost differences inside the carrier's own book across multiple policy periods.

The Half of the Loss Cost Cameras Do Not Touch

A collision in an equipped fleet generates the same post-accident legal exposure as one anywhere else. A bodily injury claimant with a retained attorney, litigation funding behind them and a high-verdict jurisdiction produces the same severity outcome however well the CSA score was trending. Camera footage runs both directions in discovery: a distraction clip from six months earlier can anchor a corporate-negligence narrative as readily as clean footage can exonerate a driver.

The severity numbers are moving on their own trend. Average commercial auto liability severity grew roughly 8% a year for nine consecutive years against about 3% general economic inflation, with total claim severity up 64% since 2015. In 2024, 135 corporate-defendant lawsuits produced nuclear verdicts, up 52%, with total value of $31.3 billion, a 116% jump, and 49 of those exceeded $100 million, an 81.5% increase. Auto cases accounted for 22.8% of all verdicts above $10 million, and attorney representation in commercial auto liability claims now runs above 50%.

The arithmetic closes the argument. A carrier achieving a genuine 30% frequency reduction from a mature camera program, starting at a 113 combined ratio, moves toward the high 80s. That is real. But severity compounding at 8% drifts the loss ratio upward by roughly 8 points a year, so the frequency gain is absorbed within four to five years.

Which means a camera program is a level shift, not a trend change, and it is priced against a trend. Unless the program also shortens case duration or improves fault attribution at trial, the credit a carrier files in 2027 on four years of comparative data is buying time against a severity curve that has not bent since 2015.

Further Reading

Sources