Get fresh updates from Hortonworks by email

Once a month, receive latest insights, trends, analytics information and knowledge of Big Data.


Sign up for the Developers Newsletter

Once a month, receive latest insights, trends, analytics information and knowledge of Big Data.


Get Started


Ready to Get Started?

Download sandbox

How can we help you?

* I understand I can unsubscribe at any time. I also acknowledge the additional information found in Hortonworks Privacy Policy.
closeClose button
prev slide
Big Data and Retail: 3 Ways to Improve Your Black Friday Sales
November 28, 2017
Uncovering the Future of Data Infrastructure and Intelligence
Next slide

How Big Data Can Combat Insurance Fraud

Research suggests insurance fraud costs as much as $80 billion every year, making it one of the biggest crimes in the U.S. Insurance companies looking to fight back must use the power of big data—and use it properly—to tackle the growing menace of insurance fraud.

Insurers often take a reactive approach to fraud and fail to use the data they collect to stay ahead. Instead, they need to embrace machine learning and artificial intelligence. With its proactive approach to data mining, machine learning can help insurance companies identify fraud patterns before fraud occurs, and create crisp prediction and detection patterns that can help reduce claims leakage associated with fraud and abuse.

Across the spectrum of automotive, disability worker compensation, and homeowner insurance claims, this pioneering approach to data analysis and data mining allows insurers to assess many more claims than would be possible with traditional rules-based and statistical models. What’s more, for firms willing to make this leap, the benefits are huge as it significantly reduces the amount of time spent on evaluating false positives, or rules-based statistical model data that incorrectly identifies a claim as fraud.

Understanding the Manual Approach to Fraud

Insurance fraud comes in two key forms: hard and soft. Hard fraud occurs when an individual or a group, commonly referred to as a fraud ring, deliberately plan and/or collaborate to invent a claims event. In the case of automotive fraud, this group might include individuals who stage accidents, and professionals, including lawyers and doctors, who support the false accusations.

Soft fraud or, opportunistic fraud, might involve an auto repair shop upping its estimate to cover deductions, or a member of the public adding to or exaggerating an otherwise legitimate claim. While the costs of individual soft fraud cases might be smaller than organized deceptions, the results are similar—significant outlays and losses, with estimates suggesting that fraud accounts for 5 to 10 percent of claims costs for U.S. and Canadian insurers.

It’s Easy to Mishandle Data

Insurers also struggle to make the most of the information they collect. Firms collect huge volumes of data during the application, servicing, and claims process. However, this information—collected in various structured and unstructured formats—is seldom evaluated in fraud investigation operations.

All too often, fraud detection by special investigation units (SIUs) relies on business rules and human intervention. While the professionals working in these units are highly skilled, they could be missing fraud patterns among the huge amounts of data they sift and sort manually. This results in claims adjusters not always having the most accurate data and facts to determine if the stories match the evidence of the claim.

When new predictive variables or fraud patterns are discovered, it can take months for skilled teams to change systems and incorporate insight into a form of structured data capture. By embracing the power of analytics, insurance firms can get a tighter grip on their data and start to identify fraud patterns automatically and more quickly match ‘facts to evidence’. This also allows them to pay legitimate claims faster.

The Business Case for Machine Learning

It’s crucial to recognize that fraud identification is just one element of a successful approach. While it might be attractive in theory for a business to snare those who perpetrate false claims, it can be a costly exercise to prevent deception.

Insurers can avoid this cost in the first place by ensuring that their business avoids covering fraudulent actors. Given the limited scope of traditional, manual fraud prevention techniques, this might seem dubious. However, by marrying traditional techniques with big data and machine learning, insurers can start to develop a proactive approach to claims leakage reduction as a result of fraud.

SIUs can analyze data at scale, making better business decisions on behalf of the customer and the business. The starting point for this approach is a data lake—a huge repository of claims data of all types, from notes, diaries, and social media, to sensor information, external data, and accident reports. Smart firms enrich their existing data sets with real-time information and third-party data sources, such as weather, traffic, and news feeds, reports consultant EY. Data science tools, like Apache Spark, allow firms to mine this lake of information in detail and undertake high-quality link analysis across multiple data sets. The result is a proactive approach to fraud prediction that uses continual learning to improve detection and evaluation.

Building Better Fraud Investigation

In short, effective claims fraud analysis requires a combination of techniques. The experience and judgment of your professional staff must be married to traditional business rules and indicators, including statistical analysis and red-flag indicators. Most importantly, your business must take advantage of advanced developments in predictive analytics and artificial intelligence (AI).

AI can identify underlying links, joining data across structured and unstructured sources to find common trends. Rather than relying on resource-intensive human analysis, AI can take on the hard work of processing huge amounts of information from disparate sources quickly. Advanced analytics and libraries, like Spark’s GraphX, allow insurers to query intricate and connected networks of information. With machine learning, it’s easier for insurance firms to identify patterns, such as the same witness appearing in multiple claims.

The visualization of this information means it’s easier for SIUs to look for clustering data and highlight potential collusion. Professional staff can then visualize correlations in a platform. This automation of links across multiple data sets means insurance specialists simply need to “walk the graph” to identify leakage.

The traditional, business rules–focused approach to insurance fraud detection means professional staff are drowning in a sea of data and variables. The power of machine learning gives insurers the ability not only to combat fraud but to prevent their business from covering fraudulent actors in the first place.

Find out more about how your organization can take a proactive approach to combating insurance fraud.


Leave a Reply

Your email address will not be published. Required fields are marked *