Fraud Management in Auto Insurance
A Thought Paper
Introduction
F raud, in its basic form, would involve Approximately 90 percent of the costs of insurance
misrepresentation / suppression of material facts fraud are the result of what is called “claims
which would ‘alter’ the decision of an insurance padding” – adding damage, injuries, and fictitious
company. Insurance fraud can be either “hard” or passengers to insurance claims. According to a
“soft”. Hard fraud occurs when someone survey by the Insurance Research Council, 35
deliberately fabricates claims or fakes an accident / percent of Americans think it’s all right to pad their
event. Soft fraud (also called as opportunistic fraud) insurance claims. The other 10 percent of insurance
occurs when normally honest people inflate fraud comes from organized accident staging.
legitimate claim amounts (“padding”) for life / health Innocent victims (private motorists, truck drivers
insurance, or intentionally understate the number of carrying chickens, etc.) are targeted by organized
miles they drive each year (in case of auto auto-accident rings. These rings make an “accident”
insurance), or list fewer employees or misrepresent happen by setting up an innocent person for a
the activities (in case of business owners / worker’s collision. Not only is this illegal, but tragedy can
compensation) to get lower premium. Some lines of result.
insurance are more vulnerable to fraud than others.
Health care, worker’s compensation, and Most of the insurance companies have adapted to
automobile insurance are more ‘fraud-prone’ as custom-built, rule based technologies to identify and
compared to other lines of insurance. prevent fraudulent claims from happening. However,
lack of a comprehensive centralized database of all
Insurance Fraud can happen at any stage in the the fraudulent activities in the industry and
insurance transaction by different parties – appropriate technologies that are “intelligent
applicants for insurance, policyholders, third-party enough” to prevent fraud is one major issue, which
providers / claimants and / or any other insurance companies need to address to curb the
professionals involved with the policy. Based on the ever-increasing losses.
intent and the information provided by the insured,
an insurance application (for coverage/claims) may This paper focuses on the incidence of fraud at two
be classified either as fraudulent or genuine. If different processes in the auto insurance company
undetected, fraudulent information would result in namely, underwriting and claims settlement and
huge losses. Insurance fraud costs the insurance suggest some probable techniques to mitigate
industry an estimated $30 billion each year and the losses (since elimination of fraud is almost
average American household approximately $300 impossible task!).
each year in extra insurance premiums.
2
Fraud Incidence during Underwriting
During the underwriting of an automobile insurance Systems designed by using user-defined rules and
application, the prospect may resort to fraud to get business logic based algorithms can define multiple
a ‘favorable’ underwriting decision, either by scenarios to assist the underwriters arrive at an
providing “false information” or “incomplete accurate decision by identifying the information
information”. Either way, the insurance company is inaccuracies and ‘flagging’ them as potentially
at a disadvantage since both these acts by the fraudulent. An example of one such flag would be a
prospect would result in improper risk assumption simple business rule which notifies if the client has
by the insurer. Typically the information that is purchased multiple insurance from company ‘A’, but
incompletely provided includes (a) the residential is approaching a different insurer company ‘B’ for
address, (b) personal & family medical history, and auto insurance cover only Such a behavior
(c) the occupational details. Information pertaining suggests that the client has certain undesirable
to (a) previous / concurrent insurance information, history (pertinent to automobile insurance) and is
and (b) the vehicle details like state of registration, confident about Company ‘A’ NOT providing him
the VIN number, the mileage etc., may be the auto insurance cover. He would therefore
incomplete. approach a different insurer, Company ‘B’ and
apply for automobile insurance without disclosing
The illustration below shows that information the adverse information. In such a scenario, a
required to monitor the fraudulent intent is available simple business rule designed to capture and
from both external and internal sources. While most compare all the personal insurance details of the
of the information is available internally or from the client should be able to alert the underwriter of
insurance application, monitoring of fraud also company ‘B’. This rule would be designed to
involves interfacing with external agencies like the interact with a “centralized database” of the
MIB or the MVR for obtaining certain sensitive insurance industry and map the prospects
information. An underwriter has to use his/her due information to the available details. This database
diligence by applying analytical and judgmental would be very similar to the Medical Information
capabilities before concluding on the decision to Bureau (MIB) database often used by the life
either accept or not accept the application for ‘Auto insurance companies. Such rules are very basic,
Insurance’ coverage based on the ‘face-value’ of easy to understand and implement. Since there are
information provided. a plethora of possible combinations for verifying
various information, availability of the relevant data
Insurance companies can, however, identify is very important.
inappropriate information most of the times by using
expert underwriting systems or technology based Predictive analytics based decision tree models are
solutions that result in minimizing the losses. another technology that an underwriter can depend
on to make precise and consistent decision about behavior. These models are created from
the prospects. These models are easily understood predefined data elements where the ‘outcomes’ are
and are based on a complex set of business rules known. Any new information can therefore be ‘run’
that produce a fraud score. This technology using this model to view the probable outcome of
incorporates compiled information from multiple the decision being taken.
sources rather than relying on a simple red-flag
system to provide an insight on future customer
Internal Sources
Personal Information
Vehicle Information
• Age / Gender
Interface
• Residence Location Applications MVA/ Dealers
• Work Location
• SSN / Credit History
Insurance Information Medical Information
• Database
Interface
• Claims History Applications Physician / MIB
• Other Company’s Cover
• Sales Person
External Sources
Fraud Incidence during Claims Settlement
The claims handling process of (automobile) compared to underwriting since it is the final stage
1
insurance policies is more prone to fraud as of an insurance policy resulting in payment to the
policyholders. Once an insurance claim payment is
1 made, the claim becomes ‘legitimate’ and the
There are basically six different types of coverages that are usually
available under an auto insurance policy. Some may be required by law. money paid is lost ‘forever’. Research has shown
Others are optional. These coverages are: (a) Bodily injury liability - for
injuries the policyholder causes to someone else, (b) Medical payments or
Personal Injury Protection (PIP) - for treatment of injuries to the driver and
that the magnitude and frequency of occurrence of
passengers of the policyholder’s car, (c) Property damage liability - for
damage the policyholder causes to someone else’s property, (d) Collision an insurance fraud is greater at the claims stage as
- for damage to the policyholder’s car from a collision, (e) Comprehensive
- for damage to the policyholder’s car not involving a collision with another
car (including damage from fire, explosions, earthquakes, floods, riots, an accident involving a hit-and-run driver or a driver who does not have
and theft) and (f) Uninsured motorists coverage - for costs resulting from insurance
4
compared to the time of applying for a policy. Fraud • Injuries (payable under PIP) claimed to have
may be committed at different points in the happened as a result of a certain accident are
insurance claims transaction and different parties not in tune with the severity of the accident
may be involved – the insured, third-party claimants • Considerable delay between date of the
and professionals who provide services to the accident and the date of filing the claim
claimants. Common frauds include: • Diagnostic tests suggested by the physician
does not justify the intensity of the accident
• “Padding” or inflating the actual claims
• In case of claims arising from a specific area,
• Misrepresenting of facts on the claim application reference is made repeatedly, to a specific
• Submitting claims for injuries or damages that physician, or a specific garage located in that
never occurred area
•
2
“Staging” accidents • Repeat claims from a specific / group of
individuals at regular intervals
Claims’ department is typically the first line of
defense an insurance company has, to combat Knowledge of subject, experience and attention to
fraud. Manual detection processes used by the detail are essential in detection of fraudulent claims.
claims departments in most of the companies are While human experts are capable of identifying
ineffective and most of the fraud slips-through some of the red flags and simple fraud patterns,
undetected. Even the best claims adjuster cannot sophisticated modeling techniques are required to
perform the detailed analysis needed to find the find more complex patterns of fraud. Advancements
complex patterns that indicate fraud. It is also made in technologies like data mining and neural
estimated that industry wide, claims adjusters networks can help in the accelerated decision-
handle upto 250 claims at any given time. These making abilities of systems and empower the
loads only add to compounding the problem further. insurers to do better risk assessment. Extensive
Based on the claims handling experiences of Data modeling, data warehousing, and data mining
insurance industry, some indicators to suggest a capabilities are essential for establishing trends and
potential fraudulent intention have been identified. performing analysis. However, all these techniques
are data intensive and the availability of client
• Loss Value does not justify the location, time
relevant information is very critical for their success.
and nature of the accident involved
2 Staged accidents usually follow some basic schemes like (a) Stopping
suddenly for no apparent reason, (b) Disregarding the right-of-way
intentionally, or giving up the right-of-way on purpose, (c) Reporting
passengers who were not in the vehicle at the time of the accident, and/or
witnesses that were not actually on the scene, (d) Claiming excessive
bodily injury not commensurate with vehicle damage, (e) Claiming
property that was not in the car at the time of loss, (f) Falsely reporting a
car as stolen, (g) Fake hit-and-run claims, and (h) Destroying a vehicle by
setting it on fire
5
Use of Technology in Handling Insurance Fraud
Many of the actual and potential indicators of fraud • Transaction Based Profiling – The ‘credit-
and misconduct reside within the insurers financial, card’ vendors, for tracking the spending
operational, and transactional data and can be behavior of their customers, very effectively
identified using appropriate data analysis tools and employ this technology. This would involve
techniques. Fraud detection systems operate by collecting, analyzing and transforming client
consistently and dependably relying on diverse data into a set of features to describe the
mathematical techniques like probability, statistics, behavior of insured. The insurance history
and artificial intelligence. The propensity of an details of an insurance policy holder like the
applicant to involve in fraud at a later date may be number and type of policies held, the number of
‘guesstimated’ using data analytics techniques. An claims made etc., when mapped to the financial
effective solution to insurance fraud and abuse profile of the client may present a clear picture
relies on a comprehensive approach and not limited regarding the intent of the insured. Access to a
to a single technology. These techniques can vary centralized data repository having the ability to
from a simple business rule based “red-flag” provide the said information and containing the
triggering system to very complex pro-active data transactional history of (all) the insureds is
analysis systems that use sophisticated analytical essential for ‘tracking’ the profile of the insured.
tests, computer-based cross matching, and non- • Neural Networks – Neural Networking Models
obvious relationship identification techniques to replicate the functionality of human brains and
highlight potential fraud and misconduct unnoticed help in stimulating the situational behaviors of
by management for years. Among other things the insureds, based on certain preset “stress
these techniques provide for: factors” or claim parameters. The output is
usually in the form of scores ranging from 0 to
• Identification of hidden relationships between 1000 where a higher score indicates a greater
people, organizations, and events to analyze likelihood of the malafide intent. These systems
suspicious transactions are particularly suitable for claims processing
• The potential to continually monitor fraud threats since the systems can adjust according to the
and vulnerabilities. capacity of the investigative units. Scoring
• The ability to consider and analyze thousands of methods employed by these technologies help
transactions in less time, more efficiently, and in identifying the claims with highest fraud
cost-effectively than using more traditional propensity.
forensic sampling techniques • Data Mining technologies – This technology
employs the mining (searching) of all the data
Some of the emerging technologies that can be
sources for insurance information and
considered by the insurance industry to handle the
correlating the results with known fraudulent
claims fraud include
claims to indicate relationships that can be used the reason for conducting the search. . As in the
to monitor future claims or enrollments. The case of transaction based profiling, access to
volume of data that needs to be searched overwhelming amounts of claims data is
depends on ‘predefined’ set of parameters and required to identify the required information.
A simplified model based on the above three techniques and useful for identification of potential frauds is presented
below
Input 1
Logic
Rule 1
Debit 1
Input 2
Debit 2
Logic
Rule 2 = Fraud Score
Input 2
Debit 3
Logic
Risk
Rule n
Identification
Business / Mathematical
Logic
Input n
Data Input
User data inputs are captured and used by the experience of insurance industry, (b) the % of
business/mathematical logic layer. Each node in the claims involving the said distance & locations (c)
mathematical layer has embedded business logic the policies sold vs. the claims experienced and (d)
and indicates a unique combination of the inputs the risk appetite of the insurer. This process results
and the expected output. This logic could be in marking the ‘intelligent’ leads required to identify
anything like a simple ‘if…then…else…’ type of the possibility of a claim. A multiple regression
business rule or a complex regression model. algorithm written (instead of the rules) taking the
Suppose, the input ‘A’ is the residence location and said factors into consideration would then be used
input ‘B’ is the work location, and a business logic to arrive at a variable value (probable claim value).
pertaining to distance (parameter) would probably This number indicates the probability of an event
state that if the distance between residence and happening. This approach is similar to the
work is more than a given certain value (say 25 ‘exception reporting’ mechanism employed in the
kms) the probability of the accident is 30%. Results information technology processes. Multiple values
obtained from this activity are then compared to the can thus be generated using different parametric
(a) analytical data emanating from claims values (inputs) and stored in a database, to
7
establish a trend. Based on the risk appetite of the also. Most of the claim related information is
insurance a unique debit value can be assigned to readily available with the insurance company
each of the results. By running multiple stimulations and only certain parameters like place of
of this model through large sets of data, a accident and the parties involved, vehicle
comprehensive database of the debit values can be related information etc would differ from case-to-
built and a trend of input vs. output can also be case. Stimulations can be run for such data and
generated. This database would form the basis for a database of the fraud values may be
establishing the fraudulent intent (if any) on the generated for comparing with ‘real-time’
basis of the information supplied by the insured. situations.
• On receipt of a fresh application, the values that The above model incorporates features of
are input are run through the model to calculate predictive analytics, neural networks and
the debit value and compared to the data in the regression-based models. Since there is minimal
database to establish the “probability” of the human intervention (except at data entry stage and
case ending up as a claim. If the debit values execution of the application), results are expected
fall under the category of suspected claims, the to be accurate. However, elaborate programming
insurance company can decline cover and avoid and processing capabilities are required along with
a potential claim. large amounts of data, for the design and
• With little modification, the system can also be development of the system.
used to ascertain the “genuine-ness” of a claim
8
Conclusion
Insurance fraud is multifaceted. There is no single Unfortunately, the absence of any centralized data
typical profile of the fraudulent claimant; repository (like the MIB) is only encouraging the
professional criminals as well as an ordinary citizen fraudsters to design ingenious ways of
can commit fraud. The strategy to combat fraud misrepresenting the information from various
must not only be reliable and effective in its insurance companies. Fighting fraud requires a
prevention but also be capable of detecting commitment to process review, workflow
fraudulent activities as and when they occur. A automation, and auditing. Insurers must also
collaborative effort comprising of the insurance maintain a constant vigil for new types of fraud and
industry players, the regulators and the ‘policing’ the evolution of existing fraud schemes
authorities and all the relevant parties (attorneys,
physicians, garage owners) may help in designing A good strategy to combat fraud involves effective
intelligent systems as a part of the risk mitigation risk identification, analysis, and reporting activities
strategies against losses that happen due to and it requires the use of a combination of data
fraudulent claims. The performance of these validation and mining capabilities, visualization
systems depends on the availability of client data, techniques and reporting tools to identify
which can be shared between various insurers. A questionable behavior before a claim is paid.
proactive strategy enables insurers to remain Replacing the traditional manual processes that
vigilant against fraudsters protecting themselves involves sorting though large amounts providers
against both small and large-scale claims frauds data and claims information is not effective in
attacks. establishing potentially fraudulent, wasteful, abusive
or questionable behavior.
About NIIT Technologies
NIIT Technologies is a global IT Solutions major servicing customers in North America, Europe, Asia and
Australia. The company offers services in Application Development and Maintenance, Enterprise Solutions
including Managed Services and Business Process Management to enterprises in the Insurance, Banking
& Financial Services, Travel &Transportation, Manufacturing and Retail sectors. NIIT Technologies’
subsidiaries NIIT SmartServe Limited and NIIT GIS Limited offer Business Process Management and GIS
services respectively.
9