Editor’s observe: As we speak we’re listening to from danger administration software program supplier Solera Holdings on how they remodeled their automotive claims course of utilizing machine studying from Google Cloud.
Caught on maintain together with your automobile insurance coverage claims division? If a fender-bender isn’t sufficient to ship your stress ranges via the roof, negotiating prices and insurance coverage deductibles with a claims adjuster most likely is.
At Solera Holdings, our enterprise is car harm estimation. We take care of round 60% of the claims worldwide between insurance coverage corporations, drivers, and the automotive trade. Like something at the moment, when individuals need their automobiles fastened, they need it executed as quick as doable. However not like different trendy companies comparable to rideshare or meals supply, claims departments at your insurance coverage firm doubtless aren’t fairly up to the mark. That’s why we determined to remodel Qapter, our established claims workflow platform, right into a touchless clever claims answer.
Higher secure than sorry—however nobody desires sluggish
After I joined Solera in 2020, I got here with the understanding that nobody explicit synthetic intelligence (AI) or machine studying (ML) know-how might be utilized to unravel each enterprise downside, regardless of how revolutionary or disruptive that know-how could be. In my expertise, fixing points all the time requires a number of in-house and cloud applied sciences. My imaginative and prescient was to successfully implement AI applied sciences to the best issues to realize and preserve aggressive benefits for Solera. So, I used to be delighted to find my workforce was already means forward of me and had been engaged on a option to resolve one in all their greatest issues with the assistance of AI and ML.
Based mostly on enter from insurance coverage corporations over time, the Solera product workforce knew that prospects wished an AI-based claims course of. Whereas restore estimation know-how has advanced from estimation spreadsheets to three-dimensional fashions, trendy buyer expectations are quick outpacing yesterday’s options and processes. Sadly, many insurance coverage suppliers take a “higher secure than sorry” strategy to present programs, and the top result’s a buyer expertise that’s as irritating as it’s sluggish. It was clear this was an space that was ripe for enchancment, and with our lengthy historical past of remodeling the insurance coverage and automotive trade, we wished to be those to crack the case.
The problem with any AI mission is making use of the best applied sciences to the issue at hand. It’s important to know the house and scope so we are able to use know-how successfully, or danger falling quick. A number of insurers had already tried (and failed) to make use of pc imaginative and prescient to automate the collision harm restore course of. Whereas they managed to construct working in-house options, all of those AI initiatives in the end bumped into points when it got here time to scale.
What might we do in a different way to keep away from failing as an AI mission? First, we stored our focus slender, solely methods to use AI to determine automobile harm within the collision claims workflow, not the complete restore course of. We then selected to reinforce our present backend programs with ML to leverage our substantial present database of proprietary automotive photos and elements catalogs to streamline the method of providing exact strategies, value, and time estimates for repairs.
Moreover, earlier than I arrived at Solera, the workforce had already constructed a earlier model of an automatic claims system that helped eradicate a number of much less profitable approaches. The unique model gave us a robust blueprint to work off and enabled us to reimagine Qapter’s full potential when mixed with the newest cloud and AI applied sciences. We knew the place we wished to go—all we would have liked was the best AI answer and the newest cloud applied sciences to assist us rework the preliminary harm evaluation into an AI-powered course of.
Google Cloud: An AI know-how toolbox with every thing we’d like
Our workforce was already skilled with cloud know-how after we began searching for an AI/ML answer that might combine with a full suite of superior cloud applied sciences. Whereas we host our personal information lake for contractual causes with our prospects, our accident declare workflow was already cloud-based. We knew that choosing the proper know-how vendor could be crucial to a profitable end result for the next-generation platform.
After finishing an intensive know-how bake-off, we discovered that Google Cloud’s AI/ML options had been extra refined, sturdy, and scalable than what different distributors might supply. Having best-in-class applied sciences for constructing and deploying AI functions, comparable to Google Kubernetes Engine and Cloud Run, that combine with the complete Google Cloud ecosystem performed a definitive position in our choice. Briefly, Google Cloud had every thing we would have liked to take full benefit of AI and ML options for processing touchless claims whereas additionally offering us with extra refined capabilities and tooling that hurries up improvement and deployment reasonably than worrying about sustaining infrastructure.
The core worth of Qapter is its capacity to know how the automobile consists utilizing 3D automobile fashions. We repurpose this information and put it via totally different workflows, comparable to automobile inspection or collision estimation. Utilizing Imaginative and prescient API and TensorFlow, we constructed a system that permits us to gather and acknowledge claims info, comparable to automobile make and mannequin, harm info, and elements required for repairs—all primarily based on collision photos.
Beginning with Imaginative and prescient API’s easy picture processing, we used its optical character recognition (OCR) to gather license plates and VINs. We then used TensorFlow to construct customized algorithms and machine studying fashions for picture recognition and automobile information extraction, which allows us to gather different vital info like automobile make and mannequin, harm info, and elements for repairs. As well as, Cloud GPUs (Graphics Processing Models) and TPUs (Tensor Processing Models) enabled us to speed up our information mannequin processing and improve our capacity to coach massive, advanced fashions quicker.
Now, all we’d like is an image of the broken automobile—and Qapter does the remaining. As soon as Qapter has the picture, it compares it in opposition to our huge repository of claims photos to estimate the extent of the harm, acknowledges the automobile’s make and mannequin, identifies what elements are wanted, and estimates the ultimate restore value.
From breakdown to breakthrough
We began rolling out the brand new Qapter in France and the Netherlands throughout 2020, and there’s little question that it has dramatically modified the complete claims expertise. Our prospects are thrilled with the brand new AI-based strategy. As an alternative of sending a claims adjuster to look at a automobile bodily, all a driver has to do now’s take an image of the automobile, add it, and begin the method.
It’s been a game-changer—inside months of the preliminary launch, Qapter might auto-authorize 50% of harm claims, lowering estimation prices by practically half. It has additionally offered an sudden profit throughout the complete harm claims worth chain in the course of the COVID-19 pandemic. Whereas Qapter reduces time and prices for drivers, insurers, and auto restore suppliers—in the end, it additionally cuts down on the necessity for human interplay.
Even in a world of social distancing, vital companies should nonetheless be accessible. Qapter retains the automobile restore cycle working easily, so drivers can get again on the street, restore retailers can proceed working, and insurance coverage corporations don’t must ship out staff to evaluate claims in particular person.
At Solera, we wish to proceed growing and constructing new services on prime of the brand new Google Cloud framework we’ve created. Laptop imaginative and prescient has a whole lot of functions throughout the harm estimation house, comparable to window and windshield harm, insurance coverage protection assessments, rental or lease returns, and fraud detection. Google Cloud isn’t only a spot answer for fixing a difficulty, it’s a core competency for us that may be leveraged throughout the complete firm.