Page 62 -

A Company and its Data 23

1.4.2. The quality of data models

Data quality is not simply concerned with the value of the
data. If the data models themselves are not to a high
standard then it is impossible to improve the quality of the
data. Indeed, only a data model can give the meaning of data
and their validation rules, without ambiguity and across the
whole of the Information System.

Even the classic example of unduplicating of addresses is
not immune to this: how is it possible to determine whether
an email address is private, professional or temporary
without taking into account its meaning? An email address
located in a marketing database might also correspond to a
personal address. The same piece of data is also stored in a
customer orders database but, this time, corresponds to
business address. The two addresses respond to different
semantics; the data model must express these
characteristics. The first is used in marketing operations and
communicated to third parties, following customer
confirmation. The second is used exclusively for sales and
after-sales. One can therefore not be favored over the other.
There are numerous examples to illustrate this, exposed by
the same semantic faults: ambiguity surrounding the
definition of the Client concept, loss of a sense of the
Revenue concept, or of the Site concept (area, place), etc.

Furthermore, the meaning of a piece of data is likely to
depend on definitions relative to other data. For instance,
“the Product has to be linked to two Manufacturing Units in
at least one Factory”. This rule of data referential integrity is
expressed in the model through the association between
Product, Manufacturing Unit and Factory business objects.