How Data Analytics can Help Curb Insurance Fraud
There is no denying the fact that India is one of the fastest-growing markets for insurance. However, the industry is bleeding losses due to increasing insurance fraud. According to the Indian forensic center, the country loses around 6.25 billion INR to frauds annually. This is almost 8.5 percent of the total revenues generated by the industry. The situation is no different globally, where insurance scams are pegged at 80 billion USD annually. These statistics poise towards an often ignored but brutal truth that traditional methods for fraud detection aren’t working anymore. The fraudsters are getting smarter and are using technology in the best possible way to evade exposure. This presses the need for organizations to think ahead and build tech environments capable of handling such elements in a much more efficient manner. A good way here is to analyze the massive data that lies with them to derive insights for successful fraud detection.
Traditional Methods and New Frauds
Frauds happen in all shapes and sizes. However, in general, they fall into two broad categories. First is criminal fraud, where habitual perpetrators intentionally try to milk the system to their benefit by using the best means possible and second is the cultural fraud where genuine claimants turn opportunists and exaggerate their claims. In both these categories, when traditional methods like rule-based systems or prediction scorecards are applied to detect anomalies, it simply doesn’t work. The reason is being that these methods utilize manual intervention or intuitions which are prone to errors. Even in cases where specific detection tools are used, the success rate is not high as such tools cover a particular agenda only. For example, many such tools work around claim management only, which leaves other areas susceptible to deception. Recently an interesting case came up in California, where a family was arrested for organized automobile insurance fraud in the tune of USD 180K. They were running the racket for a long time and never faced detection. This showcases how tools/strategies used by insurance companies today are not efficient enough.
Using Data Analytics
One of the biggest data generators today is the insurance industry. Big data analytics, when combined with advanced analytics, can help derive relationships that could detect frauds. However, many insurers find such things confusing and turn skeptical while applying it to tackle fraud. According to a recent article in Mint, an esteemed financial daily, though insurance companies are sitting on piles of data they are unable to derive any benefit out of it as not all of their data is digitized. In a report on “Evolving considerations for the Indian Insurance industry” by CII( Confederation of Indian Industries), the level of investment in digitalization and its ability to realize financial returns from that investment is severely lagging when compared to the global average. This brings another point to light that insurers are unaware of the benefits their own data could bring to the business.
It is high time that insurers should start leveraging data and applying them in detecting insurance frauds. This can be done by matching the sophistication of organized frauds through advanced predictive data analytics that have proven their abilities in predicting probable events. Here are a few data models that can help.
Social Network Analysis (SNA) – Social Network Analysis refers to the application of data mining strategies that reveal the structure and content of information by representing it as a set of interconnected or linked, objects or entities. Through SNA, insurance companies can look through large volumes of data available on social platforms and establish relationships via links and nodes. These relationships, in turn, will be helpful while validating claims. To understand this better one can compare SNA to a hybrid approach including a host of analytical methods. The methods include mining data along the lines of organizational rules, statistical techniques, network linkage analysis, and pattern analysis to uncover relationships. For instance, when one looks for evidence of fraud in link analysis, the focus is on finding clusters and how the clusters are linked to each other. Public data like judicial records, criminal records, change of name and address, bankruptcy cases, etc. all form part of social data sources that could be integrated while creating SNA models. Now, let’s take a case involving an auto accident that comes up in front of an insurance company. Here SNA can look through the above-explained social data to establish the address of the claimant or whether that automobile/person was involved with an accident in the past or not to assess whether the accident is a case of fraud or not.
Predictive Modelling – This is one great addition to the fraud prediction family and refers to creating self-learning mathematical/statistical data models by utilizing internal and external data. To understand the working, consider a scenario of a house that has caught fire. The claimant, while narrating the story, indicates that he took all the valuable items before the incident. This indicates that the house was burnt on purpose but could be missed while analyzing the report filed by the claimant. Predictive analytics involves the use of text analysis (analyzing both structured and unstructured) and sentiment analysis (analyzing expressions) to understand big data correctly and predict outcomes for fraud detection. In the above case, predictive analytics sift through the multiple paged report filed by the claimant to find clues for fraud detection and compare it to the models already built to predict fraudulent behavior.
Social CRM - Social CRM is neither a tool nor technology, but an integrated process. It recognizes the power of data from social media platforms along with existing CRMs of the organizations to enable a transparent and beneficial fraud detection system. Here, a listening tool is employed by organizations to collect data from the social chatter ecosystem. This data acts as a reference to the data which already exists in the insurance company’s existing CRM. Both these data are fed into a case management system that shares feedback by comparing the deductions to the business rules. This response is then shared with the claim investigators to ascertain the truth independently.
Overall, insurers face a choice of either absorbing these new fraud detection capabilities today which would enable them in increasing profit margins or work with the old techniques in hopes that insurance frauds would not get any complex in the future.