It’s the holy grail of the insurance claims business: The near-instantaneous cycle time of truly touchless claims. Insurers are eager to realize the benefits of reduced expenses, higher client satisfaction, and, possibly, better outcomes.
But an even more critical reason to pursue automation is the looming threat of mass retirement of insurance professionals, with more than 400,000 expected to exit the industry in the next few years.
No wonder so many insurers and insurtech companies are rushing to create solutions for automation. With fast-emerging technologies like artificial intelligence (AI) and robotic process automation (RPA), solutions are becoming more real by the day. According to the World Insurance Report 2017, 69 percent of surveyed insurers consider AI important, while 63 percent feel the same about RPA. Further, 80 percent of insurers in the same survey say they are either already investing in these technologies or plan to in the next three years. Along with predictive analytics, cloud computing, chatbots, and more, such technologies are transforming the claims management landscape.
It Begins with Data
Besides retiring claims professionals, a key driver of automation is the increasing availability of varied and voluminous data, which is beyond the realm of human ability to analyze in a timely manner.
Consider a simple workers compensation claim: In addition to claim facts about the loss event, the individual, and the injury, additional sources of data can be leveraged to determine if the claim is suspicious. These could include the individual’s claims history from more than one billion claims that might point to abnormally high frequency of prior claims; court records that might include past arrests or convictions; household circumstances that might signal stress; and financial information that might capture a bankruptcy.
Data could also be derived from social media posts, online behaviors, and internet of things (IoT) devices such as wearables. A claims professional will often struggle to manually analyze each of these data sources. Instead, systems can be built to automatically source this diverse data, analyze it, and deliver suspicion alerts for efficient action. For instance, spidering and text analysis technologies might determine that the individual, though claiming to be bed-ridden due to injury, is scuba diving in Hawaii, posting selfies, and tweeting them.
Technologies can also help automate the collection of data. With the proliferation of smartphones and the increasing use of drones, the process of gathering loss details is rapidly changing. Policyholders and claimants can now upload their own loss photos to enable claims processing, eliminating the time and expense of sending a claims professional or appraiser to view and document the damage. Drone and aerial imagery is proving particularly valuable for validating and assessing damages following catastrophic events when claims professionals are unable to reach loss sites due to flooding, road closures, or other hazards.
Understanding AI
What most people do not realize is that artificial intelligence has been used in our systems and processes long before the recent buzz. Coined as a term in the 1950s, an artificially intelligent system is essentially one that behaves as an intelligent human would in a certain context. In insurance claims, this simply means a system capable of making decisions that an experienced claims professional typically would—with equivalent or better accuracy and performance.
Consider two key decisions a claims professional routinely makes: Is this a covered loss? And is this a meritorious or suspicious claim? These days, most claims management systems perform policy matching and coverage lookup automatically. Typically, the claims professional keys in the insured’s details and circumstances of the loss and interprets coverage on the resulting policy match (or declares a nonmatch). A simple RPA can be set up to enter this information automatically and determine if the loss is covered. Such a system, despite being simple and straightforward, would be considered AI if it automatically makes decisions as well as a claims professional could.
Similarly, consider a simple suspicion scoring model composed of a set of rules (red flags, such as a loss happening within 30 days of a new policy or multiple unrelated injured claimants in the same vehicle), with each rule contributing certain points to the suspicion score. The claim could then be deemed suspicious if the points exceed a certain predetermined threshold. Such a model, combined with an RPA process to route suspicious claims automatically to an SIU queue, would also be considered an AI system, albeit a simple one.
Why All the AI Attention?
If AI systems can be so simple, then why all the recent buzz? The answer lies in the massive improvements in what AI can do, thanks to greater computing capabilities, the availability of large volumes of data, and the pay-as-you-go model of cloud computing.
Today’s advanced AI systems use more sophisticated algorithms (for example, decision trees, random forests, gradient-boosted machines, text mining, graph mining); work on diverse data sources (for example, claims notes, social media, photos, social networks); and are required to exceed human performance.
Another key distinction of AI today is the ability of the algorithms to learn from data instead of having to be programmed. When presented with historical claims containing fraudulent and meritorious claims, these techniques allow the computer (machine) to automatically learn patterns correlated with suspicion—hence, the term “machine learning.” This also allows the systems to improve continually based on actual experience.
Deep Learning and the New AI
The ability to analyze rich visual data in immensely powerful ways is one of the unique characteristics of the human brain. Expert claims professionals leverage these capabilities routinely to determine if a damaged car is a total loss or repairable, and to intuitively estimate repair costs.
Drawing loose inspiration from the neuronal structures in the human brain, a class of AI models has evolved called “artificial neural networks.” Convolutional neural networks (CNNs) are one specific type, composed of many layers of computing structures that process input data like image pixels—with each layer detecting a certain class of characteristics in the input image, such as shapes, colors, textures, and so on. When trained with labeled images, CNNs have demonstrated powerful ability in accurately classifying (cars versus trucks), recognizing (chairs, tables, appliances), segmenting (fender, door, windshield), and characterizing (damaged versus not damaged) images.
With its stellar performance—and advances in computing frameworks and open-source packages making them practical—this kind of deep learning has directly contributed to the resurgence of AI. With increasing availability of photos and videos of losses and loss locations, these technologies can be leveraged not only to discern the nature and extent of damage, but also to enable claims professionals’ decisions, such as loss estimation and totaled versus repairable.
These new AI capabilities further increase the proportion of claims that can be processed with minimal or no human touches.
The Holy Grail of Touchless Claims
While the AI systems discussed here can significantly increase the efficiency and productivity of claims professionals, the leap to fully automated claims handling won’t be a flip of a switch. Forging a path to truly touchless claims will likely require retooling processes at every stage of a claim’s lifecycle.
The first need will be to engineer systems that can interact directly with the end user to collect details of the loss. This is where chatbot/voicebot technologies come into play. With sophisticated AI techniques for natural language processing and speech-to-text and text-to-speech conversion, many insurers are using the technologies to automate customer service requests. Forbes has reported that chatbot systems are gaining mainstream acceptance due to 24/7 access convenience and “no hold music” service, among other customer benefits.
However, the systems will need to evolve considerably to handle the complexities inherent in collecting loss details. They’ll also need to better understand human emotions and respond with sensitivity, especially in interactions following a serious accident or loss.
To deliver on truly touchless claims, this array of technologies and algorithms will likely need to work harmoniously to handle the intricacies of the claims handling process. From collecting and leveraging diverse data, often in real time (chatbots, smartphones) and at high velocities (IoT/telematics); to sophisticated AI engines for image, text, speech, and video mining; to predictive models that mimic claims professional decisions—the automated claims engine is indeed complex.
Developing all this technology in-house is a daunting task for any insurer. There is a growing need for plug-and-play AI systems to appreciably shorten the time to implementation and results, and the market is responding with significant InsurTech activity.
It is critical to understand that the goal is not just touchless claims—it is really about shifting the balance. Today, about 80 percent of claims are high-touch, with roughly 20 percent being low-touch. Using the automation technologies discussed previously, we anticipate 30 to 40 percent of claims becoming touchless, with another 30 to 40 percent becoming low-touch, leaving only a fraction of claims as conventional high-touch.
This shift toward right-touch claims is crucial not only for carriers facing workforce shortages, but also for profitability and viability. Insurers that best implement the new technologies have the opportunity to propel themselves to the forefront of the market.