Thanks to advanced driver assistance systems, quantifying risk will never be the same

Advanced driver assistance systems (ADAS), which partially automate driving and boost vehicular safety, are saving lives and reducing accident frequency.

Confidently quantifying the actuarial value of these safety systems, however, has been elusive. Despite lowering frequency, ADAS generally have eye-popping repair bills that push severity upwards. ADAS are not perfect but evolving. Safety technologies change year-by-year and by make and model. As they change, they can also introduce new risks and may contribute to accidents in some situations. And the relationship between people and ADAS? Well, it’s complicated.

As for useful data, there is little to be found when only a small percentage of the United States fleet contains the automated technology and when auto manufacturers are loath to share information. Even when data accessibility improves, ever-changing technological advances will pressure both historical and new variables to be increasingly dynamic. Auto insurance actuaries may need to adapt strategies from their cyber insurance colleagues.

Monitoring and quantifying constellations of so many moving parts will not be the only important metric as ADAS become more ubiquitous. For actuaries, ADAS provide a clearer window into the safety promises being made by driverless car manufacturers.

Falling Frequency

ADAS reduce accidents and insurance claims. Roughly one-half of the 2% to 3% auto claim frequency reductions from 2017 to mid-year 2019 can be attributed to ADAS, says Susanna Gotsch, director and industry analyst for CCC Information Services Inc., which serves the automotive, insurance and collision repair industries.

Since the safety technologies are currently effective for a small percentage of use cases, a dramatic impact on accident reduction is yet to come, according to a 2018 analysis by SBD Automotive provided to Actuarial Review. ADAS will reduce accidents by 19% to 24% by 2030, plateauing at a 30% reduction by 2060, the U.K.-based international automobile research and consulting firm anticipates.

CCC’s Crash Course reports for 2018 and 2019 predict ADAS will contribute to a 5% decline in vehicular crashes by 2022 from a baseline 1995 statistic and 20% by 2030 (see Figure 1). Drivers using ADAS have observed the technologies’ ability to reduce crashes. In a Consumer Reports (CR) survey published in August, 57% of respondents say at least one ADAS feature prevented a crash. Subscribers participating in CR’s survey on ADAS own model year 2015 to 2019 private passenger vehicles with one or more of the safety systems (see Sidebar).

Manufacturers roll out ADAS by feature. It can take a while — even decades — for available ADAS to transition from optional to standard to common on U.S. roads. For example, it has taken 17 years, from 1995 to 2012 model years, for electronic stability control to become standard. It will take until the 2030s for this safety feature to be in 95% of the U.S. auto fleet, according to the Insurance Institute of Highway Safety (IIHS).¹

One much-hailed feature, automatic emergency braking (AEB), should become standard in most new cars by September 1, 2022. Thanks to commitments jointly brokered by IIHS and the National Highway Traffic Safety Administration (NHTSA) in 2015, 20 car manufacturers, which collectively produce 99% of private passenger vehicles, promised to install AEB.

Expanding the population of ADAS-featured cars, however, it not merely a matter of manufacturers installing them as standard. Assuring that the U.S. fleet grows with ADAS features is ultimately up to consumers — and they are holding on to their wheels longer than ever. In 2019, the average car in use is 11.8 years old in the United States, according to IHS Markit estimates published in June. The organization expects a 22% increase in vehicles 16 years or older from 2018 to 2023.

The trend is not surprising. Cars are better built than a generation and a half ago. Maintenance costs are lower for older cars, allowing owners to keep their vehicles longer and to avoid nagging monthly payments. Both new and late model used cars, which are most likely to have ADAS features, are financially out of reach for large portions of the U.S. population.

Consider that the median annual household income in June 2019 was $64,430, according to  Sentier Research. Kelley Blue Book estimated in August 2019 that new light vehicles cost more than half that amount —$37,169 — reflecting a 3.5% price increase for these vehicles from July 2018 to July 2019.

The average price of a used car is not cheap either, rising to a record-breaking $20,247 for the first quarter of 2019, according to Edmunds’ “Used Vehicle Report” released in June. Typical used cars are newer and have lower mileage and sport utility vehicle bodies, according to Edmunds. The report also noted that “automated driving hasn’t yet struck a chord with the masses.”

There is enough demand for cars with the Society of Automotive Engineers (SAE) Level 2 classification — the latest level of automation — by those who can afford them. A comparison of first quarter 2018 to 2019 shows sales grew by 322%, according to Canalys, a technology research firm (see Figure 2). Specifically, customers bought more than 250,000 cars with Level 2 technology such as Tesla’s Autopilot, accounting for 7% of new cars, the firm announced in May.

Overall, ADAS are making a positive impact on car safety. However, the evolving technologies are not perfect. In some cases, they can increase claim frequency, reports the Highway Loss Data Institute (HLDI). Specifically, forward collision warning (FCW) and front automatic emergency braking (AEB) combined increase the rear-end crash rate by 20%.² Lane departure warning (LDW) increases claim frequency by 2.2% for collision and 6.3% for bodily injury (BI). Parking sensors increase BI claim frequency by 1% and rear cameras push up collision claim frequency by 0.9%.³ Furthermore, The Wall Street Journal reports in August that more than 400 complaints were filed with NHTSA over the past three years mostly because brakes were unnecessarily engaging or failing to deploy.

ADAS also introduce risks similar to those presented by driverless cars. “Additional complexity of systems can increase potential for failures,” explains Jonathan Charak, vice president and emerging solutions director at Zurich North America and vice chair of the Casualty Actuarial Society’s (CAS) Automated Vehicle Task Force. For example, sensors can fail, become dirty and not work properly. However, he notes, redundancy systems that overlap with radar, lidar, cameras and other computer vision systems can help limit this risk. On top of that, hackability remains a concern.

The Severity Factor

The cost of repairing ADAS is a one-two punch for insurers. The first blow is that repair costs are pressuring claim costs upward. Repairing a minor front or rear collision on a car with ADAS costs $5,300 — almost two and half times the expense for vehicles without ADAS, according to the AAA report, “Advanced Driver Assistance Systems (ADAS) Repair Costs,” published in 2018. Damage to ADAS windshields, bumpers and door mirrors from minor incidents can cost $3,000 more to fix compared to conventional vehicles lacking the technology.

“The challenge with mapping out long-term impacts to severity,” Gotsch explains, “is there are many more variables compared to frequency.” Although Gotsch is pleased with the 20 manufacturers committing to install AEB in 2022, she is concerned about the overall lack of standardization. The technology varies and is located in different places on vehicles depending on the make and model year. So far, while there has been a small but meaningful decline in frequency from ADAS, repair costs for a vehicle equipped with ADAS are trending higher for the same damage.

The  second blow  to insurers is that expensive car repairs can encourage policyholders to file claims that they formerly covered themselves to avoid premium increases. AAA, which estimates that one out of three Americans cannot afford to cover an unexpected repair bill of $500, “strongly” urges consumers to check their insurance policies and be cognizant of ADAS-related repair costs.

Finding a technician to repair the car correctly is another challenge. ADAS repairs are nuanced, Charak explains. “Repair shops have to specialize in certain technologies or in how specific car manufacturers create systems, which lowers the availability of repair shops that can fix the highly specialized cars.”

Diagnostic tools also need to improve to ensure that cars are being repaired correctly, explains David Zuby, senior vice president of vehicle research for IIHS and HLDI.

Post-repair sensors can be misaligned because the calibration process can miss a problem, resulting in cars leaving the shop with compromised ADAS. Roosevelt C. Mosley, a principal with Pinnacle Actuarial Resources, Inc., says this is a “real issue” for insurers because “if the repair work was not done properly, the system may not function as it is supposed to,” and there is no way for drivers and their insurers to know.

There is some good news, however. In some cases, the technology can mitigate crashes, reducing claims severity and frequency, HLDI finds. Front AEB, for example, reduced overall losses by 2.6%. The calculation includes a 13% reduction of property damage liability claim frequency and a 23.2% drop in BI frequency.

CCC found that FCW with or without AEB is reducing the impact of crashes that do occur and may be preventing low-speed crashes altogether. The cars with ADAS show a smaller share of volume within both the lowest and the highest Delta-V (a measure of vehicle velocity) ranges, Gotsch explains. This suggests that “ADAS is also slowing the speed of the vehicle prior to impact for those crashes that otherwise might have had higher Delta-Vs.”

Human Interaction

The more cars operate with automated auto technologies, the more critical human interaction becomes. Determining fault and assigning liability will become more complex. This is not just a question of who — or what — was in control, but also the circumstances around vehicular accidents.

This is a point made by the NHTSA bulletin, “Critical Reasons for Crashes Investigated in the National Motor Vehicle Crash Causation Survey,” released in March 2018. The paper, which originally concluded that human error causes 93% of accidents, backpedals on the word “cause” from its much-quoted 2008 “National Motor Vehicle Crash Causation Survey,” explaining its initial study was not intended to blame drivers. Rather, it states that drivers are the “critical reason” for 94% of accidents (94% is an updated percentage from the original).

For its part, the CAS Automated Vehicles Task Force reviewed the same data in 2015 and concluded human error was the cause of 74% of accidents (AR May/June 2018). The ground-breaking CAS study also introduces the risks associated with automated technology, especially as it relates to human interaction.

So far, studies by IIHS show that drivers can rely too much on automation, allowing themselves to be distracted, miss warnings or have trouble taking control of vehicles, which suggests a need for better driver training. The CAS report also warns that automated technology can result in an overreliance on automation that could lead to driver skill deterioration, less engagement and higher pass-off risk.⁴ This is critical when inclement weather or Bambi darting across the highway requires human intervention. “Drivers have to pay attention because ADAS does not replace their roles,” Charak says.

Some driver overreliance could be the result of marketing. A 2019 IIHS study reveals that some ADAS branding names can give drivers the false impression that their attention is unnecessary with SAE Level 2 technology. The fact remains that there are limits to automation’s ability to reliably manage lane keeping and speed control all the time. Specifically, 48% of about 2,000 participants believe the name of Tesla’s “Autopilot” means it is safe to drive hands-free.⁵ For other SAE Level 2 packages, such as Cadillac’s Super Cruise or BMW’s Driving Assistance Plus, 33% or fewer believed the same.

Most fundamentally, ADAS has to be turned on to work. Nearly 100% of the blind-spot detection, 97% of rear cross-traffic alert systems and 93% FCW were operating when arriving at the dealership for service, according to a 2018 IIHS study. However, consumers are not loving LDW and lane keeping assist (LKA). Only 51% of these features were turned on upon arrival to the dealership, according to the study.⁶

Toward Quantification

ADAS study results provide actuaries with a starting point for knowing what is relevant, Mosley says. Quantifying the true insurance value of ADAS, he adds, is tough without sufficient granular data.

“Collision avoidance capabilities of each ADAS differ between manufacturers,” observes Alain Dunoyer, head of the autonomous car division for SBD Automotive. “For actuaries to correctly assess the value of current ADAS, they need to understand their capabilities.” Some systems only work correctly on the highway, only operate in good weather conditions or only react to specific “objects” on the road, he explains. There are also differences in the car manufacturer’s fitment strategy, such as optional or standard, and the activation default.

As automation evolves, risk factors and their value will change. Essentially, the less people are actually driving the cars, the lower the value of historic driver-related risk factors. Tried-and-true variables, such as driver age or moving violations, will transition from high to medium importance when cars reach SAE Level 4 automation, according to SBD Automotive.

Moreover, factors related to vehicle characteristics such as make, model, age and annual mileage will continue to be highly important. New risk factors from automation, such as driving skill requirements during pass-off or handover, previous experience with handling ADAS and the proportion of driving on different road types will also affect pricing.

Being able to identify the presence of ADAS features on a per-vehicle basis would be a big step forward, experts agree, but manufacturers generally are not sharing the information. Zuby of the IIHS and HLDI is pushing for a public national database of vehicles with ADAS and provided testimony last year to NHTSA.

Actuaries can get clued in on which cars have ADAS bundles through vehicle identification numbers (VINs), Zuby advises, because the safety technology tends to be linked to particular engines. However, he cautions, “A lot of things that we have learned that affect risk of vehicles are not encoded in the VIN.” Meanwhile, vendors including Verisk Analytics,
AutoData Solutions and Swiss Re are working to provide actuarial data and tools.

Mosley is working with Carfax, Inc. to build a quantitative product for insurers. He anticipates that in the next several months, insurers will learn how many ADAS features are in the current U.S. fleet, understand specific adoption of features by vehicle, and identify the most important variables for pricing insurance. “There are so many ADAS variables that just eliminating those with little or no value will be helpful,” he says.

Ironically, as manufacturers improve ADAS, the safety promised by autonomous cars could already be available in conventional cars. “It is not at all clear to me that you need full automation to reduce the number of crashes that occur,” Zuby says.

In the meantime, SBD Automotive expects very limited SAE Level 3 technology around 2022, Dunoyer says. “It should have been two years ago,” in the Audi A8 he adds, but the United Nations Economic Commission for Europe is working out a vehicle homologation process that automakers from around the world will try to follow. Homologation is a standardization process for vehicles or particular vehicle components to meet requirements set by various statutory or regulatory parties.

Conclusion

While ADAS safety features are generally showing positive results, they will not be able to address all accident scenarios. Since Americans are increasingly holding on to older cars and ADAS standard fitment is incremental, it will take at least 10 to 15 years for ADAS to become commonplace. That timing is not much different from when autonomous cars are expected on the market.

While greatly celebrated for its advantages, ADAS are not perfect. They are limited to specific situations, can be compromised through dirt or miscalibration, and sometimes contribute to accidents. The complex interaction between drivers and ADAS also carries risk when people overestimate the safety systems’ abilities, do not take control of the wheel and are confused or distracted by stimulating dashboards, beeps or rumblings.

As ADAS evolve, the technology will do more than complicate pricing and risk selection through new and changing variables. As assumptions and rising expectations about ADAS continue, it is essential for the insurance industry, including actuaries, to educate consumers, regulators, lawmakers and other stakeholders.

Annmarie Geddes Baribeau has been covering insurance and actuarial topics for nearly 30 years. Her blog can be found at www. insurancecommunicators.com.

¹ “Predicted availability of safety features on registered vehicles — a 2016 update,” Highway Loss Data Institute Bulletin 33:15, September 2016.
² Cicchino, Jessica B., “Effectiveness of forward collision warning and autonomous emergency braking systems in reducing front-to-rear crash rates,” Accident Analysis & Prevention, February 2017.
³ “Compendium of HLDI Collision Avoidance Research,” HILDI Bulletin 35:34, September 2018.
⁴ According to the CAS Automated Vehicles Task Force report, pass-off risk is “the risk that is created when the vehicle goes from technological control back to human control.”
⁵ Teoh, Eric R., “What’s in a name? Drivers’ perceptions of the use of five SAE Level 2 driving automation systems,” Insurance Institute for Highway Safety, June 2019.
⁶ Reagan, Ian J., et al., “Crash avoidance and driver assistance technologies — are they used?” Transportation Research Part F, January 2018.

Some ADAS Features At-A-Glance