Summary: Insurance people are slow to change. For years claims adjusters took a statement from an insured or claimant in long hand on lined paper. Premium was calculated using a ten-key calculator. Claims were analyzed based on the experience, knowledge, and professionalism of the claims person, and still are. Computers were first used by claims people only to enter log notes of the activities of the adjuster, which were first written on lined paper in an adjuster's working file. Field adjusters got their phone messages by stopping into an agent's office and using the office phone. Photos were taken with Polaroid cameras. Today, however, computers have become an integral part of insurers' processes, from underwriting to claims adjusting to researching fraud.

Insurers realize that no one really knows the extent of insurance fraud because most fraudulent insurance claims succeed. They do know that insurance fraud is a major drain on the ability of insurers to make a profit but cannot properly quantify the extent of insurance fraud. At best, insurance fraud professionals can only guess at the total amount. The best guesses are that insurance fraud takes at least 3 to 10 percent of all claim payments or 3 to 10 percent of all premiums collected. Insurers are looking to computers, and what has come to be known as fraud analytics, to obtain new solutions to the growing threat presented by insurance fraud whether perpetrated by normally honest people or by organized fraud rings.

Insurance fraud schemes are limited only by the creativity of the fraud perpetrators. They include arson for profit, fake thefts, inflating legitimate claims, embezzlement of premiums, the swoop and squat, and many other common schemes.

The early attempts to reduce insurance fraud depended on the professionalism of the claim departments and their claims adjusters. That effort, although somewhat successful, barely made a dent in the crime of insurance fraud. Legislators, concerned about the increases in insurance premiums to cover the losses due to fraud, passed Insurance Frauds Prevention Acts that required insurers to investigate every potential insurance fraud, to institute Special Investigative Units (SIUs) and to work to prosecute insurance fraud perpetrators and punish insurers who failed to institute SIUs or effectively work against insurance fraud. See States Treatment of Insurance Fraud and Fraud Requirements for Insurance Companies.

In the 1960s, statements were taken on paper or using a ten pound recorder that used plastic disks to record the statements of the insured and/or claimant. Insurance fraud, if detected, was detected by accident or by knowledge of certain ancient red flags of fraud. Each adjuster was provided with lists of red flags of fraud and, if they appeared, the adjuster would conduct further investigation or ask a lawyer to examine the insured under oath, to gain enough information to pay or reject the claim. Analysis of a claim to determine if it is legitimate or potentially fraudulent was performed in the human brain of the adjuster, claim supervisor, or manager.

For more than a decade insurance fraud investigators have used the information in data bases like those maintained by

·the SIUs set up by insurers to detect, investigate, and prevent fraud as required by state statutes

·the forty-one state governments in the U.S. that sponsor fraud bureaus to investigate insurance fraud

·the National Insurance Crime Bureau (NICB)

·the Coalition Against Insurance Fraud

·the ISO ClaimSearch® database

·other available databases

It is now the twenty-first century. Computers are now common in underwriting and adjusting, in evaluating damages, and in determining replacement costs.

Insurers now believe that an effective fraud analytics solution can cut costs and increase customer satisfaction. Fraud analytics performed by a computer software program are assisting the claims person in making the decision to perform additional investigation directly or through the services of the insurers SIU. Data analytics are being used to detect and, hopefully, defeat fraud.

Insurers have learned that in order to effectively test and monitor claims situations for potential attempts at fraud, they need to look at every transaction that takes place and test them against established parameters, across applications, across systems, and from dissimilar applications and data sources.

One of the key aspects of data analytics is the ability for the technology to maintain, review, analyze, and access comprehensive logs of all activities performed by the claims staff. Management can run an application or a script, enter some data, and find anomalies if they exist. But it doesn't stop there; the insurer still needs to gather evidence that can be used as proof of what the insurer did to uncover that fraudulent activity, and what that fraud activity was. That proof has to be specific and detailed enough to stand up to further fraud investigation, litigation, or even prosecution. Anomalies are not enough. There must be proof of a misrepresentation or concealment of material fact provided to the insurer with intent to deceive the insurer to its detriment before a person can be accused of fraud and a claim denied because of the fraudulent actions.

For fraud analytics to work it is necessary to collect sufficient data to predict frauds. Over time insurers have developed lists of fraud indicators, or red flags. When the red flags are available, the software can compare the red flags to the claims information, adjuster's log notes, transcripts of recorded statements, and transcripts of examinations under oath and depositions. See [Master Red Flag Checklist.xml^Master Red Flag Checklist^Master Red Flag Checklist].

Analytics solutions have evolved from the basic list of red flags to predictive and prescriptive model-based solutions to help insurers fight fraud. Predictive models are available that can attach a fraud propensity score to each claim so that the insurer can then assign an SIU investigator to gather evidence that can be used to make a decision with regard to a suspected fraudulent claim. These algorithms learn from positive detection of fraud that was confirmed by a thorough SIU investigation. The self-learning algorithms help to reduce false claim notations to limit the need for a field SIU investigation. This also speeds processing as legitimate claims can be paid quickly and don't get caught up in unnecessary investigation if one or two parameters are askew.

Computers use many parameters to analyze information to detect possible fraud. They include the following:

1. claims history analytics that reveal frequency, type, and overall claim amounts

2. analysis of first notice of loss to compare red flags to information given

3. analysis of statements of value to separate genuine claims from exaggerated and fraudulent claims

4. analysis of social network posting of the various entities involved in a claim

5. analysis of social media of claims for red flags of fraud

6. analysis of loss documentation to find exaggerated claims

7. analysis of the location of the claimant to the geographical extent of a disaster to eliminate claims that are filed from areas that are not located in the area impacted by the disaster

Big data processing takes into account transaction and interaction, as well as observational data. In a typical big data architecture all of these data types are arranged in forms of structured, semi-structured, as well as unstructured formats through advanced data transformation techniques.

With anomaly detection, key performance indicators associated with tasks allow various thresholds to be set. When a threshold for a particular measure is exceeded, then the event is reported. Outliers or anomalies could indicate a new or previously unknown pattern of fraud.

This type of tool is straightforward, easy to implement, and useful for evaluating individual performance and identifying employee training opportunities. Once in place, the system functions automatically. Adjuster activities are monitored, and problems can be identified and corrected.

Another anti-fraud tool combines ad hoc query and online analytical processing, enabled by databases that summarize across many different dimensions. The program allows analysts to search through huge volumes of adjudicated claims, make comparisons, identify exceptions, and find unusual situations in a dynamic environment. An experienced analyst can take the data and quickly generate reports that identify potential problems and direct future investigations more effectively.

In recent years, many insurers have turned to predictive modeling processes, reducing the need for tedious hands-on account management. Quantitative analysts use data mining tools to build programs that produce fraud propensity scores. Adjusters simply enter data, and claims are automatically scored for their likelihood to be fraudulent and made available for review.

Predictive modeling tends to be more accurate than other fraud detection methods. Information can be collected and cross-referenced from a variety of sources. This diversity of resources provides a better balance of data than the more labor-intensive flag system. However, model performance deteriorates with age. As criminals adopt new approaches, models must be updated to reflect new patterns. In spite of these limitations, predictive modeling shows great promise.

There have already been many success stories that have come out of big data analytics. Various products like Hadoop, Lexis Nexis HPCC, Teradata AsterData, Kognito, and Microsoft Dryad are available in the market and all of them have been quite successful in providing valuable insights in various industries.

Insurance companies should consider investing in technology to prevent claims fraud before it reaches epidemic proportions. Technology-based tools to fight insurance fraud can be used individually or in combination to help companies detect and prevent criminal claim activities.

Some fraud detection techniques screen claims during processing and help prevent improper payments. Others involve retrospective analysis of adjudicated claims and help uncover the activities of fraud rings, internal fraud, and leakage. Together, these techniques are powerful deterrents for would-be fraudsters who seek to profit at the expense of insurance companies and their policyholders.

Insurers must recognize that data analytics is not a panacea. It is a tool, like the adjuster's tape measure or recorder. It will not detect fraud by itself. It will delve through a great deal of data and give the insurer sufficient information to allow the insurer to better use its SIU investigators. It is the claims person and the SIU investigator working with an insurance coverage lawyer who must put together sufficient admissible evidence that can be presented in court to defeat what the insurer believes is a fraudulent claim.