Implementing Analytics for Claims Fraud Title Investigation

Transcription

1 CONCLUSIONS PAPER Implementing Analytics for Claims Fraud Title Investigation Considerations, change management and new capabilities for insurance company SIUs

2 ii Contents It s a victimless crime, right?... 1 Analytics for fraud detection... 1 Start with a focus on data quality...2 Assemble an interdisciplinary project team...3 Prioritize requirements...4 Acknowledge the need for change management...4 Grange Mutual Casualty Company before and after... 5 How do you measure success?... 6 Closing thoughts... 8

3 1 It s a victimless crime, right? An exaggerated accounting of losses; an inflated value for stolen property; an auto claim from a staged accident; medical charges for nonexistent conditions it s all small potatoes, a victimless crime, fair compensation for rising premiums and deductibles, right? That attitude seems to prevail among business and consumers these days but fraud actually makes victims of all of us. According to FBI estimates, the total cost of insurance fraud (non-health insurance) in the US is more than $40 billion per year losses that must be made up in premiums. 1 Chances are, at least 10 percent of insurance claims contain some flavor of fraud. Insurance companies have responded by establishing special investigative units (SIUs), many of them armed with computer-based tools to detect and prevent fraud. Yet the problem continues to grow. Why is that? For one, many insurance companies rely on rules-based systems that assess claims against if/then business rules and preset thresholds. Traditional rules and outlier detection methods are useful to address known patterns of fraud but are not very good at addressing the sophisticated and evolving fraud schemes we re seeing today. Second, insurance companies often operate with silo data systems, making it difficult or impossible to get big-picture perspective a complete view of a customer, account history or transaction path. The company will miss fraud that is only apparent when viewed across entities. Analytics for fraud detection Three forms of advanced analytics are taking center stage in the war on fraud: Predictive modeling reveals patterns among data elements that point to a high propensity for fraud. With predictive insight, insurers move more into prevention mode versus pay-and-chase. Social network analysis, also called link analysis, reveals connections among entities to expose organized fraud rings or collusive activities. Machine learning adapts to changing behaviors in a population through automated model building. With every iteration, the algorithms get smarter and deliver more accurate results. If your company hasn t already invested in some form of fraud analytics solutions, it will. Even if your company already has some type of fraud detection analytics, these solutions always evolve, so you will need to upgrade and implement new technology. Whether for a new installation or an upgrade, a tremendous amount of investment and C-level scrutiny goes into these projects, so it s critical to get the implementation right. 1 accessed October 12, 2016

4 2 In an event co-sponsored by SAS and the International Association of Special Investigation Units (IASIU), Kim Kuster, manager of the intelligence unit within the SIU at Grange Mutual Casualty Company in Ohio, shared her experiences with a major system implementation. Start with a focus on data quality According to the Coalition Against Insurance Fraud, more than two-thirds of insurance companies cite data integration and poor data quality as a big impediment to implementing analytics. 2 So a focus on data is a good place to start. Limited IT resources Excessive false positive rates Data integration and poor data quality Lack of cost/benefit analysis (ROI) Delayed claims adjudication SIU cannot handle claims volume 0% 25% 50% 75% 100% Source: Coalition Against Insurance Fraud, Figure 1. More than two-thirds of insurance companies struggle with data integration and poor data quality. Invest in data quality. Data is an organization s most valuable resource, but it is the weakest link for many insurance companies. Think of it like raw oil in the ground. Analytic solutions extract and refine this raw material to produce insights. But those insights will only be as good as the input. Whether you run it through rules, anomaly detection or advanced analytics, if you re starting with poor quality data, it s the old adage: garbage in, garbage out. Break down the silos. Departments across the company actuarial, underwriting, marketing, claims, loss control, etc. are all collecting data for their own purposes. Imagine the power of integrating all this data in an enterprise repository. It makes sense to invest in the right types of data integration and data management capabilities, such as entity resolution. Entity resolution enables you to join records from one data source with another that describe the same entity, such as Jon W. Insured III, J.W. Insured and John Insured. 2 Source: Coalition Against Insurance Fraud, 2016

5 3 Insurance Information Foundation Data Warehouse Claims Fast Tracking Litigation Propensity Creeping Catastrophe Claims Fraud and Build-Up Vender Efficiency Marketing Customer Intelligence Campaigns Customer Service Customer Retention Underwriting Selection Policy ANALYTICS Time Persistent In-Memory Capabilities Agency Channels Distribution Loss Control Internal Audit Cybersecurity Actuarial Pricing Rating Figure 2. An enterprise fraud system enhances the entire organization s analytic maturity. Look at the business processes that collect the data. How accurate and complete is the information gathered on new account applications, at policy renewal and from contact center activity? Are the right questions being asked in the first notice of loss (FNOL) process? Is full and accurate information being collected and promptly updated to the claims system? Some insurance companies have fairly ad hoc processes for gathering this data, and others are more structured, said Kuster. The better and more complete that information on the front end, the more powerful the insights will be. Bring in third-party data. Companies can store and analyze more data than ever, thanks to the plummeting cost of data storage and exponential leaps in processing power with techniques such as in-memory processing. You can greatly bolster analytic solutions by bringing in relevant industry data, such as from the National Insurance Crime Bureau (NICB) and Verisk ISO research for the property/casualty insurance industry. Assemble an interdisciplinary project team Often you see IT move forward with IT to address a business challenge, without engaging the business domain experts, said Kuster. That s a project that is doomed to become shelf-ware, because it didn t consider the dynamics behind the problem, the day-to-day workflow of investigators, and the business challenges, which can be unique for every area in which the company operates. If end users didn t have a stake in defining the system, they re not likely to get what they want and are certainly less likely to trust and use it.

6 4 We needed to engage our IT folks and have them weigh in, of course, but we did not want a department outside of SIU to just deliver something to us that wasn t going to be usable, said Kuster, whose SIU did a major analytics implementation in So we had SIU heavily represented on our project team. They were diving into the end design and making recommendations and feedback back to the project team. Prioritize requirements The joint project team should work with various types of end users to gather and prioritize requirements for the analytics system. Build from the business, not from the technology. Start with a deep understanding of how the SIU process works. What are the operations that are important to support? How does work flow from initial identification to detection and investigation? What matters in identifying fraud? What key performance indicators are needed? The most successful implementation will be framed not by IT or the bells and whistles of the technology, but on how best to support the work and information needs of adjusters, investigators and business leaders. Manage scope creep. Distinguish between the must haves and the nice to haves. What really adds value? It s very easy for folks to want to add into a project, and certainly you have to be flexible to make small changes in a project design as you go, but there has to be a level of discipline to manage scope creep, Kuster cautioned. Acknowledge the need for change management A successful analytics solution will radically change life for an SIU. Even if it is delivering positive results, there inevitably will be an element of change management to improve acceptance and adoption. If you re an SIU investigator or even a manager, and your new analytics system is now so effective that it identifies claims fraud in two or three days instead of four to six weeks, that s going to take some adjustment. Processes will have to adapt to investigating claims that come quite early in the claims cycle, and it may take some time for investigators to trust the analytics. Once the analytics solution goes into production, there needs to be continued oversight and monitoring to help ensure adoption, especially if the organization has not been through a major analytics project before. Start with something that s manageable, where you can show early success, and then it becomes easier to get buy-in to pursue other projects related to fraud analytics. When in doubt, get advice from the actuarial group, which is typically ahead of other functions in the company in its maturity in data management and analytics. Business processes will need to be modified now that potentially fraudulent claims can be identified in an average of three days compared to 47 days with manual detection.

7 5 Steps in a successful implementation process Engage business domain experts. Understand business results and enterprise dynamics. Demonstrate initial business case metrics. Conduct project team workshops. Engagement Deeper understanding Gain insights into data and process. Understand operations and key performance indicators. Implement change management to facilitate adoption. Monitor system performance through enterprise metrics. Identify strengths, gaps and needs. Tune the solution to sustain desired performance levels. Production Design and implementation Get end user feedback. Make design adjustments as needed. Develop solution metrics. Grange Mutual Casualty Company before and after Grange Mutual Casualty Company of Ohio is a super-regional property and casualty carrier operating in 13 states. The company has a relatively small SIU department, with 15 staff members, including 10 field investigators (one dedicated to major cases) and three analysts. Before we implemented an analytics system, investigators had the autonomy to decide whether to accept or decline assignments, said Kuster. Two analysts who were primarily dispatching the manual referrals, and one desk investigator handled nonstandard auto and lower-severity property claims. The business challenges. We wanted to increase our casualty referrals to SIU, and the methodology to detect that without analytics is very difficult to do, especially when payments are often made to attorneys, said Kuster. Our case detection was largely reactive, and we had some data quality issues as well. We wanted to resolve the issues of bad or missing data and break down the data silos within the organization to be able to holistically review our data set. We also wanted to increase our ability to bring in third-party data, whether from NICB, ICO or policy data. We had limited resources available, but as an organization we decided that implementing an analytics solution was something we needed to do. Selecting a solution. When Grange was considering a fraud analytics solution, we were looking to move beyond basic rules to a more robust solution. The solution had to be easy for end users to adopt in their day-to-day workflow.

8 6 When we looked at available vendors to partner with, we considered the impact it would create for the IT team. We evaluated vendors industry experience and analytical maturity. We really wanted somebody that was dialed into what we were trying to do. The company s SAS analytical solution has been in production for about a year and a half. Analysts use a combination of advanced search and filters to identify high-scoring claims and make referrals to SIU, claim supervisors, adjusters and underwriters. The great thing about analytics is that it s always running, so we get quality referrals 365 days a year, said Kuster. Results. Already Grange Insurance has seen a dramatic difference in the speed of referrals coming from the solution. The biggest feedback I ve gotten is that these are highquality referrals with a lot of avenues to pursue a lot of actionable intelligence to move on, said Kuster. It s not just one or two things to clear, and then back-and-forth dialog with the front-line claims folks to get answers; the referrals have a lot of meat on the bone from day one. We nearly doubled our capacity without abandoning manual referrals. The consistency in evaluating each claim is also an important benefit. We know we will eventually be deposed on some claims that have been identified through the solution. It s comforting to know that our investigators will be able to show a list of the datadriven indicators that caused the claim to be referred to them, as opposed to maybe just a gut feeling from an adjuster. Kuster also noted three ancillary benefits. The fraud analytics project has: Boosted overall analytical maturity for the department and the organization. Identified new ways to improve data quality, such as making certain fields within the claims system mandatory and having data elements selected from a drop-down box rather than being free-form. Created a common ground that has encouraged greater collaboration among departments. Our analysts are interacting with folks from underwriting, sales and marketing, and other departments in ways that we didn t have prior to having analytics, Kuster said. How do you measure success? According to the 2016 Coalition Against Insurance Fraud study, 50 percent of insurance companies measure the success of their fraud detection technology by the fraud detection rate, 16 percent look at the number of referrals to SIU, 4 percent look at the loss ratio, and 2 percent look at average days to detect fraud. 3 Surprisingly, 20 percent of respondents in the survey said their companies don t measure any of these results. At Grange, we measure all of these factors, but we also compare and contrast results achieved with the solution with the results of our previous, manual and rules-based process which leads to some interesting discussions, said Kuster. 3 Coalition Against Insurance Fraud 2016

9 7 Source: Coalition Against Insurance Fraud, 2016 Figure 3. How fraud technology success is measured Benefits of a successful SIU analytics implementation Respondents to the 2016 Coalition Against Insurance Fraud survey cited these top six benefits gained from analytics-based fraud detection: Higher-quality referrals. The feedback from users here at Grange speaks to higherquality referrals and a better understanding of why investigators are getting the referral in the first place, said Kuster. More referrals. It s key to have the ability through ranking and stacking fraud risk to ferret out that 10 percent of claims that have the highest propensity for fraud. High-performance processing for those analytics yields more quality referrals. Uncovering complex or organized fraud. Advanced analytic techniques such as machine learning and social network visualization can identify collusion that rulesbased systems would miss. Improved investigator efficiency. Compared to manual or rules-based methods, analytics delivers fewer false positives, so valuable SIU resources are not wasted on claims that prove to be legitimate. Increased mitigation of losses. Now that potentially fraudulent claims can be identified six times faster on average, how much is not being paid? Predictive analytics can prevent improper payments from going out in the first place. More consistent claims investigation. When claims are scored through technology, algorithms and analytics, there is intrinsic objectivity and lack of bias in flagging claims for review.

10 8 Higher-quality referrals More referrals Uncovering complex or organized fraud Improved investigator efficiency Increased mitigation of losses More consistent claims investigation Better understanding of referrals Enhanced reporting Other Source: Coalition Against Insurance Fraud, % 19% 38% 56% 75% Figure 4. Perceived benefits of anti-fraud technology 4 Closing thoughts Small companies might not be able to justify an enterprise fraud analytics system, but that doesn t mean they are limited to manual methods. There are out-of-the-box solutions that can be mapped to a preconfigured data model and have some analytics already in them. Implementation is relatively simple and quick, and you can start to get results within a few months. The packaged product approach won t offer the same flexibility and customization in configuration, user interfaces or modeling, but it can deliver some lift compared to basic manual and rules-based approaches. Whether you re a front-line investigator, analyst or in senior SIU leadership, it s an exciting time to be in the industry, and there are a lot of powerful tools out there. Having whatever fraud technology the budget will allow will enable you to improve detection and reduce losses. Having a comprehensive fraud analytics platform will certainly keep SIUs a step ahead of the fraudsters. 4 Coalition Against Insurance Fraud 2016

11 To contact your local SAS office, please visit: sas.com/offices SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. Other brand and product names are trademarks of their respective companies. Copyright 2017, SAS Institute Inc. All rights reserved _G