Expert Systems With Applications, Vol. 5, pp. 15-24, 1992
0957-4174/92 $5.00 + .00
© 1992 Pergamon Press Ltd.
Printed in the USA.
The Strategic Use of Expert Systems for Risk Management
in the Insurance Industry
MARC H. MEYER
Northeastern University, Boston, MA
ARTHUR DETORE
Lincoln National Risk Management, Inc., Fort Wayne, IN
STEPHEN F. SIEGEL
Fusion Systems Group, Ltd., New York, NY
KATHLEEN F. CURLEY
Northeastern University,Boston,MA
Abstract-- The applications focus and development history of two large expert systems for underwriting
life insurance cases are used to consider the linkage between business strategy and project focus for
major information systems development. It is shown that the project selection and evolution processes
of these two companies have resulted in two distinctly different expert system applications, the product
of their respective business positioning. The article also presents a detailed description of how expert
systems can be used to enhance the productivity and effectiveness of risk management in the life
insurance business.
Similarly, Meyer and Curley (1991) use a case study
of a life insurance underwriting system to illustrate a
conceptual framework for managing expert systems
development by assessing embodied knowledge and
technological complexity. DeTore (1989) described the
suitability of expert systems for the theoretical and
procedural decision processes in insurance. Hammer
(1990) focused on the opportunities for substantial
work redesign and productivity enhancement with new
information technologies. May (I 990), Spilberg et al.
(1987), and Baker (1987) provide more general descriptions of major trends in the application of new
information technologies to insurance. Additionally,
industry members now meet annually at the Expert
Systems in Insurance Conference to discuss expert system issues (IBC USA Conferences, 1990).
The ability to direct a company's energies to the
most appropriate technological efforts is an essential
part of "the art" of productive innovation. Whether it
is the result of the intuitive insight of a powerful
"'champion," or that of an effective strategic planning
process, the ability to focus a firm's resources on new
technological endeavors that "matter most" can have
a powerful competitive impact. A large number of papers have reported the use of information technologies
1. INTRODUCTION
THE PURPOSE of this paper is to consider the question
of the linkage between business strategy and the development of major expert systems in the context of
project selection and evolution. To explore this issue,
we will compare two risk management expert systems
for underwriting life insurance applications. Both systems are "successful" large-scale projects with different
loci in the workflow chain within the insurance business. These foci were the result of deliberate decision
making by senior management that was based on an
assessment of traditional corporate strengths and business positioning. In other words, each respective target
is highly consistent with overriding corporate strategy
and internal strengths of the two respective companies.
Expert systems in insurance is a topic that has been
explored by a number of authors in fairly recent publications. Davenport (1989), Jones (1987), and Conversano (1988) all presented case study descriptions of
expert systems applications for insurance underwriting.
Requests for reprints should be sent to Marc H. Meyer, PhD,
2 ! 3D Hayden Hall, Northeastern University, 360 Huntington Avenue, Boston, MA 02115.
15
16
M. H. Meyer et al.
as a competitive weapon (Bakos & Treacy, 1986; Benjamin, Rockart, Scott Morton, & Wyman, 1984; Ives
& Learmonth, 1984; Johnston & Vitale, 1988). In the
domain of expert systems technology, project selection
has been considered from several basic perspectives.
Davis (1984), Prerau (1985), and Leonard-Barton and
Sviokla (1988) have all observed the linkage between
existing competitiveness in decision-making and the
need for and subsequent application of expert systems
technology. Feigenbaum et al. (1988) showed how a
number of large corporations leveraged key human decision-making assets through expert systems to become
more responsive and productive as companies. Meyer
and Curley (1991) described how the accumulated industry insight of senior management makes its participation in the initial systems planning of major strategic
expert systems applications important. The general
thrust of all these works is that major expert system
applications are a direct outgrowth of business strategy,
enhancing existing business processes and potentially
creating new business value by utilizing the firm's core
"knowledge" assets.
What does this mean from the business manager's
perspective? How can the linkage between business
strategy and the applications of expert systems technology to problem solving be operationalized at a level
of detail that is useful for choosing between alternative
paths in developing large, costly expert systems?
Clearly, parameters such as competitive positioning,
the importance of expert system application targets
within core business processes, internal technical and
"domain" resources, and competitor's technological
endeavors are factors to be taken into account either
eXpllotly or implicitly by management. In this paper,
we will see how these factors were considered by two
companies within the same industry: life insurance.
Each firm is positioned at a different end of the business. John Hancock, is among the top ten "primary"
or direct insurers in the country; Lincoln National
Reinsurance Company is one of the largest life-health
reinsurers in the world. The result of their respective
business assessment, technology scanning, and project
selection processes has produced two very powerful and
distinctly different expert system applications.
/
.
.
.
.
.
.
2. RISK M A N A G E M E N T A N D THE
INSURANCE WORK FLOW
To understand what the expert systems of these two
insurance companies seek to accomplish, and more
specifically, where these applications fit within the insurance process, an understanding of "risk management" in insurance is useful. A description of risk
management is provided in the following paragraphs.
Risk is the possibility of loss or injury and is an
element of everyday life that cannot be avoided, be it
the loss of life, health, or property (Shepard & Webster,
1957). Insurance is the most common way that individuals and organizations protect themselves from certain well-defined potential financial losses. For a risk
to be insurable, however, the element of chance must
be present wherein the loss is caused by an uncertain
future event. The ability to effectively manage risk lies
at the heart of insurance companies. It directly translates into the margin between net premiums taken into
a firm, and claims paid out. Understanding the riskiness
of new potential business for given insurance products
also allows a company to fine-tune its pricing. Thus,
improving the management of risk can enhance an
insurer's competitiveness and profitability. Herein lies
the great opportunity for expert systems for the insurance industry.
Insurers have developed substantial expertise in the
management of risk. The insurer must make complex
decisions about the selection, pricing, and claims adjudication of insured risks. At the core of these processes, is underwriting. Underwriting operationalizes
a company's various approaches to managing risk
(Bailey, 1985). Until recent years, these decisions could
only be made by individuals with considerable skill
and experience. Today, however, these decision processes can be programmed, automating a certain percentage of underwriting decisions and providing underwriting assistance to human decision-makers for
others.
Underwriting is an information intensive process in
which decisions on individual risks are made in the
context of actuarial and business principles. At the most
general level, underwriting involves the examination
of a specific risk, classifying it into a larger group of
similar risks, and then calculating the financial impact
of the potential loss. For life insurance, these risks take
many forms: health problems (known as medical impairments), financial problems, and both occupational
and vocational hazards. Risk assessment decisions must
be made in the context of a competitive market environment, balancing the expected loss with the expected
investment income from the use of the pooled funds.
Often, the underwriting decision is based upon incomplete and uncertain information. In life insurance, this
could be observed physical symptoms that might be
indicators of a range of potentially fatal medical conditions. The quality, and more specifically, the correctness of individual decisions and the consistency of
decisions for groups of similar cases, has a direct, longterm impact on the insurer's profitability because,
within the industry, the proportion of premiums paid
for death claims is greater than that paid out to cover
the insurer's administrative costs. Consistently good
underwriting makes a company competitive; consistently poor underwriting will lead to a level of claims
that, in a period of consolidation within the industry,
will cause severe business damage.
Figure 1 shows the typical information flow through
Strategic Use of Expert Systems
17
an insurance company. Many of the information processes shown in that figure are candidates for the applications of expert systems technology. The first step
in the information flow is that agents, known in the
industry as "producers," sell life insurance policies and
complete with the applicant the standard form of the
insurance company. This information passes into
"Preliminary Processing," where routine details about
the risk from insurance application forms are entered
into a computer system. Screening may then occur (often called "Jet Screening"), where applications are
checked for completeness, and a certain percentage
"accepted" or "declined" based on simple company
rules concerning acceptable and uninsurable risks. The
Data Entry and Initial Underwriting processes can collectively be referred to as the "front end" of the insurers'
new business processing. A current trend in the industry
is to allow Data Entry and Screening to occur at agent's
offices, reducing the insurer's home office clerical requirements and improving response time to the customer, and hence, competitiveness. Additionally, insurers working with expert systems technology are
seeking to build increasing amounts of intelligence into
their Jet Screening systems. The benefits are substantial.
Large insurers, for example, can process well over
100,000 applications per year. Automatic underwriting,
combined with increased productivity in human underwriting through computer assistance, can yield savings in the order of millions of dollars per year.
Applications not accepted enter into the "back end"
of the process, called "Underwriting" in Figure 1. Specific "problems" with a case are first identified, and
Sales
L..
I"
~[ Market
--I R..... h
Investments
I Applications
I
Entry
I DN:vWel°P~m°del
/ cl
t
I
Preliminary
Processing
I
[
\
Administration
',
Adjudication
/
]
]
FIGURE 1. Insurance information flow and expert system
opportunities.
~Ofl~lCe
Cllentll~llev
j
Seine are Agents Systems)
I
Screening
lflmlnlmlwlltlGa
Systems
Decision Outputs
.
1[
~
,nitin,
Underwriting
Information
.
F ' I , Underwriting
In~
AlphaFiles
Medical Examinations
HistoricalFiles Financial Information
Productsand Pricing
FIGURE
2. Information
linkages
in the
new
business
processing.
the information needed in part or in whole to resolve
these problems is then ordered, either by computer,
telephone, or mail. Many insurers refer to this collection of steps as "Initial Underwriting" (see Fig. 2). An
important aspect of underwriting is, therefore, deciding
what types of information are necessary to resolve a
problem. Ordering and receipt of "requirements" can
become a bottleneck in the process of evaluating new
cases; managing this process becomes an essential part
of expert systems assisting in the Initial Underwriting
process. It should be noted that within "Initial Underwriting" a company can include highly complex heuristics for decision-making. These include analysis of
occupations, hobbies, certain medical conditions,
family history, and personal "build."
After "Initial Underwriting," cases not accepted pass
into the stage of underwriting that focuses on the investigation of medical problems, perhaps the most
complex heuristically in terms of procedural and theoretical underwriting knowledge. Insurers generally refer
to this stage as Impairment Underwriting (see Fig. 2).
Medical examinations, including tests such as EKGs,
are often studied by underwriters to resolve "problems." Effective underwriting may also require the assessment of interactions between individual impairments, such as the increased mortality probability of
a moderate case of diabetes and a history of coronary
heart disease. Many diseases have levels of severity,
which, once resolved, translate directly into different
levels of risk, and hence, insurance premiums for those
cases not rejected. Problem resolution then entails rejection of the application or acceptance, either at
"Standard" rates, or at surcharge over standard, based
on a mapping of the level of nonstandard risk to increased premium rates. This mapping is insurer-specific, reflecting its philosophy in terms of risk management and product pricing. Underwriters must therefore
consider the specific insurance product being applied
for. Note that improved underwriting for these problem
18
cases can also have a dramatic impact on an insurer's
profitability, albeit longer term than productivity improvements for automated underwriting. For example,
large insurers each pay hundreds of millions of dollars
in death benefits each year (AM Best Company, 1990).
Even a 1% or 2% improvement in experienced mortality (which is expected by Lincoln National from the
use of its underwriting expert system) can lead to substantial savings.
To complete the work flow in Figure 1, accepted
applications are then sent onto the company's administrative systems, "Client/Policy Issue and Administration" in Figure 1, which deal with customer billing
and agent commissions. "Claims" comes into effect
upon a policyholder's death. Claims are adjudicated,
and then studied in statistical fashion to examine the
effectiveness of product pricing and underwriting. Information from Claims is employed by an insurer's
actuaries for research and insurance product development or pricing changes. The three work-flow boxes
in Figure 1 that represent these activities are "Claims
Analysis," "Pricing," and "New Product Development." "Market Research," "Investments," and "Performance Review" comprise the remaining aspects of
the insurance work flow. While we are writing here
about the underwriting expert systems, in the future,
one will probably find applications of expert systems
technology to insurance in Sales, Claims Adjudication
and Claims Analysis (often called Comparative Analysis), Investments, Marketing Research, Product Development and Pricing, and for financial reporting with
Performance Review. These various linkages show that
whatever expert systems a company creates for any of
the respective work-flow modules, the knowledge bases
of these systems must reflect and be updated with information from one or several other operations in the
company.
Also included in the insurer's work flow is reinsurance. Every insurer establishes its own limits on face
amounts for individual life applications, be it $50,000
to several million dollars. Applications requesting
greater than these amounts are not uncommon. In such
cases, the primary insurer seeks to limit its own financial volatility by sharing part of the risk with its reinsurer. Reinsurers therefore provide insurance of sorts
for insurance companies. Dealing with large policies,
the reinsurer's business success depends greatly on its
risk management acumen, a significant part of which
is its own underwriting and actuarial expertise.
Lastly, just as the underwriting process is linked to
many information sources both within and external to
an insurance company, a comprehensive underwriting
expert system must maintain linkages with many other
distinct information systems. The major information
linkages in life underwriting can be generally conceptualized as shown in Figure 2. At the front end of the
process is the integration of the underwriting expert
M. H. Meyer et al.
system with the company's Data Entry system. In the
future, it is expected that the entry of new applications
information will be increasingly distributed to producers in the field (Life Office Management Associations
1989), making the systems integration task all the more
complex. Then, during Screening, the expert system
must access the insurer's Alpha indices (to see the proposed insured's history with the company), and other
specific historical databases. In the more extensive underwriting stages, interfaces are maintained with providers of medical examinations. At the back end of the
underwriting process, the expert system must also be
integrated with the company's product pricing databases, and ultimately, with policy issue and administration systems. Therefore, computerized risk management in insurance, if taken to its logical conclusion,
entails high degrees of systems integration and database
administration.
There is a clear strategic consequence to expert systems in the insurance industry. Faced with increasing
competitive pressures, insurance company managers,
be it in life, property and casualty, or health businesses,
know that their companies need to become more efficient and effective in key areas. These include the
selling products, providing services to customers, and
administering current business. It is recognized that a
significant part of meeting this challenge lies in the
combination of applying new technologies such as expert systems and using organizational and job redesign
techniques to achieve organizational flattening. This is
illustrated in the most general way in Figure 3, which
shows where most insurance companies stand today,
and where they wish to be in the not too distant future.
Operating expenses tend to be higher in the two middle
layers, Applications Entry and Preliminary Screening,
as compared to Sales and Underwriting. Technology
may therefore provide a way to not only substantially
reduce expenses, but also to greatly improve service to
the customer. However, this potential cannot be realized without immense effort in the areas of workflow
reengineering and large-scale organizational change.
3. AUTOMATED UNDERWRITING
AT J O H N HANCOCK
The two cases to be presented in the remainder of this
paper focus on the technological part of the challenge
discussed above. In describing these expert systems applications, we will use the histories of their development
to highlight some of the critical management decisions
that were conducive to successful outcomes.
The first case comes from John Hancock Mutual
Life Insurance Company. John Hancock processes a
very large volume of new insurance applications, in
excess of several hundred thousand cases each year.
The information from these cases comes from over
300 field offices maintained across the country. Un-
Strategic Use of Expert Systems
19
TODAY
Sales
II
Applications
Entry
II
I
Preliminary
Processing
Underwriting
FUTURE
I Producers
Spreadsheet
SalesTools
Automated Underwriting
(Reaching 50%)
i LargeDataEnvy
OftenStaffs
Redundant
EntryPoints
.i
i
I~les
Producers
Expert Systems
!
z
I
O
I
HomeOffice
SomeUnderwriting
Automation
Jet Scl'eeflefs and
HomeOffice
II SomeAutomauxl
Undenvrite~
~J~quirementsOrdering
r~
o
.omeo°ce
Underwriters
Expert Systems
Underwriting Decision Assist
Automated Requirements Ordering
FIGURE 3. The potential of expert systems to change work flew, enhance productivity, and improve service insurance processing
as an example.
derwriting had always been performed at corporate
headquarters. In 1986, an internal technology consultant in Corporate DP who was scanning a variety of
new technologies as they might be applied to insurance,
generated a game plan for applying expert systems. The
concept was appealing to senior management, and the
Underwriting Department agreed to both fund and
contribute its domain expertise to the effort. Four years
later, with the participation of five experienced underwriters and about a dozen computer programmers from
Corporate DP, the expert system is now being distributed to the field. The applications from over 170 field
offices are processed through the expert system and the
remaining offices will be provided with the technology
over the coming months. Thus, the system will be fully
installed across the John Hancock organization by the
end of 1991.
The very nature of the John Hancock's business
dictated a "front-end" focus for initial expert systems
development efforts; that is, to automatically approve
as many new cases as possible without human intervention. The nature of John Hancock's "retail" life
insurance business (i.e., the large volumes of new applications), was therefore highly amenable to a production-oriented underwriting system. In fact, it was
a greater priority to provide broader underwriting
knowledge to more cases than deeper underwriting
knowledge to fewer cases.
The initial version of the system, or Stage 1 of development targeted Intelligent Data Entry and Prelim-
inary Processing. This was achieved in two parts: a
data entry and preliminary processing system that
functioned on PCs in agent offices, and a "batch" automated expert system running on a mainframe in the
home office which, using a specific set of heuristics or
decision-making rules, can automatically approve a
large percentage of new cases by examining applicant's
occupations, financial issues, and certain medical conditions. In fact, for the current field office installations
of the expert system, the percentage of automatically
approved applications is approximately 33%.
In a subsequent stage of development, the system is
evolving towards the back end of underwriting by
computerizing more complex aspects of decisionmaking. Figure 4 illustrates the expert system's higher
level architecture. The knowledge bases in John Hancock's "Automated Underwriting" modules define
underwriting problems. Checks are made for additional
information requirements depending on policy
amount. The applicant's "insurable interest" is assessed, as are the ability to pay premiums, the riskiness
of his avocations and vocations, family history, and
certain medical conditions. New information requirements can also be ordered automatically at this point.
The collective functionality of these two modules target
the "Initial Underwriting" processes shown before in
Figure 2.
Cases that are not approved by the expert system
are available to human underwriters through an "Electronic Worksheet" the next day. Underwriters can then
20
M. H. Meyer et al.
Electronic Life Underwriting flystem
Field Office PC's
Home Office Mainframes
Automated
Underwriting
Intelligent
Data Entry
~
-
-
-
UW Requirements I
Processing [
p
Preliminary
Heuristic
Screening
Home Office
Underwriters
f
Medical Impairment
and Other Complex
Underwriting
cYYYMYMYy:eYY)'y~"
Electronic
Worksheet
I
Home Office
Mainframes
Administration Systems
for Policy Issue
Incremental Percenta2e
Automated
Underwriting
33%
of Cases Comnleted
Electronic Worksheet
Human Assisted
Underwriting
~
v
10%
Human Underwriting
with Electronic
Assistance
b~
y
57%
FIGURE 4. John Hancock's underwriting system.
approve these cases, or order additional information
requirements through the Worksheet. An additional
percentage of new cases can be approved for issue at
this point, still within the following day after initial
data entry. Currently, that percentage is approximately
10%. For example, if a proposed insured admits to a
minor illness 2 years ago, the system might automatically approve the case. A serious illness would mean
that the case would be passed onto the Underwriting
Department for human decision-making. Speed is an
important benefit: all new cases are reviewed by the
expert system within 24 hours, wherein 43% of these
cases are approved either automatically by the expert
system, or with underwriter assistance in the Electronic
Worksheet.
Standing between the field office PCs and the Home
Office mainframes are a series of modules for exchanging information from Intelligent Data Entry to the Expert Underwriting and Electronic Worksheet modules,
as well as programs for ordering information requirements, and for sending completed applications to John
Hancock's Issue and Policy Administration systems.
These data communications modules are complex, and
given the number of field offices involved, as well as
the high reliability required for transmissions, the development effort entailed substantial levels of systems
integration and other "non-AI" programming. This has
been the role of approximately a dozen DP programmers who have been assigned to the project.
A number of both specific and broader benefits are
being derived from John Hancock's expert system. A
43% approval rate for new applications within 24 hours
of initial data entry provides substantial expense reduction over noncomputerized new business processing. The ability of field offices to issue new policies so
quickly for a large number of new applications also
increases the attractiveness of John Hancock's insurance offerings, leading to new business at the retail level
that might have otherwise not been realized. "Next
day" notification on the status of new cases is provided
to field offices. Lastly, the productivity of human underwriters is enhanced because they are now able to
concentrate on complex cases filtered through Automated Underwriting. In short, this expert system exhibits a direct linkage to the firm's competitive advantage.
4. C O M P R E H E N S I V E U N D E R W R I T I N G
AT L I N C O L N N A T I O N A L
The second case involves the computerization of the
more complex elements of the life underwriting process. Lincoln National's system started out in 1986 with
a medical director in the Reinsurance Division prototyping the underwriting process for the asthma medical impairment using a simple PC expert system shell.
Like all medical directors, his job had been to advise
underwriters on risk parameters for a multitude of
Strategic Use of Expert Systems
medical impairments, synthesizing existing medical
research with a firm's mortality experience. Senior
management soon believed that his prototype was the
beginning of something much grander in scale, a technology that could have benefits not only for the reinsurance underwriting department, but also for the industry as a whole. In a decisive and rapid manner, the
company's best underwriting and computer experts
were asked to participate in extending the prototype
into a set of formal products and services to be marketed to the industry.
Understanding the complex aspects of risk management in life-health insurance had been one of the
company's recognized strengths. In fact, compared to
direct insurers, the quality and consistency of underwriting has an even heavier impact on a reinsurer's
competitiveness. Reinsurers assume larger at-risk
amounts and the proportion of premiums received that
are paid out in death benefits (i.e., mortality costs) are
much higher than in a direct insurance operation. Thus,
a successful reinsurer needs strong underwriting skills,
and stands to gain if it can leverage these skills with its
customers. Lincoln National, for example, has maintained an underwriting research and development department generally considered to be one of the most
advanced in the industry. The company's in-house
medical department also offers a substantial information bank of medical knowledge. In a market that is
extremely price-competitive due to overcapacity, Lincoln National has been able to differentiate itself on
the basis of value-added services to its reinsurance
clients. These services have included seminars, newsletters, underwriting training programs, and telephone
hotlines for clients. The underwriting research and development department has published an underwriting
impairment guide manual, containing information for
hundreds of specific medical impairments and guidelines for assessing financial and nonmedical risks. Lincoln National's expert system was destined to be the
"next generation" of these collective services.
Senior management participation in strategic technology applications is often recognized as a critical
success factor for a variety of reasons: it can provide
appropriate project focus, financial support, and linkages across the organization for gaining key individuals
required for development and implementation of
completed systems. At Lincoln National, the top management was heavily involved in both the higher level
systems objective setting and organization planning for
the expert system project. A steering committee convened regularly to map out "the future" of life insurance work flows, and how technology could enable the
placement of information, responsibility, and decisionmaking at key "points of events," be they the selling
of insurance products or the underwriting of new business.
There were high expectations. While Lincoln National's underwriting manual delivered "procedural"
21
knowledge for life underwriting, management sought
a way to also provide the more theoretical medical and
insurance knowledge that senior underwriters used in
handling difficult cases. Expert systems technology was
seen as the vehicle for achieving this. It could combine
both the procedural knowledge of the underwriter with
the theoretical inferencing of medical directors and actuaries-in a system that was highly interactive, flexible, and thus, usable by both senior and junior underwriters. The reinsurance underwriters could certainly benefit from such a system in their own work.
For Lincoln National's clients, an underwriting expert
system could provide substantial work-flow and productivity benefits as described above. Advanced underwriting expert modules could also reduce mortality
costs over the long term. A comprehensive system
might also help Lincoln National to maintain its qualitative edge over other reinsurance competition. For
all these reasons, the expert system was seen from the
beginning as a "strategic investment."
Given the nature of Lincoln National's business,
and its existing underwriting resources, it was therefore
consistent that Lincoln National started by focusing
on the back end of the underwriting process; that is,
developing knowledge bases for medical impairments.
This was the type of underwriting in which the reinsurance underwriters had always excelled. To acquire
their expertise, a formalized knowledge engineering
process was created early in the project. The process
relies heavily on group meetings attended by medical
directors, reinsurance underwriters, underwriters from
Lincoln National's "direct" life business division, and
systems personnel. This process facilitated a synthesis
of the descriptive and procedural knowledge of the
company's senior underwriters with prescriptive and
theoretical knowledge of its medical directors and actuaries (Cebul, 1988). Great care was also taken to
make the deductive reasoning in the system emulate
an underwriters cognitive processes; that is, to be consistent with the underwriter's "approach" to resolving
problems. Over time, an increasing number of medical
impairments were computerized as well as other expert
modules that assess interaction effects between impairments. The results of Stage 1, wherein six major
medical impairments were developed, convinced senior
management to allocate greater resources to the project.
Additionally, all the other modules to be described below and shown in Figure 5 had already been designed
and initially implemented by the end of Stage l, and
shown to senior management.
In 1988, Stage 2 of development proceeded. The
"team" now consisted of the medical director as the
knowledge engineer, six senior underwriters as domain
experts, and an equal number of programmers. The
medical director was clearly a "hybrid" individual. In
addition to having domain expertise, he had also been
a student of knowledge engineering, having taken
courses in "Medical Informatics" while attending
22
I
M. H. Meyer et al.
Data Entry
I
"'I
RequirementSprocessing
....
Simple
Screening
I
~Systems lnterface~
Initial
] Underwriting
Underwriting
I¸
FIGURE 5. Lincoln National's Life underwriting system.
medical school. His medical expertise has been an important element of his effectiveness as a knowledge engineer given the depth and complexity of the underwriting decision-making embodied in the expert system. The decision was also made to employ expert
systems development tools that provided portability
between PC and mainframe environments. The expert
system evolved into earlier parts of the underwriting
chain, targeting Initial Underwriting processes. Lincoln
National's Initial Underwriting module incorporates
basic underwriting heuristics for assessing occupations,
hobbies, certain medical conditions, family history, and
personal "build." Over time, however, Lincoln National has been able to move certain aspects of complex
medical underwriting earlier into the work flow. If there
is a history of a serious life-threatening disease, the
system will not try to approve the case, but pass it
along to the Impairment Underwriting module. Prior
to this exchange, the system determines any additional
information needed by the underwriter, such as blood
tests.
In the second half of 1988, Stage 3 of development
saw the expert system continue to evolve towards the
front end of the underwriting process. The programming staffdoubled in size. Another knowledge engineer
M.D. was hired and trained in Lincoln Nationars own
specific knowledge modelling processes. Once again,
domain familiarity in medical sciences was deemed an
essential component of knowledge engineering effectiveness for this particular effort. During this stage, the
Data Entry system and the Screening modules were
completed. While these modules are relatively limited
in knowledge complexity, Lincoln National has found
that actual implementations of them into client sites
entail large amounts of systems integration, particularly
to achieve high levels of decentralization to field offices.
Lincoln National's underwriting manual was also put
on-line in 1988 as a decision assistance tool for impairments not yet computerized.
Subsequent stages of development saw the creation
of the various modules surrounding the core underwriting processes in Figure 5. In 1989, the Work flow
Management module was enhanced for the individual
underwriter to manage his or her own personal case
load. For example, the front end of this module keeps
track of cases, the status of requirements that have been
ordered, and the status of specific underwriting problems. Each underwriter has an "in-tray," where cases
electronically assigned to the underwriter are presented
for evaluation. For each case, the underwriter then
works on a Problem Processing screen to manipulate,
track, change, or resolve specific underwriting problems. Electronic notepads are also provided for the underwriter, so that, through electronic mail, he or she
might have a new form of communications with agents
in the field. Also during 1989, Lincoln National finished the Management Information and Administration module. In addition to providing a series of reports
on both individual and department performance, underwriting managers can also adjust certain systems
tables to modify the processing of the entire system to
reflect the company's specific underwriting requirements and ideas. Examples might include the company's aggressiveness with respect to nonstandard ratings for particular medical impairments, or the policy
amount ceiling below which automatic case approval
can occur.
It should be noted that this expert system has an
architecture that rigorously separates data from logic
(i.e., databases from knowledge bases). Thus, the series
of expert system modules consist of knowledge bases
that read information from, and write information back
to a set of underlying tables through "C" programs
that utilize a library of database procedures. The database is therefore an important foundation of the system:
the data are shared between expert system modules,
and the database manager provides a straightforward
interface to external environments, such as other
mainframe applications or information services external to the user company. This architecture has allowed
the expert system to evolve in a more modular and
efficient fashion. Further, the database backbone has
facilitated systems integration with a user company's
existing client policy administration systems.
Thus, what started out as a simple prototype has
evolved into what may be fairly called a comprehensive
risk management system for life insurance. It is also a
new business. Lincoln National Risk Management was
formed and funded by the reinsurance parent. The
president of the new company was and continues to
serve as the vice president of underwriting for the reinsurance company. In an organizational move reflecting
the reinsurance parent's long-term commitment to the
venture, underwriting research and development personnel were shifted to support the " R & D " of the new
Strategic Use of Expert Systems
company. Leading these activities is the medical director who pioneered the underwriting system, and
working with him are a half dozen senior underwriters
and research staff formally from the reinsurance parent.
In terms of programming, approximately 25 persons
perform knowledge encoding, database programming,
and various types of systems integration. They are
managed by the former head of the reinsurance's division's own DP department. Marketing staff has also
been hired into the new company as well. The result
of these collective efforts is that installations to 20 other
direct insurance companies are now planned.
5. COMPARISONS AND CONCLUSIONS
How might a comparison of these two cases provide
insights for managers in insurance as well other industries embarking on strategic expert systems efforts?
The two underwriting expert systems described above
have clear differences. The initial focus of John Hancock's system was on the Initial Data Entry and Preliminary Processing aspects of the underwriting to yield
automatic approvals, whereas Lincoln National focused initially on Impairment Underwriting. Additionally, the systems evolved in opposite directions.
John Hancock's system expanded forward into more
complex elements of underwriting, while Lincoln National has surged backwards into the less complex elements. Each pattern is consistent with the companies'
respective business strategy: their basic positioning in
the insurance industry, the products they market, their
internal domain resources created over long periods of
time as a result of their positioning and products, and
the relative importance of work-flow improvement and
general expense reduction on one hand, and improved
mortality experience on the other. Another clear difference is that while John Hancock has developed its
expert system for internal use only, while Lincoln National has developed a new product intended for sale
to other insurance companies. This is a classic case of
"intrapreneurship," where a combination of new technologies and a new organization are employed to leverage and enhance the existing strengths of the larger
corporation. Lincoln National's drive to create a comprehensive risk management system has been channelled and accelerated by market needs. Insurers need
comprehensive underwriting systems handling both the
"front" and "back" ends of the underwriting process.
Their needs include automated underwriting and decentralized computing in addition to impairment underwriting decision assistance.
The similarities between these two expert systems
initiatives are also striking. Senior management in both
companies participated heavily in the selection and
planning of these two expert system initiatives, choosing from among the many possibilities for expert systems development shown earlier in Figure 1. The
questions that these managers considered may be gen-
23
eralized for a broader audience, one not limited only
to the insurance industry, and used to screen, set the
focus for, and staff expert system efforts:
I. Does the proposed expert system application target
core competitive processes? What are the company's
key decision-making processes, and who are the individuals who are best at making these decisions?
Can senior management get its best experts to participate in a major, multiyear systems development
effort? In both case studies, the general field of risk
management was perceived by senior management
as a core business process. As we have seen, however,
for the large direct insurer, complete automation of
a certain percentage of new cases to reduce costs
was the overriding competitive priority, whereas for
the reinsurer, initially better decision-making for
complex underwriting to improve mortality experience was the initial priority.
2. What is the feasibility of the proposed effort? Are
the systems design and technological implementation resources required to computerize these processes available internally, and if not, can they be
developed or contracted? More importantly, does
the company have the internal domain expertise
called for by the system? Lincoln National's "deep"
underwriting research and development resources,
required to be effective in the reinsurance business,
better equipped the company to tackle the backend
of the underwriting process. Both companies have
drawn heavily from central DP staffs as their expert
syster~s required substantial amounts of database
programming and systems integration effort.
3. What are the benefits? That is, what the company
stands to gain if the expert system is successfully
developed and installed? How do these benefits
compare to the costs of development, as well as the
maintenance of the system, including the updating
of knowledge bases? The benefits that such systems
may produce come at various organizational levels:
improved individual decision-making, organizational productivity through streamlined work flow,
improved customer service, and potentially the creation of new products or services. Within these two
cases, management understood the specific types of
benefits, and their scope, that could result from successful projects. While hard measures of benefits
would only be available upon completion and installation of their respective systems, the types and
ranges of financial and operational benefits could
be modelled based on extensive industry and business management experience. As described above,
John Hancock is realizing significant expense savings from automated underwriting, as well as improved customer service at the field office level. In
addition to providing these two benefits, Lincoln
National's expert system promises improved mortality experience for the underwriting decisions as
well as a new source of revenues for the parent corn-
24
pany. It also enhances the relationship between the
reinsurance c o m p a n y and its client companies.
Rather than being "surprises," the benefits in both
cases are a matter o f specific expectations, documented and well c o m m u n i c a t e d to team members,
being fulfilled.
4. W h a t is the competitive context? Are any competitors actually building certain strategic applications?
Can certain systems be identified as de facto elements o f doing business in the future? Here again,
the m a n a g e m e n t o f both companies saw early on
that a u t o m a t e d underwriting and decision support
for h u m a n underwriting comprised the future o f
risk m a n a g e m e n t in the insurance business. M a n y
other direct insurers have been developing screening
expert systems; and other reinsurers have explored
computerizing medical i m p a i r m e n t underwriting.
Thus, while these two initiatives have turned into
competitive opportunities for each respective c o m pany, from a long-term perspective, they were also
competitive necessities.
These questions and issues are certainly not unique
to the insurance industry. N o r is there a single right
answer in terms o f " w h a t expert system to build" for
every c o m p a n y , nor even for companies participating
in the same industry. These cases have shown that major expert systems development work is a function o f
business strategy, and thus, will be specific to each
c o m p a n y ' s strategic goals, customers, products, and
services. The participation o f senior m a n a g e m e n t in
strategic expert systems development is, therefore,
a m o n g the most i m p o r t a n t points that emerge from
the study o f these two cases. Both J o h n H a n c o c k and
Lincoln National still maintain "steering committees,"
formed in the first few m o n t h s o f project planning,
which review the progress o f the respective expert systems on a regular basis. The composition o f these
steering committees includes top m a n a g e m e n t from
J o h n H a n c o c k ' s "retail" business and Lincoln National's reinsurance business respectively, senior m a n agement o f each c o m p a n y ' s central D P department,
as well as the project managers and key technologists
from within each project. W i t h o u t this involvement,
it is hard to imagine that either one o f these expert
systems would have gained and maintained the appropriate business focus, budgetary and organizational
sponsorship, or the h u m a n resources that have propelled both to the level o f success each enjoys today.
Acknowledgements--Theauthors acknowledge the participation of
the followingindividuals in the preparation of this article: Raymond
BAIT,James Callahan, Brian Moody, and Michael Rich of John Hancock Mutual Life Insurance; and David Hopper, Russell Suever, and
M. H. Meyer et al.
Daniel Pieper of Lincoln National Risk Management, as well as the
industrial sponsors of the Expert Systems Research Interest Group
of the Center for Technology Management, Northeastern University.
No inference regarding participation by the authors in the actual
development of these two expert systems, nor any collaborativework
between John Hancock and Lincoln National in that development,
should be made by the reader.
REFERENCES
AM Best Company. (1990). Best's insurance reports: Life health 1990
(85th Annual ed.). Oldwick, NJ: Author.
Bailey, R. (1985). Underwriting in life and health insurance companies.
Atlanta, GA: Life Office Management Association.
Baker, A.H. (1988, October). Information technologies: Strategic opportunities for the life insurance industry. Washington, DC: Trend
Analysis Program: American Council of Life Insurance.
Bakos, Y. & Treacy, M. (1986). Information technologyand corporate
strategy: A research perspective. MIS Quarterly, 10(2), 107-119.
Benjamin, R., Rockart, J., Scott Morton, M., & Wyman, J. (1984).
Information technology: A strategic opportunity. Sloan Management Review, 25(3), 3-10.
Cebul, R.D. (1988). Decision making research at the interfacebetween
descriptive and prescriptive studies. Medical Decision Making,
8, 221-232.
Conversano, J. (1988, May/June). Expert system lightens underwriters' work load. Resource, 6-8.
Davenport, T.M. (1989, May/June). The case of the soft software
proposal. Harvard Business Review, 67(3), 12-24.
Davis, R. (1984). Amplifying expertise with expert systems. In P.
Winston & K. Prendergast (Eds.), The AI business (pp. 17-40).
Cambridge, MA: MIT Press,
DeTore, A.D. (1989). An introduction to expert systems. Journal of
Insurance Medicine, 21,233-236.
Feigenbaum, E., McCordick, P., & Nii, H. (1988). The rise of the
expert company. New York, NY: Times Books.
Hammer, M. (1990, July/August). Reengineering work: Don't automate, obliterate. Harvard Business Review. 68(4), 104-112.
IBC USA Conferences. (1990). The 2nd Annual Expert Systems In
Insurance Conference Proceedings, Natick, MA.
Ires, B., & Learmonth, G. (1984). The information system as a competitive weapon. Communications of the ACM, 27(12), 11931201.
Johnston, H., & Vitale, M. (1988, June). "Creating competitive advantage with interorganizational information systems." MIS
Quarterly, 12(2), 153-166.
Jones, D. (1987, October). Insurer's underwriting staff gets expert
help. National Underwriter. 3.
Leonard-Barton, D., & Sviokla, J. (1988). Putting expert systems to
work. Harvard Business Review, 66(2), 91-98.
Life Office Management Association. (1989, March/April). Study
helps track information processing costs. Resource, 14(2), 61.
May, K. (1990, March 15). Knowledge-basedsystemsin the insurance
industry. Systems AL 4-9.
Meyer, M.H., & Curley, K. (1991, Winter). Putting expert systems
technology to work. Sloan Management Review, 32(2), 21-31.
Prerau, D. (1985, Summer). Selection of an appropriate domain for
an expert system. AI Magazine. 6(2), 26-30.
Shepard, P., & Webster, A.C. (1957). Selection of risks. Chicago, IL:
Society of Actuaries.
Spilberg, D., De Salvo, T., & Ojha, H. (1987). Expert Systems in
Insurance Industry. 1987 Survey Report Update, New York:
Coopers & Lybrand.