Fraud is an increasing problem that affects a wide range of industries including insurance where fraud cases worth millions of pounds are uncovered every day in the UK. The Insurance Fraud Enforcement Department only in the last 3 months has been bombarded with over 15 insurance cases. Cases include:
- Fraudsters defrauding £66k from dead parent’s pension payments
- Birmingham pair being arrested in crash for cash investigation
- Fraud gang being sentenced over £500k false insurance claims
- Man being sentenced over £56k of home insurance fraud
- Man being jailed over £114k insurance scam
- Insurance saleswoman being sentenced for £31k commission scam
- Man being convicted of falsifying documents in £300k insurance claim
Much of this insurance fraud occurs during the process of purchasing, selling or using insurance. For example: a) Application Fraud (aka pre) can happen when false information is provided during the application for an insurance policy with the purpose of defraud and b) Claim Fraud (aka post) where a false claim is made after the policy is purchased.
The insurance industry is inherently at risk to claim fraud which has been the subject of major concern since insurance was first written. This risk has resulted in a mature use of data mining techniques to prevent and identify fraudulent activity. Data mining techniques are providing great aid in financial accounting fraud detection. However, in certain cases researching data mining based fraud detection is challenging due to legal reasons (i.e. difficulties in acquiring real-world industry data).
Fraud management strategy
A successful fraud management strategy consists of three elements: Prevention, Detection, Investigation and Resolution. The purpose of a fraud prevention strategy is to lower the possibility of fraud occurring. Technical solutions may include a scoring system that rigorously check insurance applications for possible fraudsters. In addition to technical solutions, law enforcement can impose stricter penalties to fraudsters when infringements of the insurance industry rules are detected. When prevention mechanisms fail, fraud detection mechanisms are used to identify dishonest claims. If fraud is detected, investigation types such as audits, surveillance and interviews are employed and appropriate actions are taken for the fraud to be resolved
Existing insurance fraud detection systems
Existing insurance counter-fraud solutions include the SAS Fraud Framework for Insurance. SAS is able to detect, prevent and manage insurance claims fraud. The framework depends on multiple data mining techniques such as segmentation, association rules, and classification techniques to detect fraudulent claims. This method uses several techniques such as predictive modelling, text mining, exceptions reporting and link analysis to discover the likelihood for fraud. It can process claims real-time and scores the results according to their severity. Case studies have shown that SAS can save a considerable amount of money from fraudulent claims. For example, Allianz Insurance, a well-established worldwide insurance company, have identified over 1,000 fraud cases worth more than 62 million CZK in the first six months after the application was implemented and saved circa 110 million CZK a year by reducing the number of fraudulent claims paid.
Insurance Hunter is an online fraud detection and prevention solution, hosted by Experian, a leading global information services company. Insurance Hunter can identify inconsistencies on policy and claims, combining information from multiple insurer databases. Its detections ability is based on a set of proven fraud indicators which can be customised to meet business rules. It checks claim data for anomalies within the current application and against previous claims and matches against previous known and suspected fraudulent claims. The check can be done in real-time and the results are displayed with red/amber/green status. In addition, Insurance Hunter supports the analysis of fraud trends and patterns which enables the evaluation and fine-tuning of the fraud indicators as well as the analysis of the value of the fraud prevented.
One of the main challenges on building insurance claim fraud detection models is
misclassification errors which can be either of the following types:
- Type I: the model misses to detect a fraudulent claim.
- Type II: the model falsely classifies legitimate claims as fraudulent.
Type I errors incur more expenses to the insurer due to the claim settlement and lays the path for more fraudulent claims. Type II errors can clog up the list of claims that need further investigation and slow down the process of legitimate claims which can have a negative impact to the business overall. Thus, it is vital for a fraud detection model to minimise both types of misclassification errors.
Another issue insurance fraud detection models are faced with is to distinguish an honest claimant from a fraudster based on demographics and personal characteristics. Professional fraudsters tend to use false identities and evolve their methodologies which makes them indistinguishable from authentic claimants and counter detection systems.
Fraud is constantly evolving. Once a fraud technique is uncovered, fraudsters conceive a new tactic and find new ways to trick the detection systems until the systems detect them. Thus, fraud detectors must be adaptable and evolve with the constant changes on fraud techniques. Detection models must be continually updated on time with new information to ensure accuracy. Also, several styles of fraud can occur at the same time. Each style can have unique characteristics. Fraud detection models must be able to detect anomalies for each style.
Challenges in developing fraud detection models also include the time that it takes the model to detect fraud and whether the detection can be performed real-time or in a batch process.
Patterns as well as the cost of fraud continuously evolves any fraud system could quickly become obsolete. Thus, a robust fraud management system should incorporate a number of different fraud detection methodologies, each one contributing towards calculating a claimant’s fraud likelihood.
However, fraud detection models cannot give definitive proof that fraud has been committed without domain experts’ feedback. Therefore, any fraud detection system should be used with care and should complement and not replace existing human specialists.
This article was written by Christina Maniati from Capgemini: Insights & Data Blog and was legally licensed through the NewsCred publisher network.