If artificial intelligence is changing the whole insurance industry, it surely starts with car insurance, the most standardized, and therefore easiest segment for machines to take over.
Since 2017, the likes of Ant Financial, PICC, Ping An, Sensetime (an A.I. company), Tencent and ZhongAn have launched A.I. tools that assess motor damage in real time, and offer fast payouts and ancillary services such as helping motorists locate nearby garages or get a tow.
Out of this scramble, the winners so far appear to be Ping An, an actual insurance company, which now claims 23.2% market share in the motor insurance in China which accounts for Rmb 615 billion annual premium, and Ant Financial, which operates a platform for smaller insurers.
Zhu Yougang, deputy managing director of Ping An Property & Casualty, is confident that other internet tech companies lack enough data on car accidents to beat Ping An’s offering.
“We deal with 50,000 cases each day, and now we have accumulated more than one billion photos of accident claims,” he said.
He says being an insurance company with a strong tech side has been a competitive advantage, despite some early friction. His team, responsible for P&C business, used to quarrel with the technology team.
“We fight hard with each other,” Zhu said. “We think they don’t know business, they think that we don’t know technology. We couldn’t have made it without the powerful push of our Chairman [Peter] Ma.”
What this means in practical terms is that, unlike some internet companies competing for a single solution, Ping An develops solutions across the entire chain of insurance.
“We are not only looking to automate damage assessment,” Zhu said. “We also include insurance underwriting, policy application and all related procedures. The whole group is moving towards the aim of ‘no paper issued.’”
Creating a model
To create a reliable A.I. model for car insurance, Zhu explains, it needs three different kinds of data.
First, driver information, including gender, profession, and past records of traffic violations or previous insurance claims.
Second is data for managing risk and controlling against fraud. This involves drivers’ credit data and scanning social media and other readily available sources to develop profiles.
Third is data related to car accidents. This is the part that represents a barrier to internet companies, as histories remain proprietary information of the insurers.
We have accumulated more than one billion photos of accident claims.Zhu Yougang, Ping An
Zhu told DigFin that his team has built a data library around car models, with a growing set of photos taken by drivers of damage, and a list of repair fees by region. Ping An has labelled and categorized all these images. The firm’s model will only improve over time as more people use it.
Ping An and PICC (People’s Insurance Company of China, the country’s oldest and largest P&C provider) together have more than 50% market share in motor insurance.
Ant Financial is the most advanced in breaking the duopoly, thanks to its dingsunbao service (“loss assessment master”), a platform in which insurers can leverage to assess the damage through A.I. instead of manual work. And the accuracy rate is about 98%, claimed the company.
Ant financial applied the same strategy that Taobao used to grow up to the country’s biggest e-commerce site: opening for free.
So insurers can use a free service to save manual work and cost. A typical claim settlement in China requires 5.18 manual interventions, from initial reporting to sending personnel to the scene, and then identifying invoices sent from an auto repair shop. It’s a pain point for the industry.
So far, China Taiping, China Continent Insurance, Sunshine Insurance Group and AXA Tianping participate, while Ant Financial gathers data at an industry level. Eventually it can build a platform for the whole industry to outsource claims management.
At the same time, it is pushing its electronic payments rail, AliPay, to manage car-insurance compensation.
Three months after Ant Financial released Dongsunbao in 2017, Ping An responded by opening its own A.I. tool to the whole industry, via One Connect, the group’s technology arm.
PICC, meanwhile, is developing its own A.I. solution in partnership with ArcSoft, a software company.
Zhu told DigFin, that before Ping An opened the solution for damage assessing to the industry, it has been used in Ping An’s back office for more than one year.
Only Ping An’s solution has been tested in production environment before industry-wide release, Zhu said, and the accuracy rate is 98.75%.
Ping An’s solutions are having an immediate impact. In 2018, the company’s combined ratio for car insurance (a measure of insurer profitability, it’s the sum of costs devided by the earned premiums) fell 0.1 percent to 97.4% while some insurers are suffering conbined ratios higher than 100%.
Over the same period, the company’s market share grows 0.5 percent.
But operations are just one aspect of what digital insurance can do. More important is restoring trust to a system in which, traditionally, insurers have to assume customers might be lying about a claim, Ping An’s Zhu told DigFin.
In December, Ping An piloted a service called Trust Claim in Guangdong province. The algo offers a line of credit for claims on a shifting basis, depending on drivers’ habits on the road and their credit histories.
The first version of the pilot is for unilateral accidents only, that is, those that involve just one car.
We let them decide for themselves.Zhu Yougang, Ping An
The algorithm covers up to Rmb4000 for the best profiles, so drivers can submit a claim and get paid within three minutes after sending through images of the damage on their mobile phones.
The solution also settles the kind of disputes between the car owner and the insurer on whether a small part like a rearview mirror should be repaired or be replaced after the accident.
“As long as the customer has enough credit, we let them decide for themselves,” said Zhu.
Ping An won’t release the number of users on its Trust Claim, but Zhu says 50% of users have posted about their experience on WeChat (the messaging app run by rival Tencent), and that 30% of users are new to Ping An.
DigFin has talked with several car owners in mainland China about their small accident experience. Ping An, PICC, and SunShine are among the brands mentioned by these car owners.
But A.I. technology still seems to be far from ordinary people’s lives. All these interviewees told DigFin that their claims are still handled by insurance agents sent to accident scenes. And the whole compensation procedure still takes days or weeks.
This seems likely to change as the quality of companies’ data libraries improves.
With more usage of Ping An’s app and its Trust Claim, it will accumulate more data per incident.
But how long will mass adoption take? Two years into the launch of the first A.I. for auto insurance, it’s still niche in China. Insurers like Ping An and PICC don’t have to fear an internet company…yet.