There’s no doubt that machines are getting smarter, and the brightest new graduates from the “academy of machine learning” may be programs such as Google’s Cloud Vision – software capable of identifying and classifying the content of photographs.
According to Google, Cloud Vision allows users to “quickly classify images into thousands of categories (e.g., ‘sailboat,’ ‘lion’, ‘Eiffel Tower’), detect individual objects and faces within images, and find printed words contained within images.” The makers of these AI tools are touting potential applications that range from identifying offensive content on social media to developing marketing campaigns based on analyses of consumer attitudes toward certain images. AI is being implemented in many solutions every day.
How Much Is My Damage?
When it comes to auto collision claims there’s reason to be excited with machine learning. Although these computer programs will require more “schooling” before they’re able to generate a perfect vehicle repair estimate, there are several near-term applications for the auto claims sector – applications that could help reduce costs, improve efficiency and enhance customer satisfaction.
By scanning and analyzing millions of historical photos of damaged vehicles, and comparing those photos to images and of undamaged vehicles, AI software could be trained to assess many different types of damage, assess the level of damage severity, and then make initial assessments and recommendations based on the data. Eventually, this AI could even begin to make assumptions of potential hidden damage based on historical outcomes and further direct an auditor or desk reviewer of potential areas to further investigate.
By no means am I suggesting that a computer could somehow detect hidden damage or damage that doesn’t appear in a photo. What I am suggesting is that AI could soon reach the point where it could accurately assess narrow categories of losses within a matter of seconds – e.g., where damage is limited to the vehicle’s exterior. Perfect accuracy will take time, but three auto-claims applications could be “ready for prime time” in short order.
Near-Term Applications
The first application is at the First Notice of Loss. One or two photographs taken by a vehicle owner could enable the AI to quickly triage the vehicle to the right Method of Inspection.
The second, which could save insurance companies hundreds of thousands of dollars in wasted tow, storage and tear-down fees to assess vehicle damage, is using the technology to determine whether a vehicle is repairable or a total loss based on the year, make, model and extent of the damage.
Third, the technology could serve as a quality assurance mechanism and auditing function – one that assesses the accuracy of estimates written by field appraisers and body shops, ensuring that proper repair versus replace decisions are being made while flagging potential mistakes. Even with some limitations, an AI tool could help human auditors focus on what needs further examination while sorting through images that are irrelevant to the loss.
Despite such promising and exciting possibilities, it’s important that we not get carried away by some of the hype coming from the Insuretech world. Many news stories make it seem like these technologies are already perfected when, in fact, they may require much more testing and refinement. So it’s important to curb some of our enthusiasm about these deep-learning tools and remember the reality and application of such technology is still years away.
That said, we are definitely crossing the threshold into a very exciting time. The insurance industry is on the verge of adding a variety of new tools to the auto-claims tool box. At ACD, we are testing and developing some of these forward-looking technologies, and are excited to integrate them into our industry leading material damage workflow platform, AutoLink®.
