AI Aims at Fraud
The Complete Guide to AI in Insurance Fraud Detection
By Kumar Patel, Founder and CEO of Omnidya.
The insurance industry is rapidly evolving. Emerging technologies are breathing new life into daily operations; and artificial intelligence (AI) has the singular potential to pervade every facet of insurance processes with advanced analytics, powerful predictions, and more robust risk management strategies. It’s no wonder that, according to a 2018 Accenture survey, four out of five insurance executives already believe that AI will work alongside their human employees within the next two years as a co-worker, collaborator, or trusted advisor.
The high amount of fraud in insurance, however, isn’t changing. From stolen identity to false claims to exaggeration of damage, insurers lose astronomically large amounts of money due to fraud every year. In fact, the FBI reports that the total cost of insurance fraud is estimated to be over $40 billion per year, excluding medical insurance. This translates to a significant sum, added to the premium costs of every policy issued and negatively affecting both macroeconomic prospects and each individual customer. Scam expenses can cost an average US family up to $700 per year on increased premium fees.
AI has the potential to become a true watchdog of insurance, uncovering fraud and preventing it in the first place. The disruptive technology has the power to monitor an unlimited number of claims at the same time to find suspicious activities, unveil hidden relationships, and quickly identify behavioural patterns. With over half of life insurers witnessing a 30% rise in fraud, is AI the solution? If so, then how do we utilize its full potential?
Can’t Fool A Bot?
Leveraging AI to detect suspicious activity is quickly becoming standard. 75% of the industry reported using an automated fraud detection technology in 2016, which is understandable as AI can be applied in several ways. Thoroughly analysing data from diverse sources can result in fraudulent causes being swiftly spotted – and with greater precision. AI also helps to uncover hidden correlations and discover new fraud schemes by mining information from both current and past claims.
Crunching mass amounts of past data and new data in real-time from various sources while identifying key connections is beyond human capability. AI systems can analyze social media accounts, communications, bank transfers, and websites such as eBay to look for listings that match stolen items. So if you were tagged in a Facebook photo of you dancing on a stolen table right after you reported a broken limb, watch out. Insurers can prefill claims in real-time using data from satellites, networked drones, weather services, policyholders, and more.
With AI’s capacity to analyze unlimited amounts of information, the claim settlement process is made simple. AI-powered bots can review claims, verify policy details, and run data through a fraud detection algorithm before sending instructions to the bank to pay for the claim settlement – or if anything is detected, forwarding it to a human employee for further investigation. One chatbot currently on the market can run claims through 18 different anti-fraud algorithms. With such scrutiny, any meddling with personal data or other suspicious activity can be halted immediately, since any red flags are promptly reported.
AI is also veering companies away from manual verification methods. Forget about home visits — now, it’s all about picture and video analysis, facial recognition, and drones. The latter are rapidly gaining momentum with a few companies already sending them to generally inspect properties and confirm the extent of damage.
But it’s crucial to realize that AI can not only reveal current fraud, but prevent fraud altogether. It can identify dubious records by analyzing individuals’ data before any contracts are signed. Implementing a robust analytics scheme can also deter any devious actions. Interconnectedness of devices generates real-time insights and promotes absolute transparency, making it difficult for anyone to manipulate information. Customer behaviour is also easier to monitor than ever before. A simple car sensor can now tell us that a driver was driving irresponsibly above the speed limit, which directly affects claims assessment. Or in another scenario, AI can analyze facial patterns of an individual to uncover signs of life-threatening habits or illnesses – which is why hiding these realities won’t be easy.
It’s Not All Roses
Clearly, AI empowers insurers to detect fraud more effectively, however, there are still many challenges. The data that’s analyzed must be completely reliable. If an individual is falsely accused of fraud, the company’s reputation can be significantly damaged and the relationship with the customer completely obliterated. Insurers should modernize and optimize their data collection and management strategies to avoid this. Too often, companies are still running on legacy software with ancient infrastructures of ambiguous databases.
Privacy is also a massive concern – and it’s no wonder why. Technology brings countless benefits but consumers often don’t like seeing a drone flying over their backyard, no matter how noble the intention. Social media scanning also inspires an ethical dilemma. Companies believe that they can predict fraud based on client behaviour – for example, an individual Googling debt solutions or mentioning an unfortunate personal situation to a friend in an email. The shift towards a person-centric fraud detection method lends significance to every single one of our actions. But this method raises important questions about whether companies should even be allowed to access these types of personal data.
There’s an interesting example of a British car insurance company that developed an AI-powered product to provide personalized car insurance. The program scanned various data channels, including the driver’s Facebook posts, to build their customer profile – complete with personality traits and susceptibility to risk. Based on this information, it would offer the right package. But this service was halted just hours before its official launch due to privacy concerns. There are still other companies balancing on the line, however, clearly showcasing a need for industry regulations and standards to define the gap between all-pervasive control and satisfactory privacy.
Another challenge is balancing human tasks with implemented AI. Insurers must realize that the technology isn’t here to completely take over their work, and that AI systems do require detailed human oversight. While AI can automate many tasks, the technology is not omnipotent. It can be flawed, especially in early implementation, and result in situations where it might not be able to differentiate between several typos or attempted fraud. Humans also play an irreplaceable role in teaching the system and conducting manual supervision to ensure that algorithms are running smoothly. An ideal scenario would be to make AI focus on detecting simpler frauds, allowing human investigators to prioritize more complex situations.
A Solution To Any Problem?
We could continue to outline the pros and cons, but there’s one undeniable truth: implementing AI to streamline fraud detection allows for a diverse scope of application. There are a variety of potential setups, including anomaly detection. This consists of teaching the AI what a normal claim looks like, thus making it treat any deviation as an anomaly – while also leveraging predictive analytics, using a pool of past resolved claims to develop a unique algorithm that distinguishes between regular and fraudulent claims.
Whether developing an in-house solution or outsourcing the setup to innovative startups, this means that AI has enormous potential to resolve the industry’s long-lasting and situation-specific challenges.
In healthcare insurance, for example, the problem of billing for services that were never actually received can be mitigated. AI can comb through a vast amount of data and report the exact procedures that were carried out, along with keeping a database of healthcare professionals that have been tied to suspicious activity. Likewise, insurers struggle to prevent “upcoding” – trying to “sell” a procedure as more complex by using a different categorization code. Here, AI can use anomaly detection to assess what the typical treatment is for a specific condition and detect any deviations.
The status quo of insurance is being torn down and AI is at the forefront. With fraud detection being the priority, the technology is seen as a powerful weapon in response to scam strategies. While concerns remain, AI has the advantage of being able to turn into a holistic solution that can be applied to any insurance sector.
Kumar has an extensive background in business and growth strategy, business analytics, change management, and more. He’s currently the Founder and CEO of Omnidya, an InsurTech startup that leverages artificial intelligence to generate bot versions of consumers.