Enterprise Wide Data Modeling
Introduction
Arguments Against Enterprise Wide Data Models
Enterprise data modeling has been in the crossfire of several
articles recently. These name many shortcomings of data modeling
and quote lots of negative experiences. This leads some authors to
the final statement, that Enterprise Data modeling is completely
useless.
Most critics say, that an inflexible, top-down and centralist
approach like enterprise wide data modeling is not fitted to deal
with problems like ever faster changing environments, pressure
from global markets and decentralization. A central data model is
said to be a contradiction to a decentralized organization. That
rapid change, the critics say, will make the data model outdated
before the data analysts can bring it to the market.
We hope you can avoid some typical errors, when you take the time,
to evaluate the process of data modeling in your company. After
all the criticism one thing should not be forgotten. Data modeling
has been invented with a purpose. The purpose to avoid the
problems with isolated, non integrated systems, we all still know
from the stoneage of data processing. So by going back to the
roots because we have some difficulties with the better technique,
some of them caused by our own faults in using the method is not
the way to improve things.
As we want to provide a checklist, we should start with the
criticism and all the things that can go wrong. We will quote a
whole list of criticism first, to derive a list of questions that
should help you to detect some very common problems with data
modeling, that can be easily avoided.
The following is a list of common arguments against data modeling.
Proper Enterprise Data Modeling
Rapid Change
Accelerates Projects
Right Amount Of Detail
Summary
Enterprise data modeling has become subject to a crossfire of
criticism. Critics say, a centralized approach to data modeling
cannot reflect modern requirements for decentralized structures
and rapid organizational change as forced by global markets. Most
of the criticism stems from real bad experiences with data
modeling. But the conclusion should not be to condemn data
modeling, but to do it right. Netmation consultants have had
experience with various methods in quickly and efficiently
gathering critical business data and constructing a data model to
accurately reflect the business ignored to develop and implement
mission critical information systems.
The following arguments will show, that most of the problems
mentioned can be fixed by a data modeling process. We will use the
arguments against data modeling to show, that data modeling is
considerably better than the critics argue.
It should not be tried to model fast changing facts in a data
model. Hard wired organization schemes are not subject of data
modeling. But the core of business activities and business rules
is subject to data modeling, such as a bank account or an external
partner will stay very much the same for years. These objects are
not subject to rapid change. Data modeling should concentrate on
those core entities. The so called reference models take this as
their objective.
Data modeling can accelerate projects by being a service function
for projects. You visit your data modeling group and they tell you
about other peoples efforts going on, about entities, that already
exist an about how other people solved the same problem in a
reference model. A data administration department that does
reviews only, after it's to late is indeed a slow down. A data
administration that helps projects in the above way is a very
important step towards reuse and reusable objects.
There is no law that says: "You have to work top down when
practicing data modeling. There should indeed be a framework,
called a top level data model. But it should not be assumed that
you need to expand the model to the 3rd and 4th level of detail.
This detailed analysis should be part of the individual project
data models and should not be included in the enterprise data
model.
The question, most of the critics leave unanswered is, what do we
do if we don't do data modeling. The first alternative is to build
island solutions like in the stone age of data processing. These
are connected via interfaces. The lack of common terminology will
lead to some problems and considerable effort. The negative
experiences with this solutions led to an approach to fix those
problems - "data modeling".