Words like “disruption” and “revolutionize,” and expressions like “catching fire,” typically conjure up images of hip new IT startups—not an insurance company. But that’s exactly how the application of artificial intelligence (AI) in the insurance industry is being described. And we’re not talking about hype like robots replacing humans, we’re talking about the practical application of AI in all kinds of insurance from health and life to property and casualty.
So get ready! AI has the potential to impact the entire insurance industry to the same degree as driverless cars will impact transportation—it’s a game changer. Although it won’t happen overnight, and it’s still early days, leveraging the power of AI is revealing many advantages. But first, what exactly is AI?
AI isn’t just one technology
AI is defined in many ways, but the common thread is that AI is not just one specific type of technology, but rather an umbrella term that captures a variety of different technologies. However, all AI technologies have one fascinating thing in common: they aim to imitate or augment human intelligence.
For example, the goal of an AI technology might be to try to mimic a certain human behaviour or human thought process and try to behave intelligently or rationally. AI technologies try to learn, reason, and problem solve—just like humans.
Overall, a main benefit of the various AI technologies to all types of insurance is that it has the potential to enable insurers to continually improve the customer experience. And along the entire continuum of the customer experience:
- Risk and pricing: AI technologies can build more accurate predictive risk models to more effectively and efficiently assess risk and in turn, enhance pricing strategies. For example, a property insurer uses AI technologies that combine satellite imaging and machine learning to determine roofing conditions of houses, decreasing the need for human assessment and more accurately pinpointing properties that present higher risk.4
- Sales and marketing: AI technologies can find and compile such a wide range and large volume of data that it can create a full profile of a potential customer and then match it with suitable products, resulting in a more targeted approach and ideally more receptive sales targets. For example, an auto insurance broker used natural language processing to analyze over 20,000 car insurance chat conversations to build a virtual assistant (VA).5 The VA acts as a broker by interacting with consumers via Facebook Messenger to assess their policy requirements and cost preferences. Based on these preferences, the VA than reviews each insurer’s service and product reviews to make recommendations of potential insurers to the consumer.
- Customer service: AI technologies can provide customers with more ways to get help and information beyond meeting with or talking with humans. In fact, there is now even a chatbot that helps people understand and track all of their insurance policies. Based on data from more than 9,000 insurance policies, the chatbot intuitively gauges the intent of the user’s questions and answers them super quick.6 It’s even trained to explain complex insurance jargon.
- Claims management: AI has the potential to automate a range of administrative processes that traditionally have been done manually. For example, many insurers along all lines are now using AI technologies that pre-populate application forms with customer data.7
- Fraud prevention and detection: AI technologies are now able to not only find and compile masses of data like never before, but also to find patterns at a level beyond human capabilities. For example, an auto insurer’s AI technology may pick up a simple tweet of a photo of a car accident that is then used to verify accident details like the accident location and people involved.8
Next generation of fraud busters...Next generation technology
Today, an increasingly more sophisticated approach to fraud prevention and detection is critical because of the way fraud is evolving.
In the past...
Fraud was mainly small-scale like individual plan members or health care providers acting alone to commit fraud, not as part of a group.
Today, it’s organized fraud—collusion in all shapes and forms—collusion among plan members, collusion among health care providers, and collusion among health care providers and plan members. This may sound familiar as 2017 seemed to see its share of headlines about collusion-style health benefits fraud as a group of employees of a government organization colluded with an orthotics provider to pull off what is described as a multi-million-dollar insurance scheme.
The orthotics provider allegedly conspired with employees to submit benefits claims for orthotic devices—like compression stockings, sleeves, orthotics, and therapy services—that were never actually provided, totaling more than $5 million. The orthotics supplier and the employees then allegedly split the insurance payments received for the fictional orthotic devices. “Allegedly” because, although the orthotics provider pled guilty and has been charged, there is no verdict yet for the employees who were charged. In addition, over a hundred employees resigned or retired, which may have been to avoid dismissal.
Just as fraudsters have become more sophisticated, so too has fraud prevention and detection—thanks to AI.
In the past, preventing and detecting fraud involved manual processes, essentially the focus was on following a paper trail. Now the ability of certain types of AI to find patterns in data at both the individual and aggregate levels, means that amazingly, AI can tease out correlations typically hidden to the human eye. For example, the GSC fraud detection and prevention strategy applies a type of AI called machine learning or ML to do exactly that.
Machine learning means look out fraudsters!
Simply put, ML is the computerized detection of data patterns and analytic processes that is essential to cope with the broad range—and massive volume—of new types of data now available (known as “big data”). Okay, even the simplest descriptions of ML aren’t that simple.
Basically, today’s range of data—that comes in a huge volume in all forms and from all directions—is impossible for us mere mortals to compile, let alone make sense of it all. By contrast, ML technologies can not only find and compile all kinds of data—at a tremendous volume—it can also identify patterns at a much more sophisticated level than we can.
Part of this is because as humans, we often unwittingly have biases that skew what we see and in turn, what we find. In addition, humans get tired and have to eat—and although we find it hard to relate to, apparently some actually get bored analyzing data—meanwhile, ML technologies just keep chugging along.
And it just gets better and better—literally with “self-education”
And there’s more, ML technologies also “self-educate”—another fancy term; it means that ML continuously learns as new data presents itself. As a result, the learning curve for ML technologies is automatic and in real time, much faster and more sophisticated than possible for humans. Without being explicitly programmed like technologies of the past, ML technologies continually revise their analysis as they find new connections—they automatically get smarter and smarter.
For example, GSC’s ML technology continually adapts so it gets increasingly better at spotting potentially fraudulent activity. Specifically, because it can amass such a broad range of data, at such high volumes, and perform very sophisticated analysis, it is able to unearth non-obvious connections.
Sussing out the non-obvious connections is essential because collusion means that the fraud is very organized. The actions by anyone in the group are orchestrated so that everyone in the group responds methodically and identically. As a result, fraud strategies need to catch the colluders off guard by finding non-obvious connections that allow us to surprise them. For example, a non-obvious connection may lead us to contacting someone who isn’t even aware that their data is being used fraudulently.
Plus, because big data means fraudsters have to cover their tracks across all this high volume of data, GSC’s ML technology will make it increasingly difficult for fraudsters not to slip up. This is especially significant regarding catching collusion because these larger-scale fraudsters cover their tracks over a much larger scale of information.
Brent Allen, GSC’s vice president, Service Operations sums it up: “Smaller-scale fraud where individual plan members or health care providers act alone or team up to a limited degree continues to represent the highest incidence of fraud. Whereas, the collusion that is now part of the fraud scene is less frequent, but it by far represents the highest dollar amount. Fortunately, by applying ML technologies, we will be able to continually enhance our ability to catch both small-scale fraud and collusion, especially due to ML’s self-educating capabilities.”
So what about ML actually replacing humans? This is definitely a misconception as actual human brainpower will always be essential for ML technologies to be successful. For example, regarding fraud prevention and detection, humans still need to manage the ML applications and then as ML identifies potentially fraudulent activities, fraud analysts and investigators are essential to follow-up on leads. Accordingly, although ML is taking fraud prevention and detection to a whole new level of sophistication, the ideal is still human intelligence combined with artificial intelligence.
Just the beginning...
We’re sure to continue to see applications of AI cropping up in increasingly innovative ways. Like given that AI aims to mimic human behaviour—and health management is all about human behaviour—it’s no wonder developing virtual health assistants is one of the hottest trends in health right now. Given that a virtual assistant is essentially a chatbot that can intuitively converse with you to help you accomplish tasks, virtual health assistants can help with any number of health-related tasks—like effectively taking your medications or even helping you select the best playlist for your morning run.
It’s clear that what we’re seeing today in AI is really just scratching the surface of what the future will hold —and what future editions of The Inside Story will cover. Definitely more to come!
1“Seeing through walls,” Emily Finn, MIT News, October 18, 2011. Retrieved October 2017: http://news.mit.edu/2011/ll-seeing-through-walls-1018
2“New robotic 'muscle' is a thousand times stronger than a human's and capable of hurling an object 50 times heavier than itself,” Daniel Miller, The Daily Mail, December 21, 2013. Retrieved October 2017: http://www.dailymail.co.uk/sciencetech/article-2527612/New-robotic-muscle-thousand-times-stronger-humans.htm
3“Scientists predict AI will allow us to translate dolphin language by 2021,” Luke Dormehl, Digital Trends, May 10, 2017. Retrieved October 2017: https://www.digitaltrends.com/cool-tech/dolphins-natural-language-processing/
4Artifical Intelligence Impact on Insurance, Max Kraus, The Think Blog, Oct 28, 2016. Retrieved October 2017: http://www.logiq3.com/blog/artificial-intelligence-impact-on-insurance
5“The Impact of Artificial Intelligence on Selling & Distributing Insurance (Part 2),” Max Kraus, The Think Blog, June 13, 2017. Retrieved October 2017: http://www.logiq3.com/blog/artificial-intelligence-impact-selling-distribution
6“IBM helps launch insurance chatbot,” Maria Terekhova, Business Insider, June 8, 2017. Retrieved October 2017: http://www.businessinsider.com/ibm-helps-launch-insurance-chatbot-2017-6
7“Insurance sector tunes into artificial intelligence,” Oliver Ralph, Financial Times, March 29, 2017. Retrieved October 2017: https://www.ft.com/content/a0d9aa8a-1494-11e7-80f4-13e067d5072c?mhq5j=e5
8“3 Things The Insurance Industry Can Learn from Silicon Valley Insurance Accelerator,” Chris Murumets, The Think Blog, Oct 20, 2016. Retrieved October 2017: http://www.logiq3.com/blog/3-things-the-insurance-industry-can-learn-from-svia