STRATHMORE UNIVERSITY
A FRAUD INVESTIGATIVE AND DETECTIVE FRAMEWORK IN THE
MOTOR INSURANCE INDUSTRY: A KENYAN PERSPECTIVE
By
KISAKA GEORGE NGOSIAH
A thesis submitted in partial fulfillment of the requirements for the award
of the Master of Science in Information Technology degree
I certify that this thesis is my original work and all material in this thesis which is not my own work has been identified. I further certify that no material has previously been submitted and approved for the award of a degree by this or any other University. This thesis is available for library use on the understanding that it is copyright material and that no quotation from the thesis may be published without proper acknowledgement.
SIGNED……….DATE……….. Kisaka George Ngosiah
FIT/060478/10
This thesis has been submitted with my approval as the University Supervisor. Prof. Vitalis Onyango-Otieno.
Director, Centre for Applied Research in Mathematical Sciences Strathmore University
First I am thankful to the almighty God who has given me life and good health throughout my academic program. Without Him I would not have reached this far. I thank my research thesis supervisor Prof. Onyango-Otieno, for the invaluable time he spent guiding my efforts. Special thanks to Dr. Marwanga and Dr. Ateya of Strathmore University too for the invaluable input during the seminars.
I would also like to acknowledge the remarkable input received from Mr. Wandera, Chairman of the Association of Kenyan Insurers (AKI). Thank you for making it possible to access invaluable data for analysis.
Finally I would like to acknowledge the input from friends and colleagues that lent a hand directly or indirectly towards this effort. For the fantastic support you gave me please accept my sincere gratitude
Insurance fraud is a serious and growing problem, with fraudsters’ always perfecting their schemes to avoid detection by the basic approaches. This has caused a rise in fraudulent claims that get paid and increased loss ratios for insurance firms thereby diminishing profitability and threatening their very existence. There is widespread recognition that traditional approaches to tackling fraud are inadequate. Studies of insurance fraud have typically focused upon identifying characteristics of fraudulent claims and putting in place different measures to capture them.
This thesis proposes an integrated framework to curtail insurance fraud in the Kenyan insurance industry. The research studied existing fraud detection and investigation expertise in depth. The research methodology identified two available theoretical frameworks, the Bayesian Inference Approach and the Mass Detection Tool (MDT). These were compared to comprehensive motor insurance claims fraud management with respect to the insurance industry in Kenya.
The findings show that insurance claims’ fraud is indeed prevalent in the Kenyan industry. Sixty five percent of claims processing professionals deem the motor segment as one of the most fraud prone yet a paltry 15 percent of them use technology for fraud detection. This is despite the fact that significant strides have been made in developing systems for fraud detection.
These findings were used to determine and propose an integrated ensemble motor insurance fraud detection framework for the Kenyan insurance industry. The proposed framework built up on the mass detection tool (MDT) and provides a solution for preventing, detecting and managing claims fraud in the motor insurance line of business within the Kenyan insurance industry.
TABLE OF CONTENTS
DECLARATION ... i ACKNOWLEDGEMENT ... ii ABSTRACT ... iii TABLE OF CONTENTS ... iv LIST OF TABLES ... viii LIST OF FIGURES ... ix DEFINITION OF TERMS ... x LIST OF ABBREVIATIONS ... xii CHAPTER 1 ... 1 INTRODUCTION ... 1 1.1 Background ... 1 1.2 Problem Statement ... 3 1.3 Research Objectives ... 4 1.4 Research Questions ... 4 1.5 Justification ... 51.6 Scope and Limitations ... 6
1.7 Theoretical Frameworks ... 6
CHAPTER 2 ... 7
LITERATURE REVIEW ... 7
2.1 Introduction ... 7
2.2 Insurance Fraud in the Kenyan Industry ... 10
2.3 Motor Insurance Fraud in the Kenyan Industry ... 11
2.4 Current Investigative and Detective Methods ... 13
2.4.1 Auditing and the Optimal Contract ... 14
2.4.2 Co-opetition and Common Databases ... 15
2.4.3 Rule Based Approaches ... 15
2.4.4 Government Regulation ... 16
2.5.1 The Mass Detection Tool for early detection of insurance fraud ... 18
2.5.2 Bayesian Inference Approach ... 18
2.6 Conclusion ... 19 CHAPTER 3 ... 21 RESEARCH METHODOLOGY ... 21 3.1 Introduction ... 21 3.2 Study Population ... 22 3.3 Sampling Frame ... 23
3.4 Sample Size and Error ... 23
3.5 Research Design ... 24
3.6 Data Collection Methods and Techniques ... 24
3.6.1 Questionnaires ... 25
3.6.2 Secondary Sources ... 25
3.7 Data Analysis ... 25
3.8 Framework Validation ... 26
CHAPTER 4 ... 27
THEORETICAL MODELS FOR FRAUD DETECTION ... 27
4.1 Introduction ... 27
4.1.1 Bayesian Inference Approach ... 27
4.1.2 The Mass Detection Tool ... 28
4.2 A Summary of the two Models ... 30
4.3 Conclusion ... 31
4.4 The Conceptual Framework ... 31
CHAPTER 5 ... 33
ANALYSIS AND PRESENTATION OF FINDINGS ... 33
5.1 Introduction ... 33
5.1.1 Respondents’ Demographics ... 33
5.2 Existing Forms of Insurance Fraud ... 34
5.2.4 Life Insurance Fraud ... 36
5.2.5 Conclusion ... 36
5.3 Existing Methods for Detecting and Preventing Fraudulent Activity ... 37
5.3.1 Detection of Fraudulent Activity ... 37
5.3.2 Prevention of Fraudulent Activity ... 38
5.3.3 Other Approaches to Detecting Potential Fraud and Fraudsters ... 40
5.4 Use of Information Systems to curtail Motor insurance fraud ... 41
5.4.1 Current Forms ... 41
5.4.2 Future Expectations ... 42
5.4.3 External Sources ... 42
5.5 Interpretation of Findings ... 43
CHAPTER 6 ... 45
FRAMEWORK FOR FRAUD DETECTION BY KENYAN INSURANCE COMPANIES ... 45
6.1 Current Practice in Relation to Theory ... 45
6.2 Gaps in Current Theory and Practice ... 46
6.3 Conclusions Relating Theory and Practice ... 47
6.4 New Framework from Research Findings and Current Theory ... 47
6.5 The Proposed Framework ... 48
6.6 The Proposed Framework’s Execution / Validation ... 49
CHAPTER 7 ... 52
DISCUSSION, CONCLUSION AND RECOMMENDATIONS ... 52
7.1 Discussion ... 52
7.2 Conclusion ... 53
7.3 Recommendations ... 54
7.4 Directions for Future Research ... 54
REFERENCES ... 55
APPENDICES ... 59
Appendix A – Research Questionnaire ... 59
Table 1.1 - 1989 Bodily Injury Liability Claim Sample 5
Table 2.1 - General Business Incurred Claims Trend 11
Table 2.2 - Incurred Claims Ratio per Class of Business 13
Table 3.1 - Insurers Registered for Each Class of Business 22
Figure 2.1 - Increase in Questionable Claims in the US 7
Figure 2.2 - Detected General Insurance Claims Fraud 8
Figure 4.1 - The Mass Detection Tool 29
Figure 4.2 - The Conceptual Framework 32
Figure 5.1 - Insurance Industry Experience of Respondents 33
Figure 5.2 - Forms of Insurance Fraud and Their Prevalence 34
Figure 5.3 - Methods by Which Fraudulent Activity is Detected 38
Figure 5.4 - Methods by Which Fraudulent Activity is Prevented 39
Figure 6.1 – The Proposed Framework 48
Apparent Fraud: A claim in which there was no injury/loss or the injury was unrelated
to the accident
Buildup claim: A claim in which the injury/loss is exaggerated and/or the treatment is
excessive
Carrier: Insurer or Insurance Company.
Claim: An insurance claim. A formal request or application to an insurance company
asking for a payment or benefit based on the terms of the insurance policy.
Co-opetition: Insurance industry term referring to the sharing of information on claims
and other data across the industry
Ethnography: The branch of anthropology that provides scientific description of
individual human societies
Insurance Fraud: Any act committed with the intent to deceitfully obtain payment from
an insurer can be described as insurance fraud. Insurance fraud occurs when the insurer does not know all the facts about the insured and the claim, and when the fraudster believes that some monetary benefits can be gained by distortion of such facts
Insurance Fraud Bureau:A not for profit making organization in the UK funded by the insurance industry, specifically focused on detecting and preventing organized and cross industry insurance fraud
Insurer: See Carrier
Railway spine: A nineteenth-century diagnosis for the post-traumatic symptoms of
passengers involved in railroad accidents
Trip and fall: Accident that involves pedestrians getting their feet caught on an object
which causes them to fall
Whiplash: A range of injuries to the neck caused by or related to a sudden distortion of
ABI: Association of British Insurers AKI: Association of Kenya Insurers BBN: Bayesian Belief Network CRB: Credit Reference Bureau IFB: Insurance Fraud Bureau
IMIDS: Integrated Motor Insurance Data System IRA: Insurance Regulatory Authority
MDT: Mass Detection Tool for early detection of insurance fraud NICB: The United States National Insurance Crime Bureau PwC: PricewaterhouseCoopers
CHAPTER 1
1.0. INTRODUCTION
1.1 Background
The essence of fraud is deception (Blan and Hart, 1985). Whatever industry the fraud is situated in or whatever kind of fraud you visualize, deception is always the core of fraud (Jans, Lybaert and Vanhoof, 2010). Insurance fraud has existed ever since the beginning of insurance as a commercial enterprise and is a major problem in the United States at the beginning of the 21st century. It has no doubt existed wherever insurance policies are written, taking different forms to suit the economic time and coverage available.
From the advent of “railway spine” in the 19th century to “trip and falls” and “whiplash” in the 20th century, individuals and groups have always been willing and able to file bogus claims (Derrig, 2002). Many definitions of insurance fraud are in common use.
In this study, insurance fraud or claim fraud is used to refer to criminal acts that involve making the willful act of obtaining money or value from an insurer under false pretenses or material misrepresentations (Derrig, 2002). This therefore includes cases where a claimant fabricates or causes an accident to happen, to obtain payments they might otherwise not deserve.
In recent years, economic analysis of insurance fraud has developed along two lines. The first one is mostly theoretical and its foundations may be found in the theory of optimal auditing. It aims at analyzing the strategy of insurers under claims fraud or application fraud. This approach mainly focuses on questions such as: What should be the frequency of claim auditing and how do opportunistic policyholders react to the
auditing strategy? What are the consequences of potential fraud on the design of insurance contracts, especially with regard to the indemnity schedule? What is the deterrence effect of an auditing policy? What is the role of good faith when insurance applicants may misrepresent their risk?
The second branch of the literature on insurance fraud is more statistically based. This focuses mainly on the significance of fraud in insurance portfolios; on the practical issue of how insurance fraud can be detected; and on the scope of automated detection mechanisms in lowering the cost of fraudulent claims (Dionne, Giuliano and Picard, 2003).
In this paper, the researchers explore the possibility of an integrated framework encompassing these two complementary approaches to insurance fraud. As in other forms of fraud in today’s organizations, fraud detection and control will involve both manual intervention and system support. Manual intervention in insurance fraud detection involves picking up claims randomly for intensive investigation.
In a non-commitment Costly State Verification setting insurers can only detect fraudulent claims by performing costly audits, and policyholders are overcompensated by the optimal insurance contract (Schiller, 2006). The focus is on the deterrence effect of the auditing strategy and on the consequences of insurance fraud on the design of insurance contracts
System intervention on the other hand has taken several forms. Co-opetition, which is an industry term referring to the sharing of suspect claims data across the industry is one way. It is like having a central repository of suspect data. This has however had its challenges since regulatory support is needed. Rule based approaches leveraging on experience have been used too. These are based on a growing set of business rules, to warn about suspect claims. To avoid same tooth extraction more than once!
Statistical approaches that compare claim values against historical data of similar claims have also had some success. Here, outliers or claims exceeding some predefined variances are flagged. Poor tuning however leads to several false positives and false negatives being encountered. Pattern mapping is tipped to be the future. Using analytics, clustering and data mining to unravel suspect patterns and compare a given case against such patterns. This has however not been well researched on and few implementations exist.
In this thesis the researchers determine the types of fraud that are currently encountered and present a framework to curtail fraud using information systems that investigate and / or detect incidences of fraud. Fraud inhibits a carrier’s ability to charge lower market-leading premiums and has a detrimental impact on loss ratios. Using a combination of business rules, social networking analysis, predictive modeling and other techniques, it is possible to detect and prevent fraudulent claims before they are paid.
1.2 Problem Statement
Insurance fraud has increased following fraudsters’ improved schemes to avoid detection hence the need for an integrated framework to curtail fraud. Information systems used in Insurance have been used mainly for administrative support, product development, new business processing, product distribution and client service. They have contributed very little to the fight against fraud in the industry at best.
A study carried out by Neirotti and Paolucci (2007) reveals that technological and business path dependencies, along with time compression diseconomies, resulted in diversities in IT adoption dynamics due to their differences in IT governance and management practice. The study also showed that competitive advantages were not correlated with IT spending levels nor with the kind of IT investments that made general productivity growth in the industry possible (Neirotti and Paolucci, 2007)
In this thesis the researchers study the use of information technology systems in the insurance industry. The researchers then propose a framework that incorporates fraud investigation in the information systems. The thesis recommends a framework that introduces fraud detection capability in insurance information systems to investigate potentially fraudulent claims.
1.3 Research Objectives
The purpose of this research was to determine and analyze the perennial problem of insurance fraud and develop a framework to curtail it. The framework introduces fraud detection capability in insurance information systems to identify potentially fraudulent claims.
Specifically, the research objectives were:-
i) To identify the existing forms of insurance fraud in the Kenyan insurance industry
ii) To find out what models exist for detecting Motor insurance fraud in the global insurance industry
iii) To propose a framework by which Information Systems can be employed to curtail Motor insurance fraud in the Kenyan insurance Industry
iv) To validate the framework
1.4 Research Questions
To achieve this, the researchers sought to answer the following questions:-
i) What forms of insurance fraud have been detected in the insurance industry? ii) What models exist for detecting Motor insurance fraud in the insurance
industry?
iii) Based on what structure can new and existing Information Systems be applied to curtail Motor insurance fraud in the Kenyan insurance industry?
iv) How does this new structure meet the desired goal?
1.5 Justification
Clearly both the local and global insurance industries have been hit by massive incidences of Motor insurance fraud. Owing to the material nature of the losses encountered due to insurance fraud, the industry is greatly concerned with the detection of fraudulent behavior.
A lot has been written about how to detect fraud. However many authors state that prevention should take precedence over detection (Jans et al., 2010). Statistics from the US Insurance Fraud Bureau (IFB) show that although fraud is detected, prosecution success is very little. There is therefore need to prevent rather than detect fraud after the fact.
Table 1.1 - 1989 Bodily Injury Liability Claim Sample
1989 Bodily Injury Liability Claim Sample
"Fraud Definition" Approx Claim Count (%)
1. Apparent Fraud or Build-up 43.80%
2. Apparent Fraud Only 9.10%
3. Apparent Fraud Referable for Criminal Investigation 1.00%
4. IFB Referrals Qualifying for Active Investigation 0.50%
5. IFB Investigations Referable to Prosecution 0.10%
6. Prosecution Successes 0.09%
Source: AIB Studies of 1989 BI Claims; RAD estimates of IFB Data
The table shows that it is almost impossible to recover cash paid out through fraudulent claims, even when litigation is used. Predictive approaches to fraud detection therefore provide a more realistic and cost-effective avenue to curtailing fraud.
Carriers use adjusters to routinely investigate claims and negotiate settlements based on the adjuster’s opinion of what the actual loss is. The adjuster’s skill is therefore refined
with experience posing the problem of older experienced adjusters who are nearing retirement. They are taking with them the experience it takes to identify fraud hence the need for preservation of this capacity within information systems
Certainly, there was a business case to explore the opportunity to use Information Technology to curtail the rising number of incidences of insurance fraud.
1.6 Scope and Limitations
The scope of our research was motor insurance claims fraud. Raw data collected from published studies was reviewed as part of the study and cited. Data from the Association of Kenyan Insurers, the Insurance Regulatory Authority and five insurance companies was collected and reviewed as part of the study too.
The study was limited only to the output that the respondents provided. This includes data and information from questionnaires administered. In addition, information from various items of literature was gathered and analyzed alongside questionnaire results. All this was in the area of motor vehicle Insurance claims. The researchers did not study claims in other lines of insurance business and alternative investment / savings products such as unit trusts and pensions
1.7 Theoretical Frameworks
The Bayesian Inference Approach and the Mass Detection Tool (MDT) are used to provide theoretical basis to this thesis. The Bayesian Inference approach falls under the predictive modeling category of frameworks. It uses a Bayesian learning algorithm to predict occurrence of fraud. The MDT on the other hand is an experience based approach that uses a growing set of business rules, to warn about suspect claims. This thesis uses the MDT as a basis for the proposed framework.
CHAPTER 2
2.0. LITERATURE REVIEW
2.1 Introduction
Insurance fraud is indeed a growing problem for the global insurance industry. The Association of British Insurers estimate that fraud from personal lines cost the UK industry the equivalent of Kshs. 119 billion in the year 2001 alone. Insurance fraud is difficult to deal with, partly because it has many varieties, for instance, bogus or inflated claims, staged incidents and systematic ‘scams’. Additionally, protective measures delay payment of genuine claims, impacting negatively against sales volume and company reputation (Ormerod, Morley, Ball, Langley and Spenser, 2003).
The American Insurance industry studies indicate that 10 per cent or more of property/casualty insurance claims are fraudulent. Fraud is the second most costly white-collar crime in America behind tax evasion (Jans et al., 2010)
Figure 2.1 Increase in questionable claims in the US (Jans et al., 2010). 0 500 1000 1500 2000 2500 3000 3500 4000 4500
Tampa Miami Orlando New York Los
Angeles
2008 2009 2010
Further statistics from The United States National Insurance Crime Bureau (NICB) showed a steady increase in questionable vehicle claims from January 2009 through December 2011. There were 84,407 questionable claims referred to the NICB from its member insurance companies in 2009. In 2010, that number increased to 91,797. In 2011, that number increased again to 100,450, a record level. This represents a 9.4 percent increase from 2010 to 2011. Over the two year timeframe from 2009 to 2011 there was a 19 percent increase. In the vehicle category, “questionable vehicle theft” logged the most referrals in 2011. This segment had 11,451 referrals in 2011 after posting 2,182 referrals in 2010, a 450 percent increase from 2009 (NICB, 2012).
According to the Association of British Insurers (ABI), undetected general insurance claims fraud totaled around £1.9 billion pounds a year. Fraudulent claims now account for a significant portion of all claims received by insurers.
The value of detected general insurance claims fraud reached £730m in 2008 having steadily increased from 2004 (Association of British Insurers, 2009).
Figure 2.2 Detected General Insurance Claims Fraud (Association of British Insurers, 2009) 0 100 200 300 400 500 600 700 800 2004 2005 2006 2007 2008 Fraud savings (£ m)
The most common and costly form of general insurance claims fraud is opportunistic retail fraud. Opportunistic retail fraud is where individuals exaggerate or inflate genuine claims to increase the value of a payout. In a minority of cases opportunistic fraudsters would fabricate an entire claim, including, for example, deliberately causing damage so as to be able to claim. Opportunistic fraud in commercial general insurance is similar to opportunistic retail fraud but the policyholders are firms, rather than individuals (“Association of British Insurers”, 2009)
Historically Insurance has been a leading industry in the utilization of Information technology (Harris and Katz, 1991). The focus in these early years was primarily on administrative support, product development, new business processing, product distribution and client servicing. As a result, today many banks and insurance companies still depend on systems first developed over 30 years ago (Ward and Peppard, 2002, p.1) which have very little if any focus on security. These systems are almost useless in the fight against insurance fraud.
Insurance claim fraud takes the form of nonexistent claims, adding items to a genuine claim or inflating the estimated loss value. Whichever the form, insurance fraud drains insurers.
Carriers are well aware that a claimant has numerous opportunities and financial incentives to take advantage of accidents. The claimant knows exactly what happened whereas the company only has some information about what happened. Claims thus need to be reviewed and adjusted before payment, which is done at a cost. There is therefore need to sort incoming claims efficiently into categories that require the acquisition of additional information at a cost. This is known as costly state verification. For many insurers, using costly audits and explicit contracts to combat fraud has been the predominant approach (Bond and Crocker, 1997). This has led to fraud
investigations being handled in a reactive manner. The only proactive initiative in this case is dependent on the experience or intuition of the person processing the claim.
2.2 Insurance Fraud in the Kenyan Industry
Insurance fraud is perhaps more of a nightmare for insurance companies in Kenya and indeed Africa as would be in most emerging economies. As carriers in this market leverage more on electronic distribution, volumes grow and with it fraud cases increase as do other related challenges, including data and transactions security and real time processing. Micro-insurance, a type of formal insurance mechanism that protects low-income people against specific perils, had also led to drastic increase in volumes. This is so due to the high number of low-income people in the agricultural sector which has, for many years, formed the backbone of Kenya's economy.
The Kenyan insurance industry has recently become concerned with the detection of fraudulent behavior. Of particular concern is fraudulent behavior in the Motor and Health lines of business. Statistics from the Association of Kenyan Insurers’ Insurance Industry Annual report 2009 reveal that loss ratios have been on the rise. Private and commercial motor claims ranked the highest with the loss ratio related to Private Motor insurance at 80.8%. A Medical Insurance underwriting loss of Kshs. 236M was registered in 2009 as compared to a profit of Kshs. 33.1M in the year 2008 (Association of Kenya Insurers, 2010). The industry estimates that 35% of all non-life claims are fraudulent.
Data from the Association of Kenya Insurers (AKI) shows that Kshs. 20 billion was collected from both private and commercial motor vehicles in the year 2010 and Kshs. 13 billion was paid out as claims. According to the AKI Annual report for 2010, the health insurance segment made the highest loss ratio, a massive 81.5 per cent, followed by motor private insurance at 74.9 per cent (Association of Kenya Insurers, 2011).
Table 2.1 General Business Incurred Claims Trend (Association of Kenya Insurers, 2010).
Class of Business Years
2005 2006 2007 2008 2009 Aviation 9,020 1,417 10,065 2,737 -4,780 Engineering 86.446 98,767 81,340 147,220 184,396 Fire - Domestic 107,195 111,847 115,465 167,815 171,407 Fire – Industrial 148,535 265,889 446,366 462,920 517,239 Liability 147,613 151,475 109,400 228,390 229,877 Marine 229,775 262,774 373,134 428,137 481,942 Motor – Private 3,204,388 3,286,171 3,565,915 4,502,851 5,282,589 Motor – Commercial 3,002,312 3,634,622 4,032755 4,875,612 6,317,808 Personal Accident 1,879,278 2,769,091 3,232,202 3,490,256 4,604,216 Theft 366,789 487,050 512,778 696,488 752,651 Workmen’s compensation 1,090,642 1,182,637 1,542,377 656,465 1,002,722 Miscellaneous 111,830 107,821 213,608 209,437 228,254 TOTAL 10,383,822 12,359,561 14,235,405 15,868,328 19,768,322 Figures in thousands Kshs
A survey by PricewaterhouseCoopers (2011) on risk in East Africa's financial services sector had identified fraudulent claims as one of the major risks facing insurance firms in the region. Carriers estimated that they lost a total of Sh. 4 billion, paid every year to undeserving parties. Motor insurance is the worst hit industry segment and a bulk of the money lost in fraudulent claims by insurers is through rampant fraud in motor insurance. In insurance markets where the industry is still relatively immature, like East Africa, an increasing incidence of fraud will test the capacity of insurance claims settlement procedures and consumer protection laws (“PwC Risk Survey 2011”, 2011).
2.3 Motor Insurance Fraud in the Kenyan Industry
The Kenyan Medical and Motor insurance business segments, the mainstay of the industry, are the worst hit with fraud with medical underwriters making loses as a result of increasing fraud. Fraud in the Insurance Industry is probably as old as the industry itself and, it is a worldwide problem. Motor Insurance remains one of the classes of insurance where fraud is rampant. Statistics point to fraud accounting for as much as between 30 to 40 percent of all motor claims paid by the insurers in Kenya.
Year 2009 statistics by Association of Kenya Insurers (AKI) indicate that the total claims paid by the industry under the motor class of business were up to Kshs.10 billion. If the 30 to 40 percent estimate was to be confirmed it would mean that the industry paid as much as Kshs. 3 to 4 billion by way of fraudulent claims (Gichuhi, 2011). Most of the fraudulent claims in the industry are from public service vehicles and involved inflation of passenger numbers and exaggeration of injuries that in turn saw carriers pay hefty amounts.
Following increase of insurance related fraud, the Insurance Regulatory Authority (IRA) introduced reforms in the industry. These included a standard contract for insurance across different insurance companies and the establishment of an Insurance Anti-Fraud Police Unit in October 2011. The IRA is the industry regulator tasked with the mandate of regulating, supervising and developing the insurance industry in Kenya. It therefore creates an appropriate legal framework to ensure efficient and effective supervision of the industry.
According to the IRA Chief Executive Officer, fraud is rife in the insurance industry, especially in the motor vehicle class of insurance business. This led to the collapse of some insurance companies in the market segment. This trend was attributed to the firms’ involvement in the public service vehicles insurance business beset by soaring claims, fraud and litigation (Oyuke, 2012).
Table 2.2 Incurred Claims Ratio per Class of Business (Association of Kenya Insurers, 2010).
Class of Business Years
2005 2006 2007 2008 2009 Aviation 33.3 11.5 51.6 41.6 -48.8 Engineering 45.1 52.1 29.5 55.0 52.8 Fire - Domestic 28.1 26.9 26.3 35.4 30.6 Fire – Industrial 20.2 33.8 56.4 52.8 53.7 Liability 51.4 49.2 27.5 47.4 45.8 Marine 34.4 38.5 46.8 51.2 51.6 Motor – Private 83.1 75.8 75.8 83.2 82.4 Motor – Commercial 49.3 51.1 54.0 57.8 63.6 Personal Accident 62.5 71.8 68.0 62.0 69.1 Theft 51.1 59.1 51.8 61.1 59.4 Workmen’s compensation 92.5 89.1 106.2 38.0 37.3 Miscellaneous 26.9 24.4 39.5 31.3 29.7
Total Industry Average 59.3 60.9 62.9 61.0 63.7
From Table 2.2, Motor Private, Personal Accident, Motor Commercial and Theft had the highest claims incurred ratios in the year 2009. These classes of business had claims ratios above 55.0% and showed a rising trend over the years. Motor Private came out as the most loss making class of business under general insurance business. Without Motor Private the industry would have made an underwriting profit of over Kshs. 1.7 billion during 2009 but made a paltry Kshs. 401,806.00 instead.
2.4 Current Investigative and Detective Methods
Insurance claim payment is considered a simple multiple step process. It involves claims submission, review, and approval processes that result in a claims being paid or rejected. The process starts when an auto claim form is submitted. Evaluation is then done where the carrier verifies the claim applicant’s information and this is reviewed by the claims adjuster. The adjuster assesses the damage to the vehicle. Information including claim forms, adjuster reports, police reports and photos will determine whether to reject, accept and pay or request more information.
The responsibility for detecting fraudulent claims in insurance companies rests heavily with staff at the front line of the claims handling process. Claims handlers are often inexperienced, with typical company lifetimes of less than one year, and they often lack sufficient or appropriate training in fraud detection. In order to increase the chances of detecting fraudulent claims by inexperienced staff, companies have traditionally provided claims handlers with lists of fraud indicators against which to check incoming claims (Doig, Jones and Wait, 1999).
Carriers are bound by regulations with regard to how long the evaluation process can take as Insurance law requires firms to process claims within 90 days. The evaluation stage is the point at which fraud is either detected or missed. If missed here, the fraud may be detected after payment. According to Bond and Crocker (1997), using costly audits and explicit contracts to combat fraud has been the predominant approach (Bond and Crocker, 1997).
2.4.1 Auditing and the Optimal Contract
In a non-commitment Costly State Verification setting, insurers can only detect fraudulent claims by performing costly audits, and policyholders are overcompensated by the optimal insurance contract (Schiller, 2006). Since this is a costly approach, a claim adjusting process is involved.
The claim adjusting process is in theory a narrowing of the information asymmetry that exists for every claim. The claimant knows exactly what happened whereas the carrier only has bits of information about what happened. The carrier needs to ascertain the information’s accuracy and determine the appropriate payment to be made or if the claim is to be denied. Carriers have the discretion to spend as little as possible on a claim or invest in acquiring information to resolve the asymmetry (Derrig, 2002).
2.4.2 Co-opetition and Common Databases
Where historical information is available, a common database from all insurers containing the details of all the declined risks and reasons behind the same can suffice. This, coupled with an internal database that maintains details on all frauds which have happened over the years, could go a long way in isolating claims for further analysis. Since all the insurers in the particular market are involved, the cost is shared or borne by the regulator.
This approach makes not just detection easy but improves underwriting capabilities by flagging potentially fraudulent business before signing the contract. Experience however is costly to obtain as it involves learning from mistakes, which are costly in the first place.
2.4.3 Rule Based Approaches
Rule based approaches leveraging on experience have been used too. These are based on a growing set of business rules, to warn about suspect claims. To avoid same tooth extraction more than once!
There are two main rule based approaches; clustering and expectations. In the clustering approach, normal patterns are grouped together to form a cluster and any deviations from the norm flag suspicious cases. These are then further investigated for fraud. They indicate outliers only and not necessarily fraud cases.
On the other hand, the expectations approach focuses on expected values, for instance the expected value of a windshield claim, and compares it with the actual value. Large deviations are suspicious. This approach requires a predictive model that generates the expectations. Here too, outliers or claims exceeding some predefined variances are flagged.
With rule based approaches, poor tuning leads to several false positives and false negatives being encountered.
2.4.4 Government Regulation
Insurance companies, through the Insurance Regulatory Authority (IRA), have taken further proactive steps to improve fraud detection during the claims-handling process. For example, the industry has developed a database that is shared across all motor business carriers to assist in the detection of anomalous information at the claims stage. The database however presents a problem for the detection of fraud. In particular, the quality of data held within the database is not verified.
Since the data is gathered by different carriers based on information supplied by the client, there is always the risk that data entry will be done independently and repetitively for each transaction that the same customer has with a company. This introduces noise and diminishes the perceived gains.
2.4.5 Predictive Analytics
Predictive Modeling and Predictive Analytics have been buzzing around the claims industry for years. This approach constructs a predictive model that predicts the probability of fraud. Such a model attempts to differentiate fraud from non-fraud cases and hence requires data from both categories to facilitate learning.
The return on investment (ROI) for predictive analytics is pretty easy to calculate. It is simply the percentage uptick in number of claims flagged and percentage of those claims mitigated, denied, or reduced. False positives are always an issue here too and create the need to consistently tweak parameters in analytics. Using analytics, clustering and data mining to unravel suspect patterns and compare a given case against such patterns has not been well researched on and few implementations exist.
2.4.6 Ethnography
Ethnography provides perhaps the best way to profile fraud based on a particular market. Ethnography involves the immersive study of work practices in realistic contexts, in which the observer works within the system under study for extended periods of time, observing and documenting everyday activities as well as exceptional events (Ormerod, Ball and Morley, 2010).
Employing an ethnographic approach complements the statistical, interview and questionnaire methods used in studies of fraud types and fraudster profiling by providing a detailed, longitudinal and independent evaluation of issues and activities surrounding how the industry deals with fraud (Ormerod et al., 2003).
2.5 Fraud Detection Frameworks and Models
Different frameworks and models have been developed to try and curb the issue of motor insurance fraud. These can mainly be classified into predictive and reactive. Predictive models detect and deter fraud before it occurs while reactive ones detect fraud after the fact.
Generally, fraud detection can be done using one of three approaches; Clustering, expectations or predictive modeling. The clustering and, expectations approaches highlight suspicious cases for further fraud investigation while the predictive modeling approach directly predicts the probability of fraud. The researchers analyzed two models, The Mass Detection Tool (MDT) for early detection of insurance fraud (Ormerod et al., 2003) and the Bayesian Inference approach. The MDT is a clustering approach while the Bayesian Inference approach falls under the predictive modeling category.
2.5.1 The Mass Detection Tool for early detection of insurance fraud
The Mass Detection Tool (MDT) for early detection of insurance fraud is an experience based approach that uses a growing set of business rules, to warn about suspect claims. The aim of the MDT was to act as a filter for all claims, providing the claims handler with the confidence to pay genuine claims quickly while selecting out suspicious claims for further investigation.
The core to the MDT is a Bayesian Belief Network (BBN) that calculates the probability of the current claim being fraudulent, based upon the prior probabilities of claims possessing the same range of fraud indicator ratings being proven as fraudulent. The calculations are based upon a store of prior probabilities of frauds given indicator ratings that is assembled from feedback of previous claim outcomes (pay, refer or refuse). So, indicators become more or less predictive of fraudulent versus genuine claims over time.
The MDT is non-coercive in that although it recommends whether a claim should be paid or referred, it leaves that decision to the user. This is important since the MDT’s capacity to learn is partly driven by the user overriding its advice and partly by the final decision taken downstream by the fraud investigators.
2.5.2 Bayesian Inference Approach
Here, a Bayesian learning algorithm to predict occurrence of fraud. In the Bayesian Inference approach, several state-of-the-art binary classification techniques are experimentally evaluated in the context of expert automobile insurance claim fraud detection (Viaene, Derrig and Dedene, 2005). Two Bayesian networks are created to describe the behavior of auto insurance.
First, a Bayesian network is constructed to model behavior under the assumption that the driver is fraudulent and another model under the assumption the driver is legal. The fraud net is set up by using expert knowledge. The legal net is set up by using data
from legal drivers. By inserting evidence in these networks, we can get the probability of the measurement E under the two mentioned assumptions. This means, we obtain judgments to what degree an observed user behavior meets typical fraudulent or legal behavior.
In general and by applying Bayes’ rule, we get the probability of fraud from inference based on a multiplicity of factors. These include driver age, driver rating, vehicle age / price, claim value and number of previous claims.
2.6 Conclusion
Growth of the insurance industry has led to the expanding application demand for data mining of massive data warehouses (Hong and Weiss, 2001). Fraud detection is one of the areas that have fueled advances in automated predictive methods. Specifically, pattern mapping and the use of analytics, clustering and data mining have been used to detect insurance fraud. This approach is used to unravel suspect patterns and compare claims cases against such patterns.
Predictive analytics can enhance the work of investigators by uncovering complexities the human eye may miss (Roosevelt, 2011)
The problem of insurance fraud is prevalent and threatens the very existence of insurance companies. Several efforts have been made to institute proactive solutions for tackling the major problem of insurance fraud. This problem however needs to be tackled using a multi-disciplinary approach ensemble as the individual documented approaches tackle only individual facets.
An Information technology framework can supplement traditional specialized investigation units, statistical analysis of claims information, mathematical models, costly audits and explicit insurance contracts in the investigation and detection of fraudulent cases by integrating the approaches into a framework. The framework can
also be used to capture expert best practice while observing pitfalls that prevent the successful detection of fraud.
CHAPTER 3
3.0. RESEARCH METHODOLOGY
3.1 Introduction
Research design involves linking of research questions to empirical data so as to be able to come up with tangible research conclusions. It sets out the logic to the enquiry. Several key components comprise a practical research design. These include the research questions, the study's proposition, the study's units of analysis, the logic that links data to propositions and the criteria for interpretation of the research findings. The choice of a research methodology is thus dictated upon by the research questions and purpose. This research intends to propose a framework intended to provide a solution for preventing, detecting and managing claims fraud in the motor insurance line of business within the Kenyan insurance industry.
The research questions raised in this thesis required a combination of various techniques to be put in use so as to answer the questions. The first question sought to establish what forms of insurance fraud have been detected in the Kenyan insurance industry. This is available from reports and media as well as subject matter experts in the industry. Article reviews and questionnaires were used to gather this information. The next question dealt with the approaches to fraud detection in the industry. This gives a view of how the carriers currently handled the problem of insurance fraud. This was elicited from insurance and ICT experts in the insurance sub-sector. This question also sought to address the issue of existing theoretical frameworks and models for detecting Motor insurance fraud in the global insurance industry. Here the aim was to find theory supporting a comprehensive view of motor insurance fraud detection, prevention and management. These were available in academic published literature.
Two available theoretical frameworks were identified and compared to comprehensive motor insurance claims fraud management with respect to the insurance industry in Kenya. This led to the establishment of a framework that was best suited for the insurance industry in Kenya. A review of what extensions were needed and what the theoretical frameworks overlooked led to the proposed framework thereby answering the third question. The researchers then tested the framework for validity of results to answer the fourth and final research question.
3.2 Study Population
For purposes of this research, stratified random sampling of industry experts was used. The insurance industry had 46 insurance companies registered to transact insurance business in Kenya in 2009. Of these, 35 were licensed to transact in the motor segment.
Table 3.1 Insurers Registered for Each Class of Business (Insurance Regulatory Authority, 2009).
Category Number
Long term business insurers 10
General business insurers 20
Composite insurers 14
Reinsurance companies 2
TOTAL 46
Table 3.2 General Insurance Business Insurers (Insurance Regulatory Authority, 2009). Serial No. Description No Serial No. Description No
1 Aviation 7 2 Engineering 33
3 Fire - Domestic 33 4 Fire – Industrial 33
5 Liability 33 6 Marine 33
7 Motor – Private 33 8 Motor – Commercial 34
9 Personal Accident 34 10 Theft 33
3.3 Sampling Frame
The sampling frame is a list of all those within a population who can be sampled. For purposes of this study, the top fifteen of the thirty five insurance companies licensed to transact in the motor segment were chosen. The criterion used for ranking the insurance companies was total gross written premium for the two motor insurance classes of business in the year 2010. The list of insurance companies in order of performance for year 2010 is attached in Appendix B. The top fifteen companies accounted for seventy two percent of the underwritten premium in 2010 and thus formed the sampling frame.
3.4 Sample Size and Error
The sample size (
n
0) used in this research was twenty respondents. These wereemployees drawn from five randomly selected insurance companies in the sampling frame. The respondents were employees in internal audit and claims processing sections of the insurance companies.
No survey can ever be deemed to be free from error or provide 100% surety. Error limits of less than 10% and confidence levels of higher than 90% can be regarded as acceptable (Hussey and Hussey, 1997, pp. 226). The confidence interval (z) corresponding to a 90% confidence level is 1.645. Assuming that the 15 insurance companies in the sampling frame have on average a total of sixteen staff doing claims handling and audit, the population (N) = 15 * 16 = 240. p is the number of responses (
n
0) expressed as a percentage of the population (N). This is calculated as (n
0/N)*100 =8.3%. The error rate is therefore calculated using the formula e² = z²p(1-p) - z²p(1-p) (Yamane, 1967, pp. 258).
n
0 N3.5 Research Design
The research was applied, analytical and cross-sectional and was developed from a qualitative point of view. Applied research plans to solve practical problems of the modern world, rather than to acquire knowledge for knowledge's sake. It is geared towards improving the human condition. In our case the problem was the prevalence of motor insurance fraud in the Kenyan Insurance industry. Analytical research attempts to establish how it came to be. This is a necessary step in the finding of solutions to it. This research took the Mass Detection Tool (MDT) as input to evaluate the practicality of a fraud investigative and detective network. The strengths and weaknesses of the MDT were evaluated and used to propose a framework that could effectively be used to address motor vehicle insurance fraud in Kenya.
Ethnographic studies of insurance fraud detection were incorporated to improve the MDT. These ultimately resulted in a framework by which insurance fraud in the Kenyan insurance industry can be curtailed through timely investigation and detection. A meta-analysis research design was used. The findings from similar studies documented in selected journal articles and books were integrated with findings from the Kenyan Insurance Industry. Being a correlational study, a meta-analysis design was chosen to allow for generalizations across studies and reveal useful patterns in the combined study data.
3.6 Data Collection Methods and Techniques
Primary and secondary data sources were used for the purposes of this research. The primary data was collected through questionnaires administered to subject matter experts in the industry. These were drawn from five of the top fifteen insurance companies in gross written motor insurance premium for year 2010. Secondary data was collected from the Association of Kenya Insurers, the Insurance Regulatory Authority, the Association of British Insurers and the US Insurance Fraud Bureau.
Responses from professionals in the industry were analyzed alongside raw data collected from published studies. This was done to compare the effect of fraud detection capability in information systems on the actual number of fraud cases reported.
3.6.1 Questionnaires
Questionnaires were chosen as the best instrument for this study because they provide an economical and convenient approach to data collection. This was necessary due to the time limitations. Questionnaires also provide anonymity of the respondents which was necessary since the subject matter is a sensitive topic. Leaked information could have serious ramifications. Additionally, since the responses needed to be gathered in a standardized way, questionnaires were more objective.
Another factor that informed the choice of questionnaires is their ease of administration. They were sent online and responses received online which ensures there was no need to visit all respondents physically. Furthermore, the results of the questionnaires could be quickly and easily quantified by the researcher during analysis.
3.6.2 Secondary Sources
Several secondary sources were consulted during the data collection process. Reports from the IRA, AKI, professional audit firms and international regulators and associations provided the basic secondary data.
3.7 Data Analysis
A standard questionnaire was administered to respondents. Descriptive and inferential statistics were used to analyze data from the questionnaire survey. For questionnaire data, analysis began after the first few responses. The first four questionnaires were for purposes of piloting and determined the relevant and irrelevant areas. This provided feedback for refinement before it was administered to the 20 respondents
3.8 Framework Validation
Elements of fraud detection from both published materials and the questionnaires were amalgamated to form a holistic framework for fraud detection. Descriptive and inferential statistics were then used to determine the importance of each element in the new framework. This was done by making inferences based on the responses. Since the framework is to be used at operational level, additional information on fraud indicators and actual data used for fraud detection was incorporated. This resulted in a procedure that represents the actual framework in practice / execution. A significant reduction in successful fraudulent payouts is therefore expected when the framework is used. This validated the framework with a statistical backing
.
CHAPTER 4
4.0. THEORETICAL MODELS FOR FRAUD DETECTION
4.1 Introduction
Schiller (2006) states that although some papers have analyzed fraud detection models, few like the paper of Dionne, Guillen and Picard (2003) have taken information systems into account (Schiller, 2006). However, subsequent research has seen the development of information system based models and tools for insurance fraud investigation and detection. Two such models are used to provide a theoretical basis to this research. The Bayesian Inference Approach and the Mass Detection Tool for early detection of insurance fraud (MDT)
This research then took one of the tools, the MDT, as input to evaluate the practicality of an information system based fraud investigative and detective network. The MDT is a suite of computer-based tools to support the early detection and subsequent investigation of potentially fraudulent claims.
4.1.1 Bayesian Inference Approach
Here, a Bayesian learning algorithm to predict occurrence of fraud. In the Bayesian Inference approach, several state-of-the-art binary classification techniques are experimentally evaluated in the context of expert automobile insurance claim fraud detection (Viaene et al., 2005). Two Bayesian networks are created to describe the behavior of auto insurance. First, a Bayesian network is constructed to model behavior under the assumption that the driver is fraudulent and another model under the assumption the driver is a legal. The fraud net is set up by using expert knowledge. The legal net is set up by using data from legal drivers.
Mathematically, Bayes' theorem gives the relationship between the probabilities of A and B, P(A) and P(B), and the conditional probabilities of A given B and B given A, P(A|B) and P(B|A). In its most common form, it is:
1. P(A|B) = P(B|A) P(A) P(B)
By inserting evidence in these networks, we can get the probability of the measurement E under the two mentioned assumptions. This means, we obtain judgments to what degree an observed user claim meets typical fraudulent or legal behavior. These quantities we call
P(E| output = legal) P(E| output = fraud).
By postulating the probability of fraud P(output = fraud) + P(output = legal) = 1 Therefore,
2. P(output = legal) = 1 - P(output = fraud )
In general and by applying Bayes’ rule, we get the probability of fraud from inference based on a multiplicity of factors. These include driver age, driver rating, vehicle age / price, claim value and number of previous claims.
4.1.2 The Mass Detection Tool
The Mass Detection Tool (MDT) for early detection of insurance fraud is an experience based approach that uses a growing set of business rules, to warn about suspect claims. The aim of the MDT was to act as a filter for all claims, providing the claims handler with the confidence to pay genuine claims quickly while selecting out suspicious claims for further investigation.
The MDT steps the user through a question-and-answer dialogue to elicit ratings of a claim against known fraud indicators. Some indicators are supplied automatically from claims information (e.g., the car registration date) while others are informed by the claims handler’s opinion (e.g. the attitude of the insured on the phone). As the indicators are rated, the system provides real-time feedback as to the probability that the current claim is fraudulent, and the probability that the claim is associated with a specific type of fraud (Ormerod et al., 2010)
Figure 4.1 The Mass Detection Tool (Ormerod et al., 2010)
The core to the MDT is a Bayesian Belief Network (BBN) that calculates the probability of the current claim being fraudulent. This is based upon the prior probabilities of
claims possessing the same range of fraud indicator ratings being proven as fraudulent. The calculations are based upon a store of prior probabilities of frauds given indicator ratings that is assembled from feedback of previous claim outcomes (pay, refer or refuse). So, indicators become more or less predictive of fraudulent versus genuine claims over time. The model includes a Suspicion building tool which is designed to merge automated learning capabilities with the MDT and with expert judgment in a coherent fashion.
The MDT is non-coercive in that although it recommends whether a claim should be paid or referred, it leaves that decision to the user. This is important since the MDT’s capacity to learn is partly driven by the user overriding its advice and partly by the final decision taken downstream by the fraud investigators.
4.2 A Summary of the two Models
The two models were developed based on similar premises and represent how information systems can be used to investigate and detect insurance fraud. The primary similarity is the use of Bayes’ rule. Some key differences however present themselves based on the way the two approaches view the problem of insurance fraud.
The Bayesian Inference approach lends itself to artificial intelligence with the premise that insurance fraud is primarily a modeling problem. It therefore models what is construed to be fraudulent behavior as well as what is deemed legal behavior of the driver based on the claims data. The result of this is a Boolean, fraud or legal.
The MDT views the problem of insurance fraud as a behavioral problem. Having been developed from ethnographic studies, the MDT uses experience and a series of fuzzy logic steps to detect fraud. At each step, responses are used to elicit ratings for a claim against known fraud indicators. These come from both the claims data and the claimant's behavior. It thus results in a probability and claims with probabilities exceeding a particular threshold are earmarked for further investigation
4.3 Conclusion
The literature review laid an academic foundation for this research in order to contribute to an understanding of how the claim processing system could be enhanced using technology. The results of the review also identified patterns, distinctions, and gaps in the literature. The literature review recognized a shortage of models and frameworks that address the problem of insurance fraud using information systems. This is surprising, because these systems are very popular and effective in existing insurance markets, with considerable size, like the auto insurance market (Schiller, 2006).
4.4 The Conceptual Framework
The conceptual framework is an intermediate theory that attempts to connect to all aspects of the inquiry. Figure 4.2 shows the conceptual framework for insurance fraud detection and investigation.
Figure 4.2 Conceptual framework
The goal is not to replace adjusters and fraud investigation personnel but to support their function by sharpening the information delivery system using more appropriate data and better technology to manipulate and deliver the data.
CHAPTER 5
5.0. ANALYSIS AND PRESENTATION OF FINDINGS
5.1 Introduction
The data collected through the administration of questionnaires is explained in this chapter. Quantitative data is presented in form of tables, charts and graphs with a brief description provided. A summary of the major findings of the study is then provided at the end of the chapter.
5.1.1 Respondents’ Demographics
The questionnaire was administered to a total of twenty staff. These were drawn from across five insurance companies. Of the respondents, eleven were auditors and nine were claims processing staff.
Details of the number of years that the respondents had worked in the insurance industry are summarized in Figure 5.1.
Figure 5.1 Insurance Industry Experience of Respondents 0 1 2 3 4 5 6 7 8 9 10
< 1 1 to 5 6 to 10 11 to 15 16 to 20 > 20
Number
of
Respondents
Insurance Experience in Years
Insurance Experience
Three of the respondents had less than a year’s experience, nine had between one and five years, five had between six to ten years experience, two had between eleven and fifteen years experience and one had between sixteen and twenty years experience. None of the respondents had over 20 years experience.
The responsibility for detecting fraudulent claims in insurance companies rests heavily with staff at the front line of the claims-handling process and internal audit. The modal age bracket for this group is therefore between one to five years. From the trend line we infer that most have less than five years experience and move from this area as they gather experience. This leaves inexperienced staff with typical company lifetimes of less than five years principally responsible for fraud detection.
5.2 Existing Forms of Insurance Fraud
Respondents were asked to identify three most prevalent forms of insurance fraud. The findings are summarized in figure 5.2.
Figure 5.2 Forms of Insurance Fraud and Their Prevalence Motor Claims, 13 Premium embezzlement, 10 Medical Claims, 11 IT Related Forms, 4 Cheque fraud, 10 Claims in other business lines, 9
Premium embezzlement, Cheque Fraud, motor claims fraud, medical claims fraud and fraudulent claims in other lines of business were the responses. Thirteen respondents had Motor claims fraud while eleven had medical claims fraud. Premium embezzlement and cheque fraud had ten responses each while none mentioned life insurance. All respondents’ companies had incurred financial costs as a result of incidents of motor claims fraud.
The researchers sought to compare this with statistics documented in literature and noted that types of insurance fraud are varied and occur in all areas of insurance. Insurance crimes also range in severity, from mildly exaggerating claims to deliberately causing accidents or damage. Insurance fraud poses a very significant problem and organizations regularly employ the services of private detectives to combat false claims. Common examples of claims fraud include;
5.2.1 Motor insurance fraud
One of the largest categories of insurance claims fraud revolves around the insurance of vehicles (Levi, 2011). This is Motor insurance fraud and includes staged collisions and exaggerated claims.
Staged collisions are a rapidly growing fraud form that involves professional fraudsters setting up a collision where a vehicle is used to stage an accident involving a random innocent party. The most widely used technique involves driving in front of the chosen victim by the scammers. They will then suddenly brake heavily or make an unexpected maneuver causing the innocent party to collide with their vehicle. Damage is claimed against the victims insurance and in addition each of the fraudsters involved may claim for injuries sustained in the motor accident, potentially even working with an accomplice who is a doctor.
Exaggerated claims are a very common too. Here, a fraudulent claim is lodged when a real accident has occurred. The dishonest owner may take the opportunity to include a
range of previous minor damages to the vehicle in the garage bill. This is then claimed in association with the real accident and/or claim for a false minor injury.
5.2.2 Health Insurance Fraud
Health insurance fraud is the act of deceiving, concealing, or misrepresenting information that results in health care benefits being paid to an individual or group. This fraud can be committed by both a member and a provider.
Member fraud can include making false injury claims, non-disclosure of pre-existing conditions, ineligible members or dependents and concealing claims that were a result of a work related injury.
Provider fraud can consist of billing for a service not rendered, alterations to claims referrals, overcharging for services and diagnosis or treatments outside the scope of the agreement.
5.2.3 Property Insurance Fraud
Possible motivations for this type of fraud can include obtaining payment that is worth more than the value of the contents stolen or destroyed or property damaged. A number of property insurance claims involve arson. The main reason for this is that the evidence of the fire being started by arson is often destroyed by the fire itself.
5.2.4 Life Insurance Fraud
Life insurance fraud includes faked deaths so that family members can make claims on policies. It could also take the form of creating a false identity and faking the death of the false person to claim on a policy.
5.2.5 Conclusion
Measuring insurance fraud is not a straightforward exercise as different forms of fraud exist. The different stakeholders gather fraud data related to their own missions thus
categorizing it depending on those missions. It is therefore not possible to find a single agency that gathers comprehensive fraud statistics for all. The kind, quality and volume of data gathered vary widely. Insurance fraud data thus are relatively piecemeal.
From the survey, thirteen of the twenty respondents had motor claims fraud as one of the most prevalent in the Kenyan insurance industry. This is a 65 percent response rate. With the error rate of 9.7% (approx. 10%) calculated earlier, we can conclude that between 55 to 75 percent of staff with the responsibility of detecting fraudulent claims in insurance companies deem motor claims fraud to be one of the most prevalent forms of insurance fraud.
5.3 Existing Methods for Detecting and Preventing Fraudulent Activity
5.3.1 Detection of Fraudulent Activity
Respondents were asked to select all the methods by which fraudulent activity detected by their businesses. They were allowed to select more than one method. All twenty respondents indicated that detection by staff other than internal investigators is used for fraud detection. With the error rate of 10 percent, this gives between 90 to 100 percent positive response on the use of staff other than internal investigators to detect fraud. This makes it the leading fraud detection method and we can infer that all insurance companies use their staff’s experience to detect fraud. The responses are summarized in figure 5.3
Figure 5.3 Methods by Which Fraudulent Activity is Detected
Eighteen respondents had reporting by the public, clients, customers and business associates. Thirteen respondents selected Use of internal investigators and nine selected use of private investigators. Only a paltry three selected use of detection technology for fraud detection. We therefore conclude that only between 5 to 25 percent of staff with the responsibility of detecting fraudulent claims use technology for fraud detection.
5.3.2 Prevention of Fraudulent Activity
Respondents were asked to select all the methods by which fraudulent activity is prevented by their businesses. They were allowed to select more than one method. The responses are summarized in figure 5.4
0 5 10 15 20 25 Reporting by the public, clients, customers, business associates Detection by staff other than internal investigators Use of internal investigators Use of private investigators Use of detection technology Internal / External audit Other, please specify
Figure 5.4 Methods by Which Fraudulent Activity is Prevented
Fifteen respondents chose securing of electronic databases as the measure in place to prevent the occurrence of motor claims fraud. We therefore concluded that between 65 to 85 percent of insurance companies secure their electronic databases to prevent fraud. The researchers also concluded that between 55 to 75 percent of the companies use on-going trend analysis of reported or suspected incidents of fraud. This is inferred from the thirteen positive responses. These two methods lend themselves to use of technology yet they that has not been adequately explored. Ten had background checks on clients using available databases.
Eight of the twenty respondents felt that their companies did not have sufficient mechanisms in place to detect and prevent motor insurance claims fraud. This represents between 30 to 50 percent of the insurance companies. This is very high, considering that statistics point to fraud accounting for as much as between 30 to 40 percent of all motor claims paid by the insurers in Kenya. Only seven respondents agreed that there is sufficient government help in the fight against insurance fraud. This is below 50 percent of the insurance companies.
0 2 4 6 8 10 12 14 16 Training and awareness programs among employees Public stance on intolerance for fraud Securing of electronic databases Use of public information on fraud scams, fraud prevention, etc. Background checks on clients using available databases On‐going trend analysis of reported or suspected incidents of fraud Other, please specify