ADAS Creates a Frequency-Severity Paradox for Auto Insurers: Fewer Crashes, Costlier Claims

From tracking the total loss rate trend across three consecutive CCC annual reports and cross-referencing with carrier earnings severity commentary, a consistent pattern is visible: ADAS is not simply reducing losses but restructuring the entire auto loss distribution. The technology prevents the frequent, low-cost collisions that once anchored the left tail of the severity curve while simultaneously inflating the cost of every claim that does occur. For personal auto pricing actuaries, this is not an incremental trend adjustment. It is a structural change in the relationship between frequency and severity that standard GLM frameworks were not designed to capture.

The data supporting this paradox arrived from two directions in early 2026. CCC Intelligent Solutions published its annual Crash Course report, subtitled “Complexity Compounds,” documenting a record 23.1% total loss rate and calibrations appearing on 28.3% of repairable estimates. Within days, the Highway Loss Data Institute released research on Mazda’s ADAS bundles showing property damage liability claim frequency reductions of up to 39% for the most comprehensive system packages, alongside severity increases driven by sensor replacement and recalibration costs. Read together, these two datasets define the paradox that personal auto actuaries must now price through.

The Frequency Side: IIHS-HLDI Quantifies the ADAS Safety Dividend

The Insurance Institute for Highway Safety and its data arm, the Highway Loss Data Institute, published a landmark study in March 2026 examining six feature bundles and four standalone systems across 2015 to 2023 Mazda vehicles. The research quantified something actuaries had theorized but lacked granular insurance data to confirm: ADAS features produce compounding safety benefits when bundled together, and updated system generations deliver better results than earlier versions.

The basic bundle, consisting of front automatic emergency braking (AEB) and forward collision warning, was associated with a 13% reduction in property damage liability (PDL) claim frequency and a 9% reduction in bodily injury liability (BIL) claim frequency. Each successive feature addition produced incremental reductions. Blind spot monitoring with rear cross-traffic alert added nearly 10% fewer PDL claims and 13% fewer BIL claims. The most comprehensive bundle, which layered on pedestrian detection, adaptive cruise control, high-beam assist, lane departure warning and prevention, rear AEB, and a driver attention alert system, produced the headline figure: a 39% reduction in PDL claim frequency and a 21% reduction in BIL claim frequency.

The 21% BIL reduction deserves a caveat. HLDI noted the result was not statistically significant at conventional confidence levels, reflecting the smaller sample size for bodily injury claims relative to property damage. But the directional consistency across all bundles is informative for trend selection even where individual coefficients lack tight confidence intervals.

HLDI chief insurance operations officer Matt Moore identified the mechanism driving these results: crash avoidance systems primarily prevent collisions at lower speeds. Front AEB prevents or mitigates rear-end crashes. Rear AEB eliminates many of the low-speed parking lot incidents that constitute a substantial share of all property damage claims. Lane departure prevention stops single-vehicle run-off-road crashes. Each system targets the high-frequency, low-severity portion of the loss distribution.

This selectivity is what creates the paradox. ADAS does not reduce crashes uniformly across the severity spectrum. It preferentially eliminates the cheapest claims, leaving a residual pool of higher-speed, higher-energy collisions that are inherently more expensive.

The Severity Side: CCC Documents the Calibration Cost Layer

CCC Intelligent Solutions’ Crash Course 2026 report, released March 31, provides the severity counterpart to HLDI’s frequency data. The report documents several structural severity drivers that compound on ADAS-equipped vehicles.

Calibration penetration. ADAS calibrations appeared on 28.3% of all repairable estimates in 2025, up from 21.8% the previous year, a 30% relative increase in a single year. In DRP shops, which handle higher volumes of insured claims, the calibration rate reached 35.6% by Q3 2025. Average calibration fees ran $485.56 per procedure. Diagnostic scan fees averaged $149.10 and appeared on 87.7% of collision repair estimates. Industry projections from Caliber Insights estimate 65% of repairs will require calibration by 2026 and 75% by 2027.

The supplement problem. More than half of calibrations (51.5%) appear as supplements rather than on the initial repair estimate. This creates a systematic pattern in loss development. Case-incurred triangles for auto physical damage will show calibration-driven supplement emergence that did not exist five years ago. For short-tail APD reserving, this means early development factors are shifting, and actuaries relying on historical supplement patterns should test whether the calibration component warrants a separate development adjustment.

As we documented in our CCC Crash Course 2026 analysis, the 1.7% headline total cost of repair (TCOR) increase, the lowest since 2017, masks divergent structural sub-trends. Newer vehicles (six years old or less) averaged $5,721 in repair costs versus $3,682 for older vehicles, a 55% gap. ADAS technology concentration in the newer fleet explains most of this differential.

Cycle time severity. Calibration requirements extend repair duration. CCC data shows keys-to-keys cycle time averaging roughly 13 days for claims with no calibrations, 15.5 days for single-calibration claims, and 17 or more days for multiple calibrations. Vehicles from model year 2019 and newer with three or more ADAS features averaged 21.5 days in the shop versus 17.9 days for pre-2015 vehicles with no ADAS. Each additional day adds rental car or loss-of-use expense to the total claim cost.

The Correlation Shift: Why Independence Assumptions Break Down

Standard actuarial pricing models for personal auto treat frequency and severity as independent random variables. Pure premium equals frequency times severity, and each component is trended separately. This multiplicative independence assumption is embedded in most GLM frameworks, in the frequency-severity split used for CAS Exam 5 ratemaking problems, and in the trend selection methodology that drives rate filings across all 50 states.

ADAS breaks this independence. When the same technology simultaneously reduces claim frequency and increases claim severity, the two components are negatively correlated at the vehicle level. A pricing actuary who selects a -5% frequency trend and a +8% severity trend for ADAS-equipped vehicles and multiplies them together to get the pure premium trend is treating those movements as coincidental. They are not. They are causally linked through the same underlying mechanism: ADAS prevents low-cost crashes and makes surviving crashes more expensive to repair.

The practical consequence is that traditional multiplicative models will overstate total loss costs for ADAS-heavy cohorts. Consider a simplified example:

Scenario	Frequency	Severity	Pure Premium
Pre-ADAS baseline	10.0 claims per 100	$4,200	$420
ADAS: independent model	7.0 claims per 100	$5,500	$385
ADAS: correlated model	7.0 claims per 100	$4,900	$343

In the independent model, the severity increase of $1,300 applies to all surviving claims equally. In the correlated model, the severity increase is smaller ($700) because the eliminated claims were the cheapest ones, so removing them shifts the average upward mechanically rather than through an actual cost increase on comparable repairs. The $42 difference per vehicle (about 10% of pure premium) is the mispricing cost of ignoring the correlation structure.

Actuaries working in personal auto pricing should decompose the severity trend into two components: (1) true repair cost inflation on claims that would have occurred regardless of ADAS, and (2) compositional severity inflation caused by the removal of low-cost claims from the frequency distribution. Only the first component should be projected forward on a compounding basis. The second represents a one-time distributional shift that recalibrates the severity level but does not generate ongoing trend.

HLDI Data: The Full Bundle Picture

The HLDI Mazda study provides the most granular insurance-data view of ADAS bundle effects published to date. The progression from basic to comprehensive bundles reveals how each additional system shifts the frequency-severity balance.

ADAS Bundle	PDL Frequency Change	BIL Frequency Change	Severity Direction
Basic (front AEB + FCW)	-13%	-9%	Up (moderate)
+ BSM + RCTA	-10% incremental	-13% incremental	Up (moderate)
Comprehensive (all features)	-39% total	-21% total	Up (significant)

Critically, HLDI found that when frequency and severity were combined into overall loss calculations, nearly all ADAS bundles were associated with lower total losses under property damage coverage. The frequency reduction outweighed the severity increase. This is the core finding that pricing actuaries should anchor on: ADAS-equipped vehicles are genuinely less costly to insure in aggregate, but only if you model the frequency and severity interaction correctly.

A carrier that captures the frequency reduction through vehicle-level ADAS discounts but fails to account for the severity composition shift will underprice the residual severity risk. A carrier that prices only on observed severity trends without separating the ADAS compositional effect from true repair cost inflation will overprice ADAS-equipped vehicles relative to their actual expected loss.

Bodily Injury: Where the Paradox Cuts Deepest

CCC’s 2026 data shows bodily injury severity rising 10.3% year over year and 32% over four years, now accounting for 52.4% of total liability dollars paid. BI frequency also increased 11% over the past two years. This trajectory, running at roughly four times general inflation, reflects social inflation dynamics that are largely independent of ADAS technology.

ADAS complicates the BI picture in a specific way. If ADAS prevents many low-speed collisions that produce minor soft-tissue injury claims, the remaining BI claims will skew toward higher-energy impacts with more serious injuries and higher litigation values. HLDI’s 21% BIL frequency reduction for the comprehensive bundle, though not statistically significant, suggests this compositional shift is occurring. The BIL claims that survive ADAS prevention are the ones involving higher-speed collisions, pedestrian strikes in complex traffic environments, and multi-vehicle incidents where ADAS mitigation may reduce but not eliminate impact severity.

For BI reserving, this means the average cost per surviving claim will rise faster than the medical CPI or general litigation cost trend would suggest. The standard approach of selecting a single BI severity trend based on historical claim data does not distinguish between true cost escalation and the compositional effect of removing cheaper claims from the pool. Actuaries should monitor the BI severity trend by vehicle ADAS equipment level, where data is available, to isolate these two effects.

Telematics as the Data Bridge

The carriers best positioned to price through the ADAS paradox are those with deep telematics data linking individual vehicle ADAS configurations to observed loss patterns. Progressive’s 21 million telematics-connected policyholders, growing at 28% CAGR since 2018, represent the largest single-carrier dataset capable of distinguishing ADAS-equipped from non-ADAS loss patterns at the individual policy level.

Progressive’s CEO Susan Griffith told investors in early 2026 that the company’s usage-based insurance capabilities allow it to understand driving behavior and technology changes through access to tens of billions of driving miles annually. The company accepts data streams directly from OEMs and third parties on an individual-vehicle basis with consumer consent, which means Progressive can observe the difference between a 2024 model year vehicle with a full ADAS suite and a 2016 model year with no crash avoidance systems, controlling for driver behavior, territory, and other rating variables.

This data advantage has direct pricing implications. A carrier with vehicle-level ADAS data can decompose the frequency-severity relationship at the policy level rather than relying on fleet-wide averages. It can identify that a specific ADAS bundle reduces frequency by 25% for a given driver segment while increasing per-claim severity by 15%, and price to the net effect. Carriers without this data must rely on crude vehicle-age or model-year proxies that blend ADAS-equipped and non-ADAS vehicles within the same rating class.

The competitive asymmetry is significant. As ADAS penetration grows toward fleet majority, carriers with granular ADAS pricing data will selectively attract the lower-risk ADAS-equipped vehicles with accurate discounts, while carriers using blended averages will retain the higher-risk non-ADAS vehicles at rates that underestimate their true loss potential. This is a classic adverse selection spiral accelerated by technology adoption curves.

ADAS Fleet Penetration: The Timeline That Drives the Pricing Urgency

The ADAS paradox is not a future concern. S&P Global Mobility forecasts that by 2029, eight distinct ADAS systems will be present in half or more of all registered U.S. vehicles, up from three systems meeting that threshold by 2027. Rear cameras will reach 78% fleet penetration, front automatic emergency braking 55%, and lane departure and blind-spot monitoring 56%. The average age of the U.S. light vehicle fleet reached 12.8 years in 2025, meaning the transition from a minority-ADAS to a majority-ADAS fleet will take the better part of a decade as older vehicles age out.

During this transition period, the personal auto book operates as two overlapping populations with fundamentally different loss characteristics. The newer, ADAS-equipped cohort produces fewer claims at higher per-claim costs. The older, non-ADAS cohort maintains historical frequency patterns but faces its own severity pressure from rising total loss rates (45.3% for vehicles 13 years and older, per CCC) and tariff-driven parts inflation.

Pricing actuaries who use a single set of frequency and severity trend assumptions for the entire personal auto book are averaging across two populations whose loss characteristics are diverging. The correct approach segments trend assumptions by vehicle ADAS equipment level, or at minimum by model year cohort as a proxy, and explicitly models the mix shift as newer ADAS vehicles become a larger share of the portfolio over the projection period.

State Rate Filing Implications

ADAS-segmented pricing introduces regulatory complexity. Several states restrict vehicle-level rating variables that regulators view as proxies for socioeconomic status. Because newer vehicles with comprehensive ADAS packages tend to be more expensive and their owners tend to have higher incomes, an ADAS equipment rating variable could face scrutiny under unfair discrimination standards.

The actuarial defense rests on demonstrating that ADAS reduces expected losses. HLDI’s data showing lower overall losses for ADAS-equipped vehicles provides the statistical foundation. Under CAS ratemaking principles and ASOP No. 12 (Risk Classification), a rating variable is actuarially sound if it is a statistically significant predictor of expected losses and is not unfairly discriminatory. ADAS equipment status meets the first criterion based on the HLDI evidence. The second criterion requires showing that the variable reflects risk differences rather than prohibited class correlations.

The AI model validation framework being developed by state regulators provides guidance here. When a rating variable correlates with a protected class but also has a direct causal relationship to loss costs (ADAS physically prevents crashes, which directly reduces losses), most regulatory frameworks permit its use with adequate documentation. The pricing actuary’s filing should include HLDI citations, internal loss experience segmented by ADAS level, and a disparate impact analysis demonstrating that any correlation with protected classes is incidental to the risk-based effect.

Some carriers are approaching ADAS pricing indirectly through telematics rather than direct vehicle equipment rating. A telematics program that observes driving behavior and captures ADAS-influenced loss patterns (fewer hard braking events, fewer lane departures) achieves a similar pricing segmentation without explicitly rating on vehicle equipment. This approach is more defensible in states with restrictive rate regulation because the rating variable is behavior, not vehicle characteristics.

Why This Matters for Personal Auto Pricing

The ADAS frequency-severity paradox creates five specific challenges for pricing actuaries filing personal auto rates in 2026 and 2027.

1. Trend selection requires decomposition. A single frequency trend and a single severity trend applied to the entire book will produce pure premium estimates that are wrong for both ADAS-equipped and non-ADAS vehicles. At minimum, severity trends should be decomposed into base repair inflation and ADAS compositional effects. Frequency trends should account for the growing ADAS share of the portfolio.

2. Loss development patterns are changing. The 51.5% supplement rate for calibrations creates a new source of case-incurred development in APD that did not exist in the historical triangle. Reserving actuaries should test whether short-tail development factors from pre-2023 periods remain appropriate when calibrations now constitute a material share of supplement costs.

3. Vehicle age relativities need recalibration. The $2,039 gap between newer-vehicle and older-vehicle repair costs documented by CCC ($5,721 versus $3,682) is wider than most class plan age-of-vehicle relativities reflect. As the fleet ages and the ADAS-equipped share grows, these relativities should be updated annually rather than on the historical three-to-five-year cycle.

4. Total loss assumptions interact with ADAS. The 23.1% total loss rate is not stable. Fleet aging and rising repair costs on ADAS-equipped vehicles push it higher. For actuaries modeling the APD severity distribution, the total loss frequency assumption directly affects the shape of the repairable claim distribution and the expected severity conditional on repair.

5. Competitive dynamics favor data-rich carriers. Carriers with telematics data can price ADAS risk accurately at the vehicle level. Carriers without it will face adverse selection as better risks migrate to carriers offering ADAS-based discounts, leaving a residual pool of higher-cost, non-ADAS vehicles priced on blended averages that understate their true loss cost.

The Forward View: From Paradox to Pricing Framework

The ADAS frequency-severity paradox is not permanent. As ADAS penetration approaches fleet saturation, the compositional severity effect will stabilize because there will be fewer non-ADAS claims to remove from the pool. At that point, the frequency benefit becomes the dominant pricing factor, and overall loss costs for the fleet should decline in absolute terms, assuming repair cost inflation does not fully offset the frequency reduction.

The transition period, roughly 2025 through 2032 based on fleet replacement timelines, is where the pricing risk concentrates. During this window, the correlation between frequency and severity is strongest and most variable. Each model year brings new ADAS features with different safety effectiveness profiles. Each year of fleet aging shifts the non-ADAS share of claims toward total loss. And each year of calibration cost growth adds a new severity layer to the ADAS-equipped cohort.

Personal auto actuaries who build explicit ADAS segmentation into their pricing frameworks now will price more accurately through this transition. Those who continue treating frequency and severity as independent will systematically misprice both halves of the portfolio, undercharging non-ADAS vehicles where frequency remains high and overcharging ADAS vehicles where the net loss cost is lower than blended trends suggest.

The data exists to do this correctly. HLDI’s bundle-level analysis, CCC’s calibration penetration tracking, OEM telematics feeds, and carrier-specific loss experience by vehicle equipment level collectively provide the inputs for an ADAS-adjusted pricing model. The actuarial profession has the tools. The question is whether carriers will deploy them before the competitive market forces their hand.