Role of Analytics in Insurance Claim Fraud Detection
Globally, insurance fraud is a major concern for Insurers which continues to increase year by year. Claims fraud is the most common buzz around the P&C Insurance industry with auto and workers compensation business segments being the major contributors. Frauds are typically an individual or a group led effort of fraudsters with an intention of inflating claims and finally making profit out of a loss. Insurers spend huge effort and manpower in detecting fraud which as a net result not only drain the dollar amount from the insurer’s kitty but also adversely affect the good risks that they underwrite. It is also a social risk as it promotes financial crimes and penalizes the society.
Thereby defying the myth that, “Fraud in insurance is victimless.”
Following are the key aspects around claim fraud, due to which imperatives of the insurance ecosystem are impacted in different degrees of severity.
- Underwriting: Claims Fraud impacts underwriting guidelines and policies and deteriorates the insurance risk pool.
- Social Costs: Due to claim fraud, the prices of insurance go up as a whole.
- Unfair with rightfully deserving: Due to claim fraud sometimes even the rightfully deserving claimants are either denied claim or have to provide additional proofs on a claim.
- Undetected fraud encourages more fraud: When to a successful fraud propensity of individuals to further indulge in fraud increases, thus encouraging more fraud.
- Loss in Reputation: Repetition of fraudulent claims for an insurance company causes loss in market reputation thus causing a decline in competitiveness.
- Customer Relationship: Fraudulent claims adversely impacts insurer’s relationship with its existing customers and with prospects.
- Regulatory Compliance: Repeated frauds or even a few major frauds might cause serious legal issues for an insurer with the regulators.
- Loss of faith: Due to fraudulent claims people’s trust in insurance declines, which is detrimental to the growth of insurance industry.
Traditionally, insurance companies have been relying on expert judgment of agents, adjusters and special investigation units to detect and deal with frauds. This approach worked to a certain degree in the past as the agents of fraud themselves were not as evolved as they are now. Also, the number of claims were relatively small which made it humanly possible to keep a track on fraud.
However, existing challenge with the expert judgment is that, not all claims can be put to scrutiny as humans can only process a given number of claims to detect frauds. A huge effort and bandwidth of insurance claim experts would go into scrutinizing the claims for fraud, as these days not only the sophistication of insurance fraud has increased but also the number of cases for fraud has been on a high as well. Thus it has become humanly difficult to deal with the Insurance fraud with the expert judgment.
How can we leverage technology to mitigate the challenge? This is where data analytics can come to rescue where an algorithm can predict a potential fraud and then an expert can look at it. This way the process is far stream lined and can utilize the expertise of an individual. Here are few key elements as per the Insurance industry that the fraud detection algorithm may consider:
- History of referrals to Special Investigation Unit (SIU):
On the basis of past referral to SIU, the probability of referral to SIU can be calculated and if it is above an acceptable value experts should be involved. Here a model can be built that can detect a probability of claim to be referred to an SIU. Using past Special Investigation Unit, referral data and pattern recognition, for example, using the automated investigation scoring if the investigation score is above a threshold flag the claim analytical techniques like Artificial Neural Networks (ANN) can be used in this case.
- History of claims that were denied:
This can also serve as an indicator as consistent claim denial can be an indicator of potential fraud. Again, past claim denial records can be easily scrutinized by an algorithm and can provide probability of its future denial, for example, Claim Negotiation Patterns, Risk Indicators for phone, SSN, Address and so on. This can be achieved by Data Mining techniques like clustering, where high claim frequency clusters might be formed around specific addresses, phone numbers and so on. This will help in classifying the claims into various bins and each bin might need a different degree of attention.
- Network of individual analysis:
There might be a certain group of people or entities that consistently come up in a fraudulent insurance claim. The presence of this entity can trigger a flag for potential fraud. There can be pattern recognition algorithms and models that can be built which pick up a patterns according to a specific group of individuals consistently appearing in insurance claims.
- Social media analytics:
There might be a mismatch between a claimant’s actual social media profile and what it should be if struck by a large accident. For example a person struck by a loss would be looking to get this life back on track. Whereas, a person who has been receiving fraudulent claim might try to flaunt lifestyle.
- Text mining:
There can be a correlation between networks discussed in point 3 and kind of messages they convey to each other in combination of point no 4 above with respect to communication pattern. This can be achieved by extracting the posts on social media and forming a bag of words or key words, and cleaning the text thereby, forming ‘text- document matrix’. Using various techniques like Logistic Regression, can be applied to generate actionable information.
The tools and techniques that were useful in the past relied largely on intuition and fewer facts. This seldom serves the purpose in current and coming times. Current frauds are largely done in an organized manner and without sophisticated analysis, it will be difficult to identify them. Also the very fact that fraudsters tend to follow the path of least resistance, the organizations that continue to rely on traditional practices to detect fraud will be the easy targets. However, the various statistical and analytical techniques discussed in this blog can be utilized to build models leveraging newer technologies to prevent, detect and filter frauds. This will act as a swift enabler and thereby improves claim adjustment expenses, reduce overall claims leakage and so on, which helps to better the loss ratio finally, improve the efficiency and performance of insurers claims unit stakeholders.
Do you think the rapidly increasing insurance claim fraud can be controlled with the above fraud detection algorithm? Share with us your thoughts at firstname.lastname@example.org