98 Part I • Information Technology
organization made a commitment, using the concept of
data steward, to actively manage the metadata for each
subject area of the business. This allowed subtle differ-
ences in customer data to be recognized and accepted and
for data to be stored accurately. Until this was done, cus-
tomers were inaccurately billed, product markets were not
accurately understood, and many employees wasted much
time trying to resolve misunderstandings.
Data Modeling
The role of data modeling as part of IS planning is essen-
tial. In practice, two rather different approaches are
followed—one top-down, called enterprise modeling, and
one bottom-up, called view integration. Many organiza-
tions choose to use both approaches because they are
complementary methods that emphasize different aspects
of data and, hence, check and balance each other.
The enterprise modeling approach involves
describing the organization and its data requirements at a
very high level, independent of particular reports, screens,
or detailed descriptions of data processing requirements.
First, the work of the organization is divided into its major
functions (e.g., selling, billing, manufacturing, and servic-
ing). Then each of these functions is further divided into
processes and each process into activities. An activity is
usually described at a rather high level (e.g., “forecast sales
for next quarter”). This three-level decomposition of the
business is depicted in Figure 4.2.
Given a rough understanding of each activity, a list
of data entities is then assigned to each. For example, quar-
terly forecasting activity might have the entities product,
customer order history, and work center associated with it.
The lists of entities are then checked to make sure that con-
sistent names are used and the meaning of each entity is
clear. Finally, based on general business policies and rules
of operation, relationships between the entities are identi-
fied, and a corporate data modelis drawn. Priorities are
set for what parts of the corporate data model are in need
of greatest improvement, and more detailed work assign-
ments are defined to describe these more clearly and to
revise databases accordingly.
Enterprise modeling has the advantage of not being
biased by a lot of details, current databases and files, or how
the business actually operates today. It is future oriented
and should identify a comprehensive set of generic data
requirements. On the other hand, it can be incomplete or
inaccurate because it might ignore some important details.
This is where the view integration approach can help.
Inview integration, each report, computer screen,
form, document, and so on to be produced from organiza-
tional databases is identified (usually starting from what is
done today). Each of these is called a user view. The data
elements in each user view are identified and put into a
basic structure called a normal form. Normalization, the
process of creating simple data structures from more com-
plex ones, consists of a set of rules that yields a data struc-
ture that is very stable and useful across many different
requirements. In fact, normalization is used as a tool to rid
data of troublesome anomalies associated with inserting,
deleting, and updating data. When the data structure is nor-
malized, the database can evolve with very few changes to
the parts that have already been developed and populated.
After each user view has been normalized, they are all
combined (or integrated) into one comprehensive descrip-
tion. Ideally, this integrated set of entities from normaliza-
tion will match those from enterprise modeling. In practice,
however, this is often not the case because of the different
focuses (top-down and bottom-up) of the two approaches.
Therefore, the enterprise and view-integrated data models
are reconciled, and a final data model is developed.
An alternative approach to data modeling, which over-
comes the difficulties of starting from a clean sheet of paper,
is to begin not within the organization but rather from
outside, using a generic data model developed for situations
similar to your own. So-called universal, logical, or packaged
data models have been developed from years of experience in
different industries or business areas. Prepackaged data mod-
els are customizable for the terminology and business rules
of your organization. Consultants and database software ven-
dors sell these starting points for your corporate data model.
The price for such a packaged data model is roughly the cost
of one to two senior database analysts for a year. Such
prepackaged corporate data models have several significant
advantages, including the following:Function 1Activity 1.1.1Process 1.mActivity 1.1.tFunctionnProcess 1.1CustomerMarketOrder ChannelProductCustomerBillFIGURE 4.2 Enterprise Decomposition for Data
Modeling