Winning the war on insurance claims fraud

Insurance claims fraud isn’t new — it’s been around since the inception of our industry in the 17^th century. Yet it is an industry challenge that continues to worsen at an alarming rate as fraudsters become more sophisticated and organized.

However, recent advancements in analytics technologies gives insurers new hope in the fight, since optimizing fraud management is an ideal use-case for the application of advanced analytics and artificial intelligence (AI). As more insurers consider fraud detection solutions, they face the age-old decision of build or buy.  The initial tendency might be to build, but today’s companies are often reconsidering, opting for a buy decision for several reasons.

Time is of the essence

Perhaps one of the most obvious arguments for a buy approach is the time involved to internally develop and deploy a viable fraud detection solution. At a minimum, assuming the data science and IT skillsets are already in place, and the project takes priority in the IT queue, it typically requires at least a year to build usable models. During that time, losses from fraud add up and new fraud schemes emerge, making the ROI and quicker implementation of a buy decision attractive.

For most insurers, building a fraud detection solution is a new experience with a steep learning curve and a higher risk of failure. Some insurers may find it faster, more efficient, cost effective and beneficial to leverage the experience of Software as a Service (SaaS) providers able to provide a production environment in under six months, requiring minimal insurer IT resources.

Related:  Taking a digital approach to claims processing

Not as easy as it might seem

A good fraud detection solution needs to go beyond simple rules-based fraud detection and take a machine learning and AI-based approach. While rules-based detection might seem like a good starting point, it fails to take into account the changing landscape, where fraud schemes continue to evolve. The system must continuously and automatically adapt using current data – and a lot of it.

Insurance data is not just voluminous, but highly complex and relational. A fraud detection solution needs to understand the context of data, the relationship between data elements, and data in its various formats such as claim forms, documents, photos, audio, claim notes, and additional structured and unstructured data elements.

The entities in the data also need to be resolved to be effectively used for fraud detection with any level of accuracy. Insureds, claimants, service providers and other entities may appear multiple times in data in various contexts and permutations.

The ability to identify and resolve what looks like multiple entities into the singular entity the data actually represents is critical to the success of claim and network-based fraud analytics. This is complex work and requires a level of expertise that not all insurance companies have in-house, while fraud technology providers will have years of experience and the algorithms already built into solutions.

Lastly, a good dataset for fraud detection needs to go well beyond an insurer’s own claims data and tap into other industry data sources such as third-party public records, police reports, criminal records, financial stress indicators, social media and more. Fraud detection technology providers may have strategic partnerships with multiple data providers and will be able to leverage this external data in a way that provides meaningful lift in how models perform.

Related:  Secrets to combating insurance fraud with data analytics

One fraud model doesn’t fit all

Types of fraud vary widely from small misrepresentations in a claim to sophisticated networks staging accidents. The system needs to detect anomalies in insurance data and determine whether that anomaly is suspicious. If suspicion is raised, the system should alert a fraud handler to review the claim for further inspection and to determine if this is a new type of fraud.

If warranted, the new fraud type should be added to the system’s fraud library, with the appropriate models and algorithms refined to detect it in the future. Models and algorithms are not static. They need to be constantly reviewed and periodically refined as fraudsters find new ways to beat the system.

The solution should include a feedback mechanism to provide for continual improvement. This work is ongoing, labor intensive and time-consuming; requiring continuous management by the insurance company choosing to build internally.

SaaS fraud detection solutions may provide the benefit of the vendor’s continuous capability and model performance enhancements, as well as management and maintenance of the infrastructure.

Related:  How insurers can identify fraudulent claims before they are paid

Don’t forget the interface

Building effective fraud models is the beginning. The system needs to provide functionality that gives the end user — the fraud handler, SIU investigator and others — a way to easily analyze the suspicious claim to understand why a red flag was raised. This comprehensive user interface needs to provide all the information in an easily digestible format to better understand the suspicion, the major fraud indicators, and to determine appropriate next steps for investigation.

Given that fraud schemes can be complex, the interface needs to both visually and textually provide the relationships, similarities, frequencies, geographic locations and anomalies when a claim is flagged for further fraud investigation. If an insurer is withholding or delaying payment due to potential fraud, or taking legal steps against a policyholder, they want to be sure they have all of the evidence before taking action.

Insurance company development and IT teams tend to struggle with this part of the fraud detection solution. They frequently provide analytic results in the form of complex reason codes that business users have trouble interpreting, frequently delivered via basic Excel spreadsheets.

On the flip side, SaaS fraud solution providers tend to invest in developing a user-friendly and highly-intuitive user interface, because this interaction point between their solution and the customer ensures effective use and adoption of the solution.

Accuracy is key

For a fraud detection solution to be successful, the users must trust it. If the models are not accurate, too many false positives will result, wasting time, frustrating users and delaying payments, or worse, creating the risk of suspicious claims being ignored rather than investigated.

Insurance fraud is every insurer’s issue

While insurance companies compete on a daily basis, they have a common enemy in insurance fraud, with industry reports estimating losses from fraud at 10% of premium for property and casualty insurers.

When an insurance company builds its own solution, the only benefits arise from the internal data applied to the continuous improvement of the system. A vendor-provided SaaS solution benefits from the cumulative experience of multiple insurers, an army of data scientists focused on improving fraud detection methods and technology, time-tested solutions, and the wealth of historical and new data used to refine existing models, build new ones, and identify new fraud schemes.

Whether an insurer decides to build its own or utilize a vendor-provided fraud detection solution, time is essential as technology and experience are currently giving the advantage to the fraud perpetrators.

Dan Donovan (ddonovan@shift-technology.com) is the global head of product marketing for Shift Technology. He is an expert in the field of insurance fraud and claims analytics with more than two decades of experience leading fraud investigation and claims operations for Liberty Mutual, and working in the insurance technology sector.