This IBM white paper introduces the field of analytics, and discusses how analytics can be utilized in claims and benefits processing systems. It also provides an example of an advanced analytics system developed for the U.S. Social Security Administration.

1. Analytics Solutions
Advanced Analytics for Smarter Benefits, Claims,
and Entitlement Management
Toward a Smarter Government
January 2010

2. On behalf of the IBM Analytics Solution Center in Washington, D.C., we are pleased to present this white paper,
“Advanced Analytics for Smarter Benefits, Claims, and Entitlement Management.” Improving the performance of the
nation’s systems that handle benefits, claims, and similar citizen requests has been on the top of many agencies’
priority lists. Recent newspaper headlines have highlighted the need to improve these systems. Fortunately, today’s
enterprise information management systems with advanced analytical capabilities can be employed to provide faster
processing, while at the same time reducing fraud and improper processing.
This white paper introduces the field of analytics, and discusses how analytics can be utilized in claims and benefits
processing systems. It also provides an example of how such a system was developed for the U.S. Social Security
Administration.
We hope that this white paper will be useful to managers across the government as they continue to transform their
agencies to meet the needs of their constituents.
Christer Johnson Stephen Brady
Partner Sr. Technologist
North American BAO Advanced Analytics Leader Analytics Solution Center
IBM Global Business Services IBM Federal
About the Analytics Solution Center, Washington, DC
The ASC mission is to:
• Help clients understand how analytics can benefit the performance of their mission through thought leadership events.
• Demonstrate analytics solutions and technology options for solving your organization’s mission problems.
• Provide collaboration to seek and explore innovative, yet realistic, technical approaches.
• Support the development of proofs of concepts and pilot analytical solutions.
Customers, Partners and individuals are invited to join this community by visiting our web site at www.ibm.com/ascdc (IBM
DeveloperWorks registration required to join).
We provide access to a range of useful content regarding information technology that can be applied to the missions of
government departments, agencies, and organizations. Contact ASCdc@us.ibm.com to get more information or to ask us a
question via email.

3. Introduction of 97 days back to only 20 days.
Claims processing is fundamentally a customer service While processing speed and throughput are important
business, evaluated in terms of effectiveness in measures of the effectiveness of a claims handling
delivering benefits to claimants. Particularly in the system, consistency in the way claims are handled and
government domain there is a tension between the view avoidance of fraud and abuse are essential as well.
of benefit programs as entitlements and the demand of Confidence in the fairness of a government program is
the electorate that public resources not be wasted. In undermined when similar claims may be honored in
times of stress, such as natural or economic disasters, one geographic region and denied or only partially
both the number of claims to process, and the urgency granted in others. This kind of inconsistency can result
of those claims in terms of human need and suffering when human beings are forced to interpret complex
are elevated. guidelines and can not easily reference previous
experience or evolving trends.
The extreme cases of problems in claims handling are
of course the ones that rise to the level of news Sources as diverse as USA Today and the US Senate
coverage. Still when some figures come to light, such Panels on Healthcare report that as much as 10% of
as the average of 196 days required from the time a the more than a trillion dollars spent each year on
claim for veterans benefits is submitted to the Veterans healthcare in the United States may be wasted on
Administration until the final disposition of that claim, or fraud and abuse. Clearly the problem is a significant
the average of 145 days required by the US Immigration one. While organizations may devote a lot of effort to
and Naturalization Service to process an application for chasing down and recovering payments in the case of
citizenship, it is hard to argue that improvements are not fraudulent or flawed claims, better still would be to
warranted. At the other end of the spectrum, when the identify the questionable claims before payment is
Center for Medicare and Medicaid Services determines made, as part of the claims processing workflow.
how to satisfy a claim within an average of 5 days, some
could wonder whether unjustified claims are being paid The solution to smarter claims handling derives from a
without being subjected to proper scrutiny. At either combination of improved process design, application of
end of the scale, applying more intelligence to the information technology, and accommodation of the
processing of claims or requests can ensure timely people who will use the technology and conduct the
handling without compromising on verification of the process. Simplifying and facilitating data entry and use
validity of those claims. The Appendix to this paper of content management systems make unstructured
outlines a solution developed by IBM for the U.S. Social data more accessible and usable. Introducing rule-
Security Administration, using the principles described based management of the workflow associated with
in this paper which allowed the processing of certain claims processing ensures greater consistency and
classes of disability benefits to be cut from an average transparency in the handling of claims, as well as

4. contributing to more expeditious processing. And the
factor that actually makes the process “smarter” is
analytics. Analytics contribute to understanding how
the system operates, allow you to develop rules for
governing and directing the flow of claims that reflect
both organizational objectives and a deep
understanding of actual and intended outcomes, and
support continual reassessment of both internal and
external factors and their impact on the effectiveness of
the claims handling process.

5. Analytics the creation of structured metadata and indices.
The term “analytics” appeared a couple of times in the Descriptive analytics help provide an understanding of
introduction, but what does this term refer to? In fact, the past. Predictive analytics use the understanding of
what we call analytics today is really an extension and the past to make “predictions” about the future. For
aggregation of efforts that have been pursued for a example, a particular type of claim that falls into a
number of years, as people have tried to apply category that has proven troublesome in the past might
computers to help improve the quality of decisions being be flagged for closer investigation. Descriptive
made by human beings. analytics may begin by providing a very static view of
We further subdivide analytics into three categories of the past, but as more and more instances are
increasing complexity and impact: descriptive, accumulated in the experience base of the system, and
predictive, and prescriptive. with algorithms that can execute in very short periods
of time, this evaluation, classification, and
Figure 1: Analytics Landscape categorization can be performed repeatedly endowing
the overall work process with a measure of adaptability.
How can we achieve the best
As descriptive analytics reach the stage where they
Stochastic Optimization
outcome including the effects of
Prescriptive
support anticipatory action, a threshold is passed into
Optimization How can we achieve the best
Predictive modeling What will happen next if ? the area of predictive analytics. Predictive analysis
Forecasting What if these trends applies advanced modeling techniques to examine
Mission Impact
Predictive
Simulation What could happen…. ? scenarios and help detect hidden patterns in large
Alerts What actions are needed?
quantities of data to project future events. It segments
Query/drill down What exactly is the
and groups individuals to predict behavior and defines
Ad hoc reporting How many, how often, Descriptive
Standard Reporting What happened?
trends. It utilizes techniques such as clustering, expert
Degree of Complexity rules, decision trees and neural networks. Predictive
Based on: Competing on Analytics, Davenport and Harris, 2007
analysis most commonly is used to calculate potential
behavior in ways that allow you to:
• Examine time series, evaluating past data and
Descriptive analytics are probably the most commonly
trends to predict future demands (level, trend,
used and understood. These techniques deal with what
seasonality)
has happened in the past and categorize, characterize,
• Determine ‘causality’, creating models from past
and classify existing data. This includes dashboards,
demand patterns while considering other relevant
budget report, and various types of queries. These
data, to help forecast future demands (such as
techniques are most commonly applied to structured
impact on demand for replacement parts at a
data, though there have been numerous efforts to
municipal bus maintenance facility caused by
extend their reach to unstructured data, often through

6. known, predicted, or seasonal changes in consider, weigh, and trade off - scheduling or work
passenger demand,) planning problems for example. Twenty, fifteen, even
• Extract patterns from large data quantities via data ten years ago these problems could only be solved
mining, to predict non-linear behavior not easily using computers running algorithms on a particular
identifiable through other approaches. data set for hours or even days. It was not useful to
In operational terms, predictive analysis may be applied embed such problem solving into a decision support
as a guide to answer questions such as: “Who are my system since it could not provide timely results. Now,
customers and what is the best way to target them?”; however, with improvements in the speed and memory
“Which patients are most likely to respond to a given size of computers, similar computations can be
treatment?” or “Is this application likely to be performed in minutes. While this kind of information
fraudulent?” might have been used in the past to shape policy and
offer guidance on action in a class of situations,
It is at this level that the term “advanced analytics” is assessments can now be completed in real time to
more aptly applied. Included are techniques for support decisions to modify actions, assign resources
predictive modeling and simulation as well as and so on.
forecasting. In simulation, you are creating a model of An example of these sorts of decision support tools
the system and inferring from the model what the comes from the U.S. Postal Service and involves the
behavior of the actual system will be. This requires assignment of mail to commercial air carriers. Years
being able to build algorithms or mathematical ago, all the carriers were viewed as interchangeable.
constructs that provide a sufficiently accurate However, once it was possible to scan mail and track
representation of the observable behavior of a system. how the mail was delivered and which carriers were
This in turn can be used to evaluate proposed changes involved, it was possible to differentiate service by
to a system before they are implemented, thus carrier and to modify the assignments so they went to
minimizing cost and risk. Much of business process better-performing carriers. This produced a
modeling falls into this category. Forecasting can be measurable improvement in overall delivery speed
applied in lots of way, not the least of which is predicting which probably resulted not only from better choices of
workload, often translated into resources required, carriers, but also from the incentive created by adding
including human resources. this factor into the assignment process.
Once you understand the past and can begin to predict
what might happen in the future, you can begin to think
about what the best response or action will be. This is
the area of prescriptive analytics. There are many
problems that simply involve too many choices or
alternatives for a human decision maker to effectively

7. Building Smarter Claims of subsequent efforts, such as process bottlenecks,
Handling processing inconsistencies, and obstacles to reaching
appropriate outcomes. Often the desired behavior can
To this point the discussion has been about the kinds of be codified in the form of policy which can then be
analytic tools that can be brought to bear on the instantiated in a set of rules to govern claims
problem of improving a claims handling process. In this processing. The entire system will be linked together
section we discuss what a process might look like that under the control of a workflow management system
would produce a system that is shaped and guided by directed by a rules management system, even while
these analytic tools. the separate underlying applications are preserved.
This addresses issues both of consistency in claims
Figure 2 illustrates a typical claims process, consisting handling and process transparency, and
of a number of separate applications and steps of accommodates regulations for auditing and oversight.
gathering and evaluating claims information and then Alternatively, the characterization of the process can be
effecting the actions required to carry out any decisions used to construct a model of the claim handling system
made as part of that evaluation. which can itself be the object of study as a way of
identifying potential for process improvement.
The process can then be further modified by applying
Figure 2: Basic Benefits/Claims Management Process
External
Input
intelligence to the information gathering stage.
Informed by historical data, customer facing
applications can be created that employ text analytics
Data and contextual information to ensure that claims are
repository
Intake Eligibility Evidence Approvals Payments Verificat
of decisions represented consistently and correctly within the
ion
handling system, contributing even more to consistency
of processing outcomes. Claims handling policy may
allow for expedited handling of some claims and
require closer scrutiny for others.
The first step in trying to improve such a process is to
analyze the store of historical data produced as a
Even within a more automated claims handling
byproduct of the operation of the process. Here, all of
process, a requirement for human evaluation and
the data analysis and data mining tools can be brought
assessment will not disappear. Still not all humans are
into play to build a detailed picture of the process,
equal in their experience and areas of competency.
including benchmarks on performance and identification
Ensuring that tasks are assigned to appropriate staff
of parameters and correlations between parameters that
members for handling is a classic workforce
affect the eventual dispensation of any claim. This may
management and scheduling problem, and hence
reveal any number of features that might be the targets

8. susceptible to mathematical optimization techniques. statistical analysis on past claim data to regularly
This area in particular has shown significant progress balance their desire for speedy processing with their
with the increase in available computing power and appetite for risk and their overall financial exposure. In
advances in the necessary algorithms. other words, these companies use past data to
Continued collection of outcome data, particularly in estimate the risk associated with each claim or
cases of fraud and abuse allows the cycle to be application and then use statistical models to determine
completed and sets the stage for continual improvement how to handle each claim or application. In some
in the claims handling process. The ease with which cases, they may automatically approve a claim and in
rule driven systems can be updated enhances the ability others, the will delay for extensive verification.
of the system to react to changes in behavior of those
who use the system and to respond when external A smarter benefits processing system is illustrated in
circumstances produce a surge in the number or a shift Figure 3.
in the type of claims that are being received.
In many cases, the key to better leveraging advanced Figure 3. Smarter Benefits/Claims Processing System
analytics for claims processing is to change the overall
philosophy from one that is based on a manual review
of each claim independently from others to one that
uses rigorous statistical analysis of past claims and
decisions to help guide the decision process on future
claims and begin accelerated payments for selected
(“low risk”) types of claims. Such a change in
philosophy should start with rigorous analysis of the
claim and decision data from previous years. The
purpose of this analysis would be to understand and
identify the most common decision or reimbursement
percentage associated with each claim type. Inherent in
the value of analyzing past data on claims and payment
decisions is the assumption that the organization would
be able to use such statistical analysis to justify
automated pre-payments for certain categories of
claims. This is not different in principle than what most
insurance companies and financial institutions do with
their “scoring models” for claims or credit card/loan
applications today. These companies conduct rigorous

9. Conclusion
Benefits processing always involves a tension between
the desire to provide prompt relief from a claimant’s loss
or expenses and the need to manage financial
resources responsibly and to avoid paying fraudulent
claims. When the balance moves too far in the latter
direction, claims processing times can become
unconscionably long. In this paper we have discussed
how the application of advanced analytics techniques
can be used to transform a benefits processing system
so that it exploits historical information to help develop
claims handling policy which can then be embodied in a
rules-driven workflow. When this is combined with tools
that improve the quality of claims information gathered
from customers, visualization capabilities to provide
transparency into the operation of the system, and
optimization techniques for fine-tuning the assignment
of work to human evaluators, the result is smarter
claims handling. Technology is an enabler, but in the
end the impact of any of these changes to the claims
handling process is limited by the ability of organizations
to modify the overall business process, the way people
work and the way they think about the work they are
doing. In the benefits and claims handling area, the
effects of the process on the end-customers, i.e., the
citizens in a benefits processing system should never be
ignored.

11. Appendix
Case Study: Predictive Modeling for Disability Claims
Client: Social Security Administration (SSA)
Industry: Government, Social Services
Challenge: A recent challenge faced by SSA focused on the both the time required for
reviewing and approving disability benefits for disabled citizens and the
consistency of these determinations across the country. The backlog generated
had come to attention of Federal oversight organizations such as the General
Accountability Office, which only added to the urgency on the part of the SSA.
Solution: Extending prior work at SSA with structured data mining tools such as logistic
regression, IBM expanded its suite of predictive models to include advanced text
analytics to infer meaning from unstructured data in disability applications so that
SSA could automatically score applications to identify potential quick decisions.
Benefits: The predictive model for reviewing new disability applications has reduced the
average cycle time for approving an application from 90 days to 20 days for
selected cases and continues to drive the agency toward higher levels of
consistency across the system. The Quick Disability Determination (QDD)
process has been featured in Congressional testimony and numerous press
releases from the SSA as one of its most successful programs.

12. © Copyright IBM Corporation 2010
IBM Corporation Route 100 Somers, NY 10589
U.S.A.
Produced in the United States of America 01-10 All Rights Reserved
IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States,
other countries, or both. If these and other IBM trademarked terms are marked on their first occurrence in this information with a trademark symbol
(® or ™), these symbols indicate U.S. registered or common law trademarks owned by IBM at the time this information was published. Such
trademarks may also be registered or common law trademarks in other countries. A current list of IBM trademarks is available on the Web at
“Copyright and trademark information” at ibm.com/legal/copytrade.shtml
Other company, product, or service names may be trademarks or service marks of others.
References in this publication to IBM products or services do not imply that IBM intends to make them available in all countries in which IBM
operates.
GQW03001-USEN-00