Leveraging AI and Analytics for Insurance Fraud Detection

New digital players face real barriers to entry such as stringent regulations, product complexity, and large balance sheets. Due to these, insurance companies have been somewhat slower than other financial services organizations to embark on their digital transformation journey. But now, as the resilience of the traditional insurance business model is being challenged in the digital world, AI, machine learning, analytics, digital distribution, digital marketing, and sensors fitted in vehicles, to name but a few, are starting to drive conversations in boardrooms.

At the most fundamental level, insurance is a risk transfer mechanism for people and organizations, the policyholders. They pay premiums, and in return, the insurer protects them against loss—of a car, a house, even a life—and agrees to pay the policyholder or the beneficiary a set benefit in the event of the agreed-upon loss.

Premiums are not the only way insurance companies generate revenue; they reinvest the premiums into a diversified portfolio of interest-bearing financial assets that will be used to pay off any future claims brought by policyholders. Obviously, in order to be profitable, the premiums collected plus the revenue from investments must exceed the claims paid plus any operational and administrative costs.

Technology-led innovation has enabled insurers to embrace change and transform their business model

While insurers endeavor to minimize these costs—process automation can reduce claim costs by 30% (source: McKinsey)—their greatest challenge is to manage the many risks they face as part of their daily operations, including asset risk (their investments), credit risk (obligations owed by customers and debtors), and liability risk (potential losses from catastrophes or inadequate pricing or reserving). To reduce their liability risk, insurers take advantage of risk pooling across a large number of policyholders, exclude specific types of coverage from a policy, and avoid policies that have a medium-to-high chance of experiencing a loss that will result in a claim. Most insurance companies also transfer some risks to global reinsurers that will indemnify the primary insurer against losses, for a premium. But technology-led innovation has enabled insurers to embrace change and transform their business model. They now place more value on preventing losses than transferring risks.

Insurers use complex actuarial models to analyze risks, including fraud, and price policies accordingly. But these models have one major disadvantage: they use sampling methods that rely on historical data and past behaviors. As criminal practices change rapidly and perpetrators become increasingly more sophisticated, these models’ analyses become almost irrelevant. AI and analytics have greatly strengthened these models in two ways: (1) by facilitating data integration from different sources, including internal systems and external sources like social media and (2) by combining predictive modeling, other disciplines like behavioral science, and technologies like image recognition and voice analytics to ensure better outcomes for both the client and the insurer.

Fraud detection has also significantly benefited from these technological advances. There are two categories of fraud. The first, hard fraud, is when someone deliberately fakes or exaggerates an incident leading to a claim (e.g., a staged car accident or severe pains). The second, soft fraud, is when people lie to an insurance company (e.g., concerning the value of stolen items or health conditions). According to the FBI, the total cost of non-health insurance fraud exceeds $40 billion per year, which translates to increased premiums between $400 and $700 per year for the average U.S. family. Hence insurers endeavor to reduce fraud and in turn reduce premiums to remain competitive.

Data is at the heart of the matter

AI-based fraud detection systems need quality data to ensure effective detection both before a policy claim is approved and, ideally, before a policy is underwritten to keep fraudulent activity completely out of the business cycle. Unfortunately, due to legislation that makes information sharing difficult, and also because insurers are still operating with some silos, data is not shared and reused between departments. Also, many legacy systems lack crucial details. The problem is further exacerbated when insurance companies that have grown through mergers and acquisitions have data repositories and systems that have not yet been integrated.

We are watching you…

In addition to internal data, new fraud detection solutions also take advantage of social networking analysis and social customer relationship management to help identify associations between individuals. For example, while a claimant has no previous claims or suspected fraudulent activities, one of his relatives has lodged multiple claims or has been a witness in a number of claims. This situation would be flagged for investigation. These solutions also have "listening" tools to extract and analyze public data from social networking platforms to complement their CRM systems.

While AI and analytics engines will be used more and more, the successful fraud detection solution of tomorrow will also feature a very well-educated staff.

Insurance companies must balance their risk management and fraud detection capability with the operational reality:

• What are the costs of advancing AI and data analytics within the company?
• Do fraud losses represent a signi?cant burden for current or future operations? 
• Is fraud detection adversely impacting the customer experience? Is it improving honest customer claim processes?