From User Requirements to Conceptual Design in Data Warehouse Design - Data Warehousing Design and Advanced Engineering Applications - page 7

Table 1. Comparison of the basic user requirement techniques

Supply-Driven User-Driven Goal-Driven

Basic approach Bottom-up Bottom-up Top-Down

Users involvement Low: DB Administrators High: Business users High: Top management

Business users must have a good
knowledge of the processes and
organization of the company

Willingness of top management
to participate in the design
process

Existence of a reconciled data
level

Constraints

Maximize the probability of
a correct identification of the
relevant KPIs.

The availability of data is en-
sured

Raise the acceptance of the
system.

Strengths

Difficulties in being supported
by top management and in
translating the business strategy
into quantifiable KPIs.

The multidimensional schemata
do not fit business user require-
ments.

Quick obsolescence of the mul-
tidimensional schemata due to
changes of the business users.

Risks

Depends on the level of the
interviewed users, typically
tactical

Targeting organizational level Operational and tactical Strategic and tactical

Moderators; Economist; DW
designers

Skills of project staff

DW designers Moderators; DW designers

Risk of obsolescence Low High Low

Number of source systems Low Moderate High

Cost Low High High

goal-oriented and the supply-driven techniques.
They initially carry out a goal-driven step based
on the GQM method (Vassiliadis et al. 1999; Basili
et al. 1994) in order to identify the business needs
the data mart is expected to meet. The outcome of
this phase is a set of ideal star schemata obtained
by a progressive top-down definition of the KPIs
necessary to measure the goals. Then, the schema
of the operational data sources are examined us-
ing a semi-automatic approach; the candidate
facts, dimension and hierarchies are extracted
and modelled using a graph-based representation
(i.e. star join graph ). Finally, the results of the
two phases are integrated, thus determining the
multidimensional model of the data mart as the
set of candidate star join graphs that is most suited
to matching the ideal star schemata.
Similarly, in GRAnD, the approach proposed
by Giorgini et al. (2006), a goal-oriented step based
on the Tropos methodology (Bresciani, 2004) is
carried out in order to identify the business goals
and the terms relevant to their monitoring. These

terms are then mapped onto the operational source
schema that is finally explored in a semi-automatic
fashion in order to build the final multidimensional
schemata.
A third approach coupling a goal-driven step
with a supply-driven one has been proposed in
(Mazon et al., 2007). The approach builds on a
Model Driven Architecture (Mazon & Trujillo,
2008) where information is formalized using
UML. In particular, project goals are modelled us-
ing a UML profile based on the i* framework that
has been properly adapted and extended to fit the
DW specificities. Starting from the collected goals
an initial conceptual model is obtained; then its
correctness and feasibility is checked against data
sources by using the Multidimensional Normal
Forms (Lechtenboerger & Vossen, 2003).
The main difference between the three methods
is that while in GRAnD the results of the goal-
oriented step are used to drive the supply-driven
one, in the other ones the goal-driven and the
supply-driven steps are almost independent and