Hi, I’m James Ruotolo, the Principal for
Insurance Fraud Solutions at SAS. Today I’m
going to talk about the value of using a hybrid
approach for insurance fraud detection.
Losses due to insurance fraud are on the rise.
Increasing economic pressures and a growing
wave of organized fraud rings make fraud detection
a top priority for insurance companies worldwide.
Traditionally, insurers have relied upon claim
adjusters to manually identify “red flags”
to detect fraud. More advanced companies have
automated this process or implemented predictive
models to detect suspicious claims. For most
organizations however, current fraud detection
methods are fraught with manual procedures,
large numbers of false positives and a failure
to integrate siloed data from various business
units or incorporate unstructured text data.
The key to effective fraud mitigation is combining
detection methods that leverage data points
from across the enterprise and constantly
add new intelligence back into the system.
At SAS, we’ve incorporated this concept
into our Fraud Framework solution with a hybrid
approach to insurance fraud detection.
Most insurers have information silos populated
by various transactional systems. For example,
an insurance company may have data on policies,
claims and vendor payments . Other sources,
like human resources systems or external third
party data could also be used.
With a Framework-based approach, the data
from each of these systems is aggregated into
an “Intelligent Data Repository”. It is
important to note that unstructured text data
needs to be included here. Some studies suggest
that upwards of 80% of insurer data in a text
format. In our experience at SAS, anywhere
from one-third to one-half of the components
in a good insurance fraud detection solution
are based on unstructured sources like claim
notes or customer service logs.
With all the source information aggregated,
claims are scored using a combination of technologies.
First, business rules can be used to flag
certain behaviors. For example, a business
rule could flag any claim filed within 30
days of policy inception, a common indicator
that a more detailed claim review may be required.
Next, Anomaly detection is used to reveal
abnormal patterns of activity compared to
a peer group. With all of the data aggregated
in the Intelligent Data Repository, this type
of analysis is now easy to conduct. For example,
anomaly detection might flag a claim with
a high ratio of bodily injury exposure compared
to the physical damage sustained in an auto
accident. This can be an indicator of staged
accident activity. Anomaly detection can also
incorporate “watch lists” to flag known
suspicious entities.
Then, Predictive Models can be used to identify
complex fraud patterns. For example, a predictive
model might identify suspicious accident indicators
based on prior investigative experience that
would otherwise have been difficult or impossible
to detect.
Finally, Social Network Analysis is used to
identify organized ring activity. Networks
are built automatically based on connections
made within the data. These connections often
go unnoticed by adjusters who typically don’t
have the luxury of analyzing aggregated data
in this fashion. For SAS customers, this has
shown results of up to 10 times more fraud
detected, and $50 Million in annual benefit
to tier 1 insurers.
While each of these detection systems is independently
powerful, the real value of a framework approach
is demonstrated by this next step.
All of these components work in concert and
are managed by an Alert Generation Process,
or “AGP”. The AGP is what generates a
fraud risk score for each claim. But instead
of relying simply on the details about a single
claim, the framework is able to evaluate other
claims in the same peer group or within the
same network and influence the fraud risk
score accordingly.
Let’s look at an example. If a claim is
flagged by a business rule, it might get a
moderate risk score. But if other claims within
the same network – say, involving the same
participants – also trigger fraud business
rules, the fraud risk score for that claim
would be increased.
The fraud risk score and relevant information
about the claim is then transferred to an
Enterprise Case Management System which can
be used by a Special Investigation Unit to
review alerts and track the progress of investigations.
A case management system should house all
investigative information including notes,
evidence and investigation outcomes. This
type of consolidation has been shown to improve
investigator efficiency by more than 25%.
The information from the Enterprise Case Management
system then becomes a new source for the Intelligent
Data Repository. As investigations are conducted,
the system learns about fraud patterns and
suspicious entities and can use that information
to improve the alert generation process for
future claims.
Good fraud mitigation capability reduces loss
costs and should be part of every insurer’s
claim management strategy. A hybrid approach
to insurance fraud detection reduces false
positives, uncovers both opportunistic and
organized fraud activity and improves the
effectiveness of a Special Investigation Unit.
To learn more about the SAS Fraud Framework
for Insurance go to sas.com/insurancefraud.
I’m James Ruotolo, the Principal for Insurance
Fraud Solutions at SAS. Thanks for watching.