What is graph theory and how can it help detect insurance fraud?
The P&C insurance industry is prime for examining data in a new way to analyze and uncover potential fraud.
Insurance fraud is a constantly moving target. It’s hard to keep up with the perpetrators’ latest ways. Given the thin margins in P&C insurance, a first line of defense is detecting fraudulent claims.
There are a number of new and effective ways for the industry to fight fraud. Because the P&C industry has so many product lines and data is often stored in silos, P&C in particular is a prime candidate for examining data in a new way, to analyze and uncover potential fraud that could have otherwise been missed using a conventional approach.
Graph theory: A new method
Graph theory is a cutting-edge analytics methodology that compresses analytics efforts to pinpoint hidden networks, leading to better outcomes — such as finding that needle in a haystack, or connecting together two seemingly unrelated data points to uncover the anomaly that warns, “fraudulent claim.”
Graph theory utilizes NoSQL databases, which allows for dynamic storing of big data (including unstructured data) that’s easy to scale, all housed in a distributed computing environment. It’s a powerful new method to visually see relationships between entities and untangle networks with multiple variables.
This methodology can help P&C insurers look at existing data in new ways, leading to improved transparency, faster detection of anomalies, better analysis and more confident follow-up actions.
Where typical analytics have hitherto relied on rules-based algorithms, pre-determined one-way connections, and tabular, cumbersome output to understand fraud, graph theory uses visual formatting to immediately triangulate suspicious activity and relationships that may not have been previously hypothesized. Graph theory can enhance and complement current fraud detection methods to help shed light in dark corners.
What is graph theory?
Graph theory draws on NoSQL databases that allow for flexible data models that scale efficiently and are cost effective for big data. It is an application of discrete mathematics to understand the interconnectivity among “nodes” through “edges.” Nodes can be defined in any way you want — e.g., policyholder, claimant, region — and the edges can link nodes with any direction — e.g., transaction, spatial relationship, qualitative factors. Put more simply, it takes the data and builds a “picture” out of the specifics you need to pay attention to. It allows for the immediate creation of highly flexible graphs to better understand hypothesized linkages and non-obvious connections in your data.
Typical analytics in insurance fraud rely on rules-based algorithms, predetermined one-way connections, and tabular and cumbersome output to understand fraud. Graph theory uses visual formatting to immediately triangulate suspicious activity and relationships that may not have been hypothesized.
One particular use case where graph theory could be put to beneficial use is opioid abuse in workers’ compensation claims. The abuse of opioids among injured workers can lead to many years of increased medical and indemnity costs for the insurer. Graph theory can untangle the data to understand hot spots and predictors related to doctor shopping and opioid abuse, which can then be used by claim departments to mitigate risk.
Across the insurance industry, data is often stored in ways that prevent analysts from making important connections, given that different systems relating to claims, policy and underwriting often aren’t able to interface. A broader approach to data analytics using more modern tools can create a variety of advantages for the companies that use such strategies. NoSQL databases and graph theory applications offer just such a method for broadening horizons and improving analysis.
P&C data exists in a variety of often-unlinked systems, stored in many structured and unstructured forms. This information holds significant value for insurers, but it can only be accessed through tools designed to bring all of the data together and conduct effective analysis. P&C businesses have a great place to start with claim analytics, but they need to eventually move past this specific realm to combine NoSQL databases and graph theory with the many other data sets they hold and derive the most possible value from it. Finding more ways to use these exciting platforms will provide the most abundant long-term benefits.
Key P&C insurer issue
Finding new and effective ways to fight fraud is a priority across the insurance industry. Property and Casualty, with its many product lines and data often stored in separate silos, offers plenty of room for adopting and implementing these cutting-edge methods. Using NoSQL databases and graph theory applications can bolster existing efforts to use various forms of analytics to detect outliers.
This ability to visually depict otherwise unrelated connections that spell fraud is an additional tool in the fraud detector’s kit bag. And it offers a less labor-intensive and time-consuming approach. Beyond this specific benefit, P&C businesses can put graph theory at the service of improving related areas such as claims underwriting, risk management, and customer engagement and sentiment. With very thin margins compared to other sectors of the insurance industry and a relatively limited number of new customers, the P&C insurance industry should welcome graph theory as a way to stay competitive.
Fusing data for more robust analysis
While only just starting to be recognized in the P&C market, these tools have already gained significant holds in a variety of other industries. In the health insurance field, for example, the payer community is moving toward the use of NoSQL databases and graph theory applications to identify potential issues with narcotics over-prescription and overuse. In this case, graph theory offers a powerful tool for finding discrepancies and identifying areas that potentially need more attention. There are very similar considerations to be made in the P&C market, as mentioned above, with issues related to narcotics prescriptions tied to worker’s compensation claims.
A more holistic and comprehensive view of various streams of information allows for faster and more effective discovery of potentially relevant linkages. This is one of the ways in which graph theory applications truly shine. Building off of existing databases and analytical tools, they allow companies to dig deeper and discover connections that may have been missed with more basic analytical models. Beyond fraud, it’s also possible to detect potential issues related to business processes and risk management. Although fraud is a major concern for P&C businesses, it’s far from the only one they face.
It’s important to note that these newer platforms for analysis and fraud detection don’t have to replace existing ones, nor do they have to displace more basic yet reliable and still effective methods of data analysis for P&C companies. Instead, they add another layer of investigation and review, increasing a company’s chances of identifying the connections between a group of possibly fraudulent claims or linking proven cases of fraud with newer, potential instances of the same.
Graph theory offers analysis in a compressed time frame that could take days to complete with traditional methods. The combination of fast processing and intuitive visual displays removes the need to put in all the time and effort of drilling down into data to find root causes for fraud and similar issues. Graph theory’s efficiency points users in the right direction from the beginning, improving resource allocation and demand planning. This is an all-around benefit, but its value is especially clear when looking at the sheer number of claims large P&C insurers face in their most popular product lines.
Mike Kim is a Director at AArete, a global consultancy specializing in data-informed performance improvement, and heads its Center of Data Excellence. He can be reached at mkim@aarete.com. Vasu Bhat is a Managing Director in AArete’s Digital and Data Services practice. He can be reached at vbhat@aarete.com.
These opinions are the author’s own.
See also: The changing face of fraud