How AI is speeding up insurance claims processing
Join today’s top leaders online at the Data Summit on March 9. Register here.
The immediate consequences of a car accident are difficult for all parties involved. Drivers need to process trauma, treat injuries with medical care, exchange information, and figure out how to get their damaged cars repaired or replaced. Insurance companies must conduct damage assessments and determine the most efficient and least expensive route to settling claims. Adjusters, body shop workers and various other parties all have a say in the process, which can often take weeks.
In addition to the time and trauma involved, accidents generate a lot of data, whether it’s images of damaged parts or associated documents from police reports. Additionally, crash frequency – 2019 saw nearly 6.8 million vehicular crashes in the US alone – means a large volume of data to deal with all the time. Car insurance claims arise not only from accidents, but also from other types of damage, such as flooding and falling trees on bumpers.
AI is on the rise
These collective factors make a particularly compelling case for implementing artificial intelligence in claims handling, says John Goodson, chief technology officer at CCC Intelligent Solutions, a provider of technology solutions for the healthcare industries. automobile and insurance. (CCC is not itself an insurance company.)
The use of AI in insurance claims processing has steadily accelerated. CCC reported a 50% year-over-year increase in the application of advanced AI for claims handling in 2021. The company reports that more than 9 million unique claims were routed through its solution of deep learning AI – a number that has increased by more than 80% in 2021.
When an accident claim arises, the insurance company must send adjusters who answer a long list of questions: is the car completely damaged or can it be repaired? How much will it cost? What is the best way to repair the car? Where do I get spare parts? Will parties need a rental?. The same questions need to be asked every time, which makes them particularly suited to a deep learning model: understanding the damages and solutions of previous accidents and applying this learned knowledge to future ones.
CCC handles around 16 million auto accident claims every year, giving it a rich database on which to base AI models. CCC’s deep learning model draws on billions of vehicle damage photos, incident reports, and claim form line items. Deep learning algorithms and computer vision are beginning to detect patterns – a bump that looks a particular way will require some type of intervention and cost a certain number of dollars – and provide recommendations for next steps.
“We have built an AI model that determines line by line the things that will have to be done: what parts will be needed? How long will it take to be repaired? We answer a lot of questions,” says Goodson. CCC’s algorithms also sort data by car model, so the algorithm can continue to learn as new claims are filed.
If a customer files a report at the scene with images of the accident, these are compared to the database to find images of the similar model as well as learned information about the various repair processes. “We can immediately say, ‘this car is going to be fixable, here’s the shop to take it to if it’s driveable, if not, call that towing agency,'” Goodson says. “It makes the whole process much faster and much less traumatic for the consumer and much less laborious for the businesses involved.”
To make its AI-provided recommendations more understandable, CCC provides its estimates with “heatmaps” that highlight damaged spots and make them easier to visualize.
Build robust ML models
To reduce bias, CCC cleans its models without identifying information such as vehicle identification numbers, address, and city names. License plates are also hidden. “It’s a very strenuous process to make sure the data is really ready to be trained,” says Goodson, estimating that nearly 35% of their time is spent preparing the data.
Natural language processing (NLP) comes into play for documents that might not be easily digestible in digital formats.
Insurance claims processing is particularly well suited for AI applications due to a large database and the ability to apply inference-based recommendations. Similar mechanisms can apply to other industries with heavy documentation tasks and a large repository of information. Goodson cautions against relying on AI to save time without relying on solid data. “Most companies want an AI practice, but they don’t have enough data or have ethical principles in place to ensure that biases don’t creep in,” points out Goodson.
“You have to train and retrain your model if biases appear, you really can’t take shortcuts, you have to pay a lot of attention to data cleaning and preparation,” Goodson says. What does “enough” data look like? “It’s subjective to the industry and hard to answer, but it’s definitely not in the hundreds,” Goodson says.
CCC’s own AI ventures will move towards information processing at the edge. Going forward, expect consumers to simply be able to live stream a video of the damage to the insurance company’s portal and receive instructions on next steps within minutes. “We are using advancements in AI to advance not only our back-office techniques, but also to leverage that technical capability to advance our [front-end] also solutions,” says Goodson.
VentureBeat’s mission is to be a digital public square for technical decision makers to learn about transformative enterprise technology and conduct transactions. Learn more