WNS DecisionPoint - Report - Fighting Insurance Fraud With Big Data Analytics

INSURANCE FRAUD DETECTION AND
PREVENTION IN THE ERA OF BIG DATA

CURBING FRAUD BY APPLICATION OF ADVANCED
ANALYTICS ACROSS POLICY LIFECYCLE
INSURANCE FRAUD DETECTION AND PREVENTION IN THE ERA OF BIG DATA
CURBING FRAUD BY APPLICATION OF ADVANCED ANALYTICS ACROSS POLICY LIFECYCLE
Overview
Every year claims and underwriting fraud cost $80 billion
and ~ $34 billion, specifically to Property and Casualty
Industry (P&C). This indicates the inefficiency of current
fraud detection techniques deployed by the insurers. The
ability to analyze only a small portion of data is one of the
main reasons for the under-performance of the existing fraud
detection techniques, especially when it comes to detecting
sophisticated fraud operations. Fraud management systems
that leverage advanced analytics tools such as predictive
modeling can greatly impact both the cost and success of
fighting fraud, which in turn will positively impact the
combined ratio and profitability.
With the growing complexity of fraud, insurers need to put in

place forward-looking fraud detection techniques that
efficiently analyze untapped data and help to minimize fraud
loss. Insurers tend to collect voluminous amounts of Big Data
in various forms (structured, semi-structured and
unstructured) during the policy application, servicing and
claims processes, which are seldom evaluated in fraud
investigation operations. Combining analysis of Big Data
along with the existing fraud detection techniques enable
insurers to augment productivity of underwriting and claims
processes and also maintain a competitive advantage.
To assess the impact and adoption of Big Data analytics on

detecting and preventing insurance fraud, WNS
DecisionPointTM conducted a survey of the US P&C insurers.
The study analyzes current fraud detection practices
adopted by insurers at three stages of the policy life cycle:
the underwriting stage, the claims handling stage, and the
investigation stage. It highlights the evolution of fraud
detection techniques from manual processes involving an
expert-based approach, to automated techniques, to data-
driven analytical models that offer more valuable insights to
identify fraudulent activities.
wnsdecisionpoint.com 1
Insurance fraud is one

of the biggest crimes
in the U.S., reporting
payout of $80 billion
every year, imposing
heavy costs on
insurers and its
customers
2 wnsdecisionpoint.com
THE COST OF INSURANCE FRAUD

Insurance fraud is one of the biggest crimes in the U.S., reporting payout Exhibit 1
of $80 billion (for the entire insurance industry) every year, imposing A Few Instances of
heavy costs on insurers and its customers. Fraud in Insurance
Claims and its Impact
on the U.S. Insurance
FRAUD AT POINT OF SALE Industry
Point of sale (POS) fraud, also Workers' compensation (part of
known as premium or underwriting commercial line insurance) is most
fraud, takes place when the exposed to such fraud, where
applicant purposely hides or employers avoid premiums by The FBI arrested 36
distorts facts/information when reporting fewer employees or people in New York -
obtaining insurance coverage. inappropriately classifying
including 10 doctors and
Fraud at the POS stage, which employees with high-risk jobs as
erodes 10 percent of insurance holding lower-risk ones. About three lawyers - for their
revenuei, is increasing due to a rise 10-20 percent of employers in the alleged involvement in a
in digital sales channels, which U.S. misclassify at least one worker $279 million auto
make it easy for individuals to as an independent contractorii.
insurance fraud
reduce premiums by easily Employers committing such fraud
manipulating data, such as age, are higher in number in the
previous claims and zip code, Southern states, where around 30
among others. To make matters percent of construction workers
worse, pressure on carriers to have been wrongly classified as
increase revenues and higher independent contractors, Marc Thompson, deeply
customer switch ratio limits amounting to annual losses of
in debt, torched his
underwriters from scrutinizing all $400 million in Florida, $467
the applications in detail. million in North Carolina, and $1.2 home for the $7.3 million
billion in Texasii. in insurance money
FRAUD AT CLAIMS STAGE

According to the survey by FICO1 about 5-10 percent of the claims Insurance fraud drains
on 'Insights into Insurance Fraud amount, and more than 30.5 more than $50 billion
Trends' in 2013, around 35 percent percent of the participants pegged
from insurers each year
of respondents estimated that the loss rate between 10 and 20
claims fraud costs their company, percent.iii and costs the average
U.S. family between
$400-700 in the form of
increased premium,
according to the FBI
1. FICO - Founded in 1965, FICO is a leading analytics software company that provides tools across multiple industries to manage risk,
fight fraud, build more profitable customer relationships, and optimize operation
3
2
Fraud in 'No-Fault' States
The U.S. is also witnessing a higher top five states that generated the impose a strict timeline within
rate of Organized Group Activity most questionable claims were: which the claim must be accepted
(OGA)3. The National Insurance Florida (3,530), California (2,679), or rejected. Although intended to
Crime Bureau (NICB)4 reported a 14 Michigan (1,080), Texas (1,050) and simplify and expedite the claims
percent rise in Organized New York (765)v. Looking closer, process, these policies make it
Group/Ring Activity from 2,310 three are 'No-Fault' states (Florida, much easier to stage accidents and
questionable claims (Qcs)5 for nine Michigan, and New York), and one, defraud insurance companies.
months (Q1-Q3) in 2012 to 2,643 Texas, although not 'No-Fault,' has Worse, the timelines make it
QCs for nine months (Q1-Q3) in partially adopted some 'No-Fault' difficult for insurers to detect fraud
2013.iv In 2012, NICB conducted a policies. 'No-Fault' policies put fast enough, and even if they do,
survey to examine OGA and much of the burden to pay on the they are often sued for negligence
reviewed 13,014 Qcs for the period injured party's own insurance in their examination of the claim.
January 2008 to June 2012. The company, regardless of fault, and
Fraud in Personal Lines
Fraud losses in personal line, mainly Worse, because of liberal rules and injury protection (PIP) claims
led by fraudulent auto injury claims, dishonest lawyers, who support (Florida - 31 percent, New York
were estimated to rise in double such crimes, 'No-Fault' states 24 percent, Massachusetts
digits over a two-year period and reported an even higher 22 percent, and Minnesota
reach $18.5 billion, in 2014vi. appearance of fraud for personal 22 percent)vii.
2. States with the no-fault system of insurance, where insurers indemnify for basic damages and injured parties are paid by their own
insurance company, regardless of fault
3. Any specific group made up of entities and/or individuals, who systematically and repeatedly conduct pre-planned activities for the
purpose of generating fraudulent insurance schemes
4. NICB is a non-profit organization that partners with insurance companies and law enforcement to help identify, detect, and prosecute
insurance criminals
5. Claims referred to NICB by member insurance companies for closer review and investigation of possible fraud
Exhibit 2
Estimated Annual Losses in P&C Lines due to Fraudulent Claims in 2014

($ Million)6
Personal Lines Commercial Lines Combined Lines
Auto Liability $8,820
Auto Physical $5,158
Homeowners Multiple Peril $4,484
Workers' Compensation $3,329
Other Liability $2,833
Commercial Multiple Peril $2,120
Commercial Auto Liability $1,461
All Other $892
Medical Professional Liability $562
Accident and Health $552
Commercial Auto Physical $411
Products Liability $349
Financial Guaranty $224
Farm owners Multiple Peril $214
Allied Lines $1,057
Inland Marine $562
Fire $547
Mortgage Guaranty $308
Warranty $130
Source: WNS DecisionPointTM estimates (based on data by National Association of Insurance Commissioners (NAIC)
Vulnerability of personal lines is Organized fraud is also high in out of 13,014 QCs were reviewed
also evident from double-digit personal line insurance. during the study period).
increase in questionable claims As per NICB's report on Organized
(QC), referred to NICB, in the last Group Activity, which stated
four years. The rise was mainly 'personal automobile' as the policy
driven by referral reason suspicious type that reported the highest
theft of vehicles. number of referrals (10,659 QCs
6. Assuming 10 percent of the Losses Incurred and the Loss Expenses incurred during respective years
5
IMPACT OF FRAUD
Fraudulent Claims Skew Combined Ratio and Underwriting Profitability
Fraudulent claims directly impact 2014. During the same period, P&C insurers to post reasonable
Loss Ratio, thereby, reducing a 1.8 pps (percentage points) levels of profitability.
profitability and also negatively increase in Loss Ratio also
impacting Return on Equity (ROE). negatively impacted underwriting This emphasizes a strong need of
As reflected in Exhibit 3, despite a profits by 26 percent in 2014. loss management to control
fall in Expense Ratio, an elevated Maintaining Combined Ratios dishonest claims payout through
Loss Ratio pulled down ROE in below 100 is a pre-requisite for better fraud detection techniques.
Exhibit 3
U.S. Property and Casualty Insurance Key Financials

Fraud Claims Ratio (Approx.) Loss Claims Ratio (Excluding Fraudulent Claims) Expense Ratio ROE
0.0 50.0 100.0
2013 6.7 60.5 28.0 95.2
2014 6.9 62.1 27.6 96.6
0.0% 5.0% 10.0% 15.0%
Change in Underwriting Profits (in $ Billion)
$0 $10 $20 $30 $40
Net Underwriting Gain/(Loss) 2013
Net Premiums Earned
Net Losses Incurred
Loss Expenses Incurred
Other Underwriting Expenses
Other Expenses
Net Underwriting Gain/(Loss) 2014
Source: Insurance Information Institute and NAIC
Fraud Also Impacts Insurer's Ability to Offer Competitive Premiums
Fraudulent claims add to premium Fault' states such as New York, no longer feasible for the insurer.
costs, since insurers are compelled Florida and New Jersey. These This trend, combined with low
to pass on the cost of such claims states have the highest auto investment yield (drop of 57 pps
to policyholders. According to the premium prices, mainly caused by from 3.74 percent in 2011 to 3.17
WNS DecisionPointTM survey, half of fake claims involving staged car percent in 2014)viii, has also
the respondents felt that fraud accidents. Fraud is also primarily impacted the profitability of the
escalates premiums by three to five responsible for rapidly rising insurer. Given all this, insurers need
percent. Nearly a quarter of the workers' compensation premiums, to shift from traditional ways of
respondents reported escalation to driven by false injury claims and detecting fraud and adopt
be over five percent; while the rest misrepresentation of facts at the advanced fraud detection and
mentioned it to be between one time of application. prevention techniques.
and three percent. The cost of
fraud added to the premiums However, given the competitive
varies across insurance lines and environment, increasing premiums
states, and is much higher for 'No- to compensate for fraud losses is
7
SHIFT IN FRAUD DETECTION

APPROACH WITH RISE OF
ANALYTICS
Traditional approach of detecting fraud is still being used by few insurers in this era of complex and sophisticated
crime. Some of the traditional methods are internal audits, red flag indicators, and scoring models, as explained in
the illustration below.
Exhibit 4
Traditional Fraud Detection Methods
Traditional Methods
Internal Audit Red Flag Indicator Scoring Models
 Random checks on sample  Rule-based indicators to detect  An advancement to rule-base

claims leave a room for some suspicious behavior techniques, wherein insurers
fraudulent claims to go provide scores to rule-based
undetected  Red flags indicate the need to indicators (either manual or
further investigate a claim and automated)
 Lack of experience of insurance does not conclude the claim to
fraud professionals results in be fraudulent  Based on fraud propensity
lower detection and scores, claims are classified into
investigation rate various segments such as high,
medium and are subsequently
referred to for further
investigations
Source: WNS DecisionPointTM
These traditional methods primarily claims. Traditional approaches are satisfaction adversely impacts
detect known fraud patterns using also known for high false-positive customer retention. Despite these
sampling techniques. Since these rates (flagging genuine claims as drawbacks, some P&C insurers in
methods require manual fraudulent), which impact customer the U.S. still continue to use these
intervention, there is a higher satisfaction. Given the competitive traditional methods as they lack
possibility of human error and environment with high customer the expertise and infrastructure to
longer lead times from fraud switching rates, especially for auto deploy advanced analytical
detection to the settlement of insurance, low customer techniques.
AUTOMATED AND ANALYTICS-DRIVEN FRAUD

IDENTIFICATION AND INVESTIGATION
Realizing the limitations of occurrence of fraudulent claims 2014 and 2016ix. Investments in
traditional methods, many carriers (notably in the case of advanced solutions such as link
have started adopting anti-fraud catastrophes, which greatly analysis, predictive modeling and
technology to determine the magnify the scope for fraud), and other sophisticated data analytics
likelihood of claims fraud. As per to drive innovations in detecting are expected to increase and
the survey report on 'Insurance organized activity early in the insurers will partner with specialist
fraud detection technology' claims cycle. service providers to drive down
conducted by Coalition Against fraud-related expenses. This helps
Insurance Fraud (CAIF) and SAS The increasing adoption is reflected insurance companies quickly and
Institute, the number of in increased investments in data accurately determine whether a
respondents using anti-fraud analytics and modeling claim needs further investigation
technology grew from 65 percent technologies by insurers. According and to determine the complexity of
in 2012 to 71 percent in 2014ix. to a survey of 42 insurers by CAIF the claim.
and SAS Institute, 25 percent
Insurers are also increasingly planned to increase investments in
adopting analytics to predict the anti-fraud technology between
Exhibit 5
Planned Investment Areas in Fraud Detection Technology

Link Analysis Automated Red Flag/Business Rule Predictive Modelling Geographic Data Mapping
Exception Reporting/Anamoly Detection Text Mining Data Visualization
2015-2017 46% 18% 38% 21% 16% 30% 21%
2012-2014 21% 25% 32% 12% 18% 30%
Source: The State of Insurance Fraud Technology Report, 2014, by CAIF and SAS, Sample Size = 42
How Fraud Analytics Add Value at Each Stage of the Policy Life Cycle?
Fraud can be detected across the enable insurers to have a
four stages of the policy life cycle comprehensive view of each
as mentioned below. Fraud policyholder, which paves the way
analytics, if deployed at various to discover organized frauds.
stages of the policy life cycle,
9
Exhibit 6
Guard your Margins by Applying Advanced Fraud Analytics over Various

Stages of the Policy Life Cycle
Point of Sale First Notice of Loss (FNOL)

Investigation Post Claims
Claims Handling
Validate a Larger Number of

Capture Precise Identity Promptly Uncover Linkage
Claims in a Shorter Period of
of Applicants of Claimants with Fraud
Time with Higher Accuracy Quickly Analyse
Organized Frauds with
Categorize Risks and Derive Higher Accuracy
Discover the Linkages of Reduce Claims
Fraud Propensity Scores Investigation Costs
Applicants with Fraud
in Near Time
Detect Abnormal Behavior Reduce

Patterns of Applications Investigation Time
Source: WNS DecisionPointTM Analysis
Point of Sale (POS)/Renewal Stage

Since a sizable part of claims fraud companies to put a check on the exposure to known fraud at the
is initiated at the application stage, underwriting fraud to decrease claims stage.
it is very important for insurance premium leakage and avoid
Exhibit 7
Fraud Detection Process at POS/Underwriting Stage
Customer Insurer
Process
Implementation Assessment-Based
Fraud Checks Score Generation
Services Action
 Application  Use of single or  Risk scores are  The decision to accept or

received combination of techniques based on affinity reject a policy is based
through for complete analyses of fraud on the outcome of the
agents, investigation
brokers, insurer  Insurers use analytics to  Applications are
website and categorized as  If scores are above the
 Capture precise identity per the scores specified limit, the
other online of applicant
channels application is accepted
 Reason codes are
 Discover the linkages of attached to the  High risk applications are
the applicant to frauds risky applications directed to experts for
 Detect abnormal investigation
behavior patterns
Apart from the traditional Red Flag DecisionPointTM survey, all Exhibit 8. As a result, insurers were
indicator method, many insurers respondents deployed predictive able to develop premiums that
are using a combination of modeling techniques at the accurately reflected the relative risk
business rules and analytics to underwriting stage to enhance characteristics of the pool of
tackle underwriting fraud. understanding of current and underlying policyholders.
According to the WNS future insured risks, as seen in
Exhibit 8
Types of Fraud Analytics Currently Used by P&C Insurers at the Underwriting Stage
Social Media Analytics 25%
Social Network Analysis/Link

25%
Analysis
Geographic Data Mapping 25%
Text Mining 25%
Data Visualization 50%
Automated Red Flag 70%
Predictive Modeling 100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Source: WNS DecisionPointTM Survey
After implementing analytics at the detect fraudulent policies and section of insurers also experienced
point of sale stage, most of the exercise better control over lower claims volumes and additional
carriers were able to promptly business rules indicators. This benefits as shown below.
Exhibit 9
Benefits Realised from Using Fraud Analytics at the Underwriting Stage
Lower % of the Policies

25%
that Get Cancelled
Higher Detection Rate of

50%
Suspicious Policy
Decline in Average Time to

50%
Identify an Issue in a Policy
Lower % of Policies
that Registered Claims 50%
within 100 Days of Sale
Reduced Premium
50%
Leakage
Lower Average Claims

75%
Value
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
11
Claims Stage
At the FNOL stage, the role of Thus, the inability of claims analytics in the fraud detection
claims handler is very important as handlers to accurately notice process at the claims stage,
they are responsible for processing doubtful patterns results in fake investigation stage, and post-
and investigating insurance claims claims getting paid. The chart claims stage.
relating to customers' policies. below illustrates important role of
Exhibit 10
Fraud Detection Process at Claims Stage, Investigation Stage, and

Post-Claims Stage
Investigation-Post
FNOL Claims Handler Adjuster Investigator
Claim Settlement
 After loss of the  Claims handler  Adjuster physically  Investigator further  Special teams
property/vehicle is identifies suspicious assesses the examines the examine large
reported by policy claims by applying damages, the loss referrals highlighted claims data using
holder, FNOL agent rule based value and estimates by claims handler data mining and
gathers details techniques the claims amount multi-variate
pertaining to the  Refers to various analytics techniques
event and  Verifies claims data  Uploads the findings databases and uses
subsequent loss against companies gathered at this a mix of statistical  Post claims analyses
records and industry stage into the and traditional enable investigators
 FNOL agent databases systems which is techniques to to reveal fake claims
uploads the then referred to by identify the fake that are not evident
information  Assigns scores to investigators claims in smaller data sets
gathered in the claims based on
above stage in the fraud possibility  Initiates further
company s system action based on the

 Directs high-risk outcome of the
which can be claims to more
referred to by claims investigation
experienced claims
handler and handlers or to SIUs
adjusters
 Informs an adjuster
 FNOL agent then for futher action
initiates claims
process by
informing claims
handler
Most claims handlers assess claims records) and external data sources as analysis of incorrect or outdated
using a set of specified indicators (NICB, Insurance Services Office information results in undisclosed
and then verify submitted details (ISO)7 Claims Search, among dishonest claims and also generates
with the data generated from others). It is very important to high false positive rate, which
internal systems of record (log provide accurate and recent overburdens the Special
book, policy data, and customer information/data to claims handlers Investigative Unit (SIU).
7. ISO provides information about property/casualty insurance risk for a broad spectrum of commercial and personal lines of insurance
In order to validate a larger number modeling, enable insurers to understand the impact of ongoing
of claims in a shorter period of time categorize risks and derive fraud fraud analysis on their business.
with higher accuracy, claim handlers propensity scores in real time, As per the survey conducted by
are required to use sophisticated which helps claims handlers adjust WNS DecisionPointTM, insurers using
analytics. This, in turn, results in the their line of questioning and route analytics reported higher benefits
identification of low-fraud incidence suspicious claims to investigators. compared to the insurers, who are
events (with probability as low as Analytics solutions, such as using traditional methods or
0.001 percent). Some analytics reporting and visualizations of data automated indicators
techniques, such as predictive patterns, also help insurers easily (see Exhibit 11).
Exhibit 11
Benefits Realized from Usage of Analytics to Detect Fraud at Claims

Handling Stage
Insurer using Analytics Insurer Using Traditional Approach or Automated Indicators
Reduced Referral
Time
More Referrals
Better Understanding
of Referrals
Enhanced Reporting
Reduction in
Assessing Time
Reduction in
Assessing Cost
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
No. of Respondents (in%)
Investigation Stage
Within the claims life cycle, others) to check the veracity of the fraudulent activities, thereby,
analytics is largely used at the information provided by claimant. decreasing the amount of time
investigation stage. At this stage, Analytics play an important role in taken to expose hidden
investigators scrutinize each claim collecting, consolidating and relationships among entities and
referred to by claims handlers. integrating scattered data from potential fraud. However, owing to
They compare claims' data against various data sources. Techniques lack of in-house expertise,
various data sources (such as such as social network analytics extensible and scalable information
policy information, claim history, enable investigators to quickly foundation and technology
medical reports, NICB, ISO, among uncover linkage of claimants with systems, adoption of analytics by
13
the U.S. based P&C insurers is low predictive analytics is being underuse of other solutions, as
and they usually rely on automated deployed at investigation stage to shown in the chart below.
red flags and business rules to improve fraud detection and speed
determine fake claims. Further, only up claims processing, indicating
Exhibit 12
Techniques Currently Employed by Insurers to Detect and Investigate

Fraudulent Claims
Business Rules 100%
Automated Red Flag 100%
Predictive Modelling 75%
Data Visualization 25%
Social Media Analytics 25%
0% 20% 40% 60% 80% 100% 120%
No. of Respondents (in%)
Post-Claims Stage
Application of fraud analytics at to trace in individual claims business rules, automated red flags,
the post-claims stage is in a investigations. This is more predictive modeling, and text
nascent phase and is mostly common in the No-fault States, mining, among others) to uncover
deployed by players with P&C where insurers are required to unrevealed inter-relationships
revenues over $5 billion. Few make faster settlements and do between claims and other entities8
insurers employ a special team not have enough time to scrutinize involved in such activities. These
within their SIU, who specially look claims thoroughly. insights enable insurers to discover
into claims after their settlement. organized fraud, and also to
This team analyzes enormous At this stage, the SIU analysts continuously improve the rules and
volumes of claims data to compare large volumes of claims models operating at the POS,
determine patterns of fraud, which data using multiple techniques and FNOL, and Investigation stages.
surface gradually and are difficult tools (such as combinations of
8. Claimant, locations where damage happened, witnesses of accidents, and auto body shops, among others
BIG DATA FRAUD ANALYTICS

Fraud analytics has evolved from data that is available within and insurance organizations and is
building models and providing outside the company. mostly historical in naturex. It is to
fraud scores to policies and claims, be noted that analysis is based on
to enabling real-time fraud Insurers mostly analyze structured historical data tends to lose its
detection by efficiently and and unstructured data, which predictive power beyond a certain
effectively processing the deluge represents just 15-20 percent of the point.
of structured and unstructured total data that is generated by
Data Sources
Internal External
CRM
Structured
Credit
Claims
History
Record
Billing
Data Application and Claims
Policy
Data with NICB and ISO
Data Types
Information
Log Medical
Notes Reports Police
Records
Un-Structured
Adjustor s
Web Chat Notes Social Media
Transcripts Interactions
15
Exhibit 13
Types of Data Analysed for Fraud Detection by Insurers Across Stages
ISO - Application Underwriting Stage

and Claims Data
Industry Fraud Alerts
/Watch List Data
Medical Report
Fraud Bureau
NICB - Application
and Claims Data
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
ISO - Application Claims Stage

and Claims Data
Police Reports
Claims Record
NICB - Application
and Claims Data
Medical Report
Policy Information
Billing Data
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
Claims Record Investigation Stage
Policy Information
Billing Data
ISO and NICB - Application
and Claims Data
Medical Report
Reports by Department
of Motor Vehicle
Police Reports
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
In the current environment, where  Applying analytics on claimant's policies/rules at underwriting stage
fraud rings adopt sophisticated social network data may reveal and also save on promotional costs
techniques, real-time data his connection with entities who by avoiding marketing to such
availability is crucial to increase are/were involved in fraudulent 'potential high-risk' prospective
accuracy. Analysis of internal and activities customers. Apart from disclosing
external unstructured data hidden fraudulent activities,
 Log notes and web interactions
provides a great opportunity to unstructured data analysis also
uncover complex fraudulent can confirm claimant's urgency enables insurers realize many
activities, which are difficult to for claim settlement, adjustor's benefits in terms of efficient
trace through analysis of the notes may point out inflated processes and reduced costs.
structured data. For instance: vehicle repairs
 Scanning through social media

Such a multi-pronged, coordinated
interactions of a claimant may approach to fraud can be used for
reveal his visit to a bar before his comprehensive risk shrinkage by
car accident altering fraud detection
Exhibit 14
Benefits of Analysing Unstructured Data as Reported by the U.S. P&C Insurers
Higher Fraud Detection

Rate of Suspicious Policy
Decline in Average Time to

Identify an Issue in a Policy
Lower % of the Policies that Registered

a Claim within 100 Days of Sale
Lower Average
Claims Value
Reduced Premium
Leakage
0% 19% 38% 57% 76% 95%

% of Respondents
Benefits accrued from the e-mails, blogs, web interaction,

assessment and evaluation of audio/video, and images, is still at a
unstructured data have nascent stage. Predictive analysis,
encouraged insurers to invest more on the other hand, uses a
on tools and techniques to examine combination of rules, modeling,
Big Data (notably social media text mining, database searches and
posts) to flag claims for closer exception reporting as mentioned
inspection, priority handling and in the charts below.
settlement. However, use of
unstructured Big Data such as
17
Exhibit 15
Extent of Usage of Unstructured Big Data in Next 24 Months
Blogs Underwriting Stage
Images
Social Networking
Websites
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
E-mails Claims Stage
Blogs
Call-Center Logs/
Agent Notes
Images
Web Interactions
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
Web Interactions Investigation Stage
Call-Center Logs/
Agent Notes
Social Networking
Websites
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% of Respondents
Big data is scattered within and external data with internal data and
outside the organization and is in obtain a complete view of the
an unstructured form, which makes policy holder, claimant, and his/her
it very difficult for an insurer to actions. Thus, very few insurers are
gain meaningful insights. Insurers able to analyze unstructured data
are required to possess robust data owing to technology inadequacies
management capabilities to be and a lack of in-house expertise in
able to cleanse and integrate the this area.
Building an Environment to Support the Use of Big Data Analytics
 The data (structured, semi- pasted e-mails, images, typos,  Techniques such as predictive
structured and unstructured) misspellings, etc. This data is also modeling, social media analysis,
collected from various sources transformed and loaded into the and social network analysis, are
are fed into the data collection data warehouse. integrated into the fraud
tool. identification modeling process,
 The analytics team uses the
which further modifies the scores
 The data is then cleansed, to be information stored in the data based on the output from these
loaded into a data warehouse. warehouse and combines that techniques.
The semi-structured data, as information into analytics models
sourced from ISO and NICA to generate fraud scores based  The alerts generated by analytics
reports, is reconstructed into the on multiple factors. The models are further investigated
structure appropriate for running information used includes and evidence collected from this
predictive models. Unstructured previous sentences, associations model is added to the existing
data sourced from log notes, with other people with past system to further build the case.
e-mails, adjuster's notes, among fraud cases, multiple rejected
others, may not be in English claims, and modifications to
and may contain fragments of personal information.
sentences, acronyms, copy and
Role of Big Data Analytics at Various Stages of the Policy Lifecycle

It is very crucial for insurers to implement Big Data analytics at all three stages of the policy life cycle to develop
an end-to-end, comprehensive fraud management programme.
Point of Sale or Underwriting stage Claims Stage
Investigation Stage and Post-claims

Prior to the Submission of a Claim FNOL/Claims Handling Stage
Settlement
 Examines policies for identifying theft  Determines discrepancies  Scrutinizes claims in detail and helps
and recognizing deceitful patterns much faster in the claims investigator to collect possible fraud
lifecycle through a evidence
 Provides real-time insights to
combination of traditional
determine and validate the true rule-based techniques and  Quickly incorporates claim data related to
identity of policy applicants advanced analytics, such as all entities involved from various
social media analytics and unstructured sources using text and data
 Discovers complex relationships and
predictive modeling mining techniques to establish
linkages with criminal activities relationships among various parties
 Reduces the average cycle involved in a claim
time for legitimate claims
 Creates leads and clues (through use of
link analysis and social network analysis)
to prepare evidence for fraudulent claims
 Promptly alters controls, based on

emerging trends and new fraud patterns
19
Benefits Generated by Insurers who Deployed Big Data Analytics
Although challenging, analyzing suspicious patterns or risks. average investigation time and 1.4x
Big Data enables insurers to Insurers adopting Big Data lower average investigation costs
understand unknown correlations analytics reported 40 percent compared to insurers, who have
much earlier in the policy life cycle improvement in the average not deployed such sophisticated
and to develop a deeper referral time, 50 percent more statistical techniques
understanding of emerging average referrals, 2.5x reduction in (see Exhibit 16).
Exhibit 16
Benefits Reported by Insurers at Claims Handling Stage

Low (deployed Big Data Analytics) High (Deployed Big Data Analytics)
Average (deployed Big Data Analytics) Average (not deployed Big Data Analytics)
Reduced Referral Time

(In Terms of # of Days)
More Referrals
(In PPs)
Reduction in Investigation Time

(In Terms # of Days)
Reduction in Investigation Costs

(In Terms of %)
0 5 10 15 20 25
Adoption of Big Data analytics at the investigation stage also enables insurer to gain similar benefits in terms of
cost and time. The average investigation cost of insurers, who use Big Data analytics, was 67 percent lower, and
the average time per claim investigation was 2x shorter, compared to the other insurers. Insurers also reported
other benefits as shown in the graph below.
Exhibit 17
Benefits Generated at Investigation Stage
% of Insurers Using Big Data Analytics % of Insurers Using Traditional Approach or Automated Indicators or Analytics
Better Understanding
of Referrals
Reduction in
False-positive Rate
Reduction in the
False-negative Rate
Higher # Investigations
Per Investigator
0% 20% 40% 60% 80% 100%

% of Respondents
Average Cost per Claim Investigation (in USD)
0 500 1000 1500 2000 2500 3000
Insurer Using Big Data

Analytics
Insurer Using Automated
Indicatorsor Analytics
0 20 40 60 80 100 120 140

Average SIU Analyst Time Per Claim (in Days)
Low Adoption of Big Data Analytics
The adoption rate of Big Data respondents have invested in crunch for funding such projects
analytics is very low across the Big Data analytics, complete and have infrastructure constraints
policy life cycle. This is especially deployment is accomplished by to support Big Data application
true at the underwriting stage, only a few of them, and most requirements, tool capabilities, and
where only one-fourth of insurers are either in the pilot stage best practices for implementation.
respondents have only invested in or at a knowledge gathering stage Adoption of Big Data analytics at
Big Data analytics and are still at a (see Exhibit 18). Insurers are the investigation stage is moderate
knowledge gathering stage. reluctant to invest in such projects, with near to 50 percent
as the practical challenge of deployment rate.
At the claims handling stage, defining ROI from such investments
although 73 percent of the remains. They also face capital
Exhibit 18
Percentage of Respondents Currently Using Big Data Analytics to Detect

Fraud at the FNOL Stage
Invested in Big Data Analytics Not Invested in Big Data Analytics
73% 27%
Stage of Big Data Adoption

25% 38% 38%
during FNOL Stage
0 20 40 60 80 100
Knowledge Gathering Piloting and Experimenting Deployed
However, increasing confidence in DecisionPointTM survey, around 33 years. Most of the claims handlers
Big Data analytics is driving a large percent of the underwriters, who and investigators, who have not
number of insurers to invest in have not leveraged Big Data invested, are also planning to
utilizing new sources of data that techniques, are planning to make deploy Big Data technologies
extend beyond traditional and such investments in the next year within two years.
internal sources. As per WNS and the rest, within a period of two
Challenges in Adopting Big Data Analytics
The challenges across the various stages of Big Data implementation need to be clearly understood and
addressed to ensure the maximum chance of success. Big Data adoption is broken into four phases:
Planning, Knowledge Gathering, Pilot, and Full-scale Deployment.
Exhibit 19
Adoption Phases of Big Data Analytics
Knowledge
Planning Pilot Deployment
Gathering
Adoption
Phases
 Sets plans to guide  Takes into account  Runs pilots to  Ensures that
project teams existing and determine the departments exposed to
throughout the untapped in-house feasibility of such Big Data analytics, have
adoption phases sources and assesses projects and its understood the
the requirement of impact on associated governance
 Determines the additional data types organizational goals and risk management
investments and practices
anticipated benefits  Analyses benefits
from such projects realized and runs  Ensures that
Key Activities
pilots again in case organizational
projects fails to requirements are in
produce expected place such as integration
results of the siloed data;
additional resources to
handle higher work
volumes; added capacity
to store, maintain and
manage such data
21% 16%
% of Respondets 37% 26%
at Each Stage
 No easy way to  Technology and  Making the  Capital allocation for

determine the ROI infrastructure infrastructure and big-data-related
of Big Data requirements to applications initiatives
investments acquire, host and production-ready in
analytically process operations  Higher levels of
the vast amount of environment technology support
Key Challenges data gathered
 Management of  Integrating multiple data
 Risk and governance large referral sources
issues(security, volumes generated
privacy, data by analytics usage
integrity)
Source: WNS DecisionPointTM Survey and Analysis
Insurers, at different phases in the optimize the usage of internal  Classify the business units as
Big Data adoption, face different and external data and owners and implementers to
problems (as stated in the above accordingly, assess the provide clarity regarding roles
diagram) and require appropriate organizational, information and responsibilities of each
solutions, which will not only management and participant
enable them to reach the technological changes needed
 Evaluate the gap between
deployment phase sooner, but also to embrace Big Data
maintain and enhance the estimated and actual benefits
 Understand the requirements derived from usage of Big Data
processes, thereafter.
of tools and techniques to analytics
 Insurers at planning phase: manage and glean relevant
 Document lessons learned
At this stage, while insurers face insights from big data sources,
many issues, their main concern application of statistical and from implementation failures
is determining the decision- predictive analysis techniques, and use that during the
support value of Big Data, as best practices and planning and design of the
reported by 75 percent of the organizational change next pilot
respondents. Nearly a quarter of readiness  Insurers, who have already
the respondents struggle to deployed Big Data analytics
 Make sure that all data points,
understand the definition and should:
use of Big Data and the analytics captured from various internal
capabilities required to enhance and external sources, are  Train SIU analyst and
decision-making speed and integrated into the company's investigator resources to
effectiveness. These insurers system to provide a unified institutionalize the usage of
should: view of policies/claims across Big Data analytics
the underwriting department,
 Ensure that Big Data strategies the claims department and the  Modify or update red flag rules
are aligned with business investigation department. If and reframe fraud detection
objectives internal IT systems are and prevention strategies,
fragmented and not geared to whenever a new fraudulent
 Conduct stakeholder
handle Big Data, the project pattern is discovered
communication/education will fail to produce desired
sessions to articulate the outcomes  Continue to explore the
business value of Big Data in untapped data existing within
improving the fraud  At the pilot stage, insurers the organization and external
management processes. Some should: data to promote higher and
of the benefits accrued to the accurate fraud detection and
 Ensure that various
insurer include increased true prevention
positive referral rate, early departments are ready for the
forewarning of suspicious pilot or proof of concept run Insurers, who are thinking of
behavior, reduced average cost especially the investigation adopting these techniques and the
of investigation per claim, and department, as they are likely underpinning technologies in the
shortened claim investigation to experience higher work near future, can consider enlisting
time by SIU analyst, among volumes the services of a specialist analytics
others service provider in order to share
 Ensure close coordination
the significant investments
 Insurers at a knowledge
between the various required in such projects and
gathering phase should: departments including SIU, acquire the right skills and domain
claims, underwriting and IT to expertise in the shortest period of
 Explore the volume and variety jointly determine Big Data time.
of unused data within and implementation blueprint
outside the organization,
THE KEYS TO SUCCESSFUL BIG

DATA ADOPTION
For robust fraud management  Ensure all the data points from evidence of fraudulent activities.
practices, a comprehensive the various departments involved
 Continue to explore untapped
framework is required, which scans in fraud detection process are
all possible areas of fraud in the integrated in a unified data and integrate such data
policy life cycle, in a unified policy/claim view. points into existing fraud
manner. Below are the emerging detection system.
 Reframe the prevailing business
best practices to successfully Since their creation, P&C insurance
operationalize Big Data analytics. rules by leveraging domain
expertise and set realistic companies have been battling
 Build a dedicated fraud threshold values for different fraudulent policies and claims.
management team to coordinate anomalies. Currently, this fraud can increase
fraud detection processes of claims expenses by up to a
 Use a combination of different staggering 20 percent. As schemes
various departments
underwriting, claims, and SIU. detection techniques to improve for committing fraud proliferate,
the efficiency of fraud traditional fraud detection
 Evaluate prevailing fraud management process. For techniques are becoming less and
detection system(s) and identify instance, business rules and less effective. Automated rules
gaps, if the system is not aligned anomalies detection can be have helped, but they are
with the organizational applied to all claims to identify inefficient and ultimately not the
objectives. discrepancies in data or to find best available solutions. The future
those claims that exceed the is clear. Big Data analytics offers
 Recognize the skill sets required
anomaly threshold limit. The underwriters, claims handlers, and
to implement Big Data analytics screened claims then can be investigators an advanced and
and accordingly, decide whether filtered by running predictive statistically reliable solution to
to build the analytics solutions models which assign a fraud revolutionize their fraud
in-house or partner with external propensity score to each claim. management processes. Adoption
vendors. If the decision is to use of Big Data analytics may be
an external vendor, evaluate  Integrate social media analytics
challenging, but the end result is
different vendors to find the and social network analytics with worth it.
solution that best aligns with the predictive modeling to uncover
company's requirements.
About DecisionPoint
Making key decisions that improve business
performance requires more than simple insights.
It takes deep data discovery and a keen problem
solving approach to think beyond the obvious.
As a business leader, you ought to have access to
information most relevant to you that helps you
anticipate potential business headwinds and craft
strategies which can turn challenges into
opportunities finally leading to favorable business
outcomes.
WNS DecisionPointTM, a one-of-its kind thought

leadership platform tracks industry segments served
by WNS and presents thought-provoking original
perspectives based on rigorous data analysis and
custom research studies. Coupling empirical data
analysis with practical ideas around the application of
analytics, disruptive technologies, next-gen customer
experience, process transformation and business
model innovation; we aim to arm you with decision
support frameworks based on points of fact. Drawing
on our experience of working with 200+ clients
around the world in key industry verticals, and
knowledge collaboration with carefully selected
partners, including Knowledge@Wharton, each
research asset draws on points of fact to come up
with actionable insights which enables bringing the
future forward .
References
i. Report by SAS, Plugging Premium Leakage, 2013 vii. Report by Insurance Research Council, Insurance
Research Council Finds That Fraud and Buildup
ii. Report by Economic Policy Institute, EPI Briefing Add Up to $7.7 Billion in Excess Payments for
Paper on (In)Dependent Contractor Auto Injury Claims, 2015
Misclassification, 2015
viii. Report by NAIC, Insurance Industry Snapshots,
iii. Survey by FICO, The three best targets for 2014
attacking P&C insurance fraud, 2010,
ix. Report by SAS and Coalition Against Insurance
iv. Report by NICB, First 3 Quarters of 2011, 2012, Fraud, The State of Insurance Fraud Technology,
2013 Questionable Claim Referral Reason 2014
Analysis, 2013, Author David Fennig, Strategic
Analyst x. Report by Coalition Against Insurance Fraud,
Journal of Insurance Fraud in America, 2015,
v. Report by NICB, Organized Group Activity in Author - James Ruotolo, Principal for Insurance
Insurance Fraud, 2012, Author - Andrea McLain, Fraud Analytics at SAS.
Strategic Analyst
vi. WNS DecisionPoint estimates based on the

report by National Association of Insurance
Commissioners (NAIC), Profitability by line by
state, 2014
Copyright notice and disclaimer:
All materials and software published on or used here are protected by copyright, and are owned or controlled by or licensed to WNS (Holdings) Limited
(WNS), or the party listed as the provider of the materials. UNAUTHORIZED COPYING, REPRODUCTION, REPUBLISHING, UPLOADING, POSTING,
TRANSMITTING OR DUPLICATING OF ANY OF THE MATERIAL IS PROHIBITED. You may use it for personal, noncommercial and informational purposes,
provided that the documents are not modified and provided you include the following copyright notice in such downloaded materials: © Copyright 2016
WNS (Holdings) Limited. All rights reserved. Some of the information contained herein is extracted from various publications and publicly available
information of other companies on their website or other resources, and WNS makes no representation as to the accuracy or completeness of the
information. WNS makes no representation that all information relating to these companies/WNS and its businesses has been included.
Image
A credible insights hub for companies looking

to transform their strategies and operations by
aligning with today s realities and tomorrow s
disruptions.
Email: perspectives@wnsdecisionpoint.com
Website: wnsdecisionpoint.com
@WNSDecisionPt WNS DecisionPoint WNS DecisionPoint

WNS DecisionPoint - Report - Fighting Insurance Fraud With Big Data Analytics

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

WNS DecisionPoint - Report - Fighting Insurance Fraud With Big Data Analytics

Încărcat de

Drepturi de autor:

Formate disponibile

INSURANCE FRAUD DETECTION AND

PREVENTION IN THE ERA OF BIG DATA

With the growing complexity of fraud, insurers need to put in

To assess the impact and adoption of Big Data analytics on

Insurance fraud is one

THE COST OF INSURANCE FRAUD

FRAUD AT CLAIMS STAGE

Fraud in Personal Lines

Estimated Annual Losses in P&C Lines due to Fraudulent Claims in 2014

Personal Lines Commercial Lines Combined Lines

Auto Liability $8,820

Auto Physical $5,158

Homeowners Multiple Peril $4,484

Workers' Compensation $3,329

Other Liability $2,833

Commercial Multiple Peril $2,120

Commercial Auto Liability $1,461

All Other $892

Medical Professional Liability $562

Accident and Health $552

Commercial Auto Physical $411

Products Liability $349

Financial Guaranty $224

Farm owners Multiple Peril $214

Allied Lines $1,057

Inland Marine $562

Mortgage Guaranty $308

U.S. Property and Casualty Insurance Key Financials

0.0 50.0 100.0

2013 6.7 60.5 28.0 95.2

2014 6.9 62.1 27.6 96.6

0.0% 5.0% 10.0% 15.0%

Change in Underwriting Profits (in $ Billion)

$0 $10 $20 $30 $40

Net Underwriting Gain/(Loss) 2013

Net Premiums Earned

Net Losses Incurred

Loss Expenses Incurred

Other Underwriting Expenses

Net Underwriting Gain/(Loss) 2014

Source: Insurance Information Institute and NAIC

Fraud Also Impacts Insurer's Ability to Offer Competitive Premiums

SHIFT IN FRAUD DETECTION

Traditional Fraud Detection Methods

Internal Audit Red Flag Indicator Scoring Models

 Random checks on sample  Rule-based indicators to detect  An advancement to rule-base

Source: WNS DecisionPointTM

AUTOMATED AND ANALYTICS-DRIVEN FRAUD

Planned Investment Areas in Fraud Detection Technology

2015-2017 46% 18% 38% 21% 16% 30% 21%

2012-2014 21% 25% 32% 12% 18% 30%

Guard your Margins by Applying Advanced Fraud Analytics over Various

Point of Sale First Notice of Loss (FNOL)

Validate a Larger Number of

Detect Abnormal Behavior Reduce

Source: WNS DecisionPointTM Analysis

Point of Sale (POS)/Renewal Stage

Fraud Detection Process at POS/Underwriting Stage

 Application  Use of single or  Risk scores are  The decision to accept or

Source: WNS DecisionPointTM Analysis

Social Media Analytics 25%

Social Network Analysis/Link

Text Mining 25%