We develop the notion of a decision requirement as the pair <decision, information> where 'information' is that required by the decision maker to assess if the 'decision' is to be taken or not. It is shown that there are two kinds of decisions, imperative and managerial. The former are decisions about which transactional service out of a choice of transactional services is to be provided. Managerial decisions determine what infrastructure out of a set of possibilities is to be put in place. It is shown that a decision is the reason why a functionality of an information system is invoked. The notion of decision requirement is clarified through a decisional requirement meta model. This is supported by a decision and information meta model.
Goal oriented requirements engineering techniques [
The role of Requirements engineering in developing Data Warehouses has been
investigated only in the last decade or so [
Evidently there is a close relationship between the information systems and data warehouse of an organization. The former are used to populate the latter through the ETL process. In the opposite direction, the decision taken by using the data warehouse has the effect of changing information system contents. This means that information systems operate in a decisional environment. We consider this environment in the next section and show that there are two kinds of decisions, imperative and managerial. In the subsequent section we develop a meta model for decision requirements. Here we also model the notion of a decision and information from the data warehouse perspective. In section 4 we discuss our proposals with other related work.
The decisional environment provides the context in which an information system (IS) operates. This is shown in Fig. 1. When the information system is sent a stimulus from the decisional environment then the functionality that responds to this stimulus is invoked.
Stimuli can be sent by two different kinds of actors, IS administrators and IS operators. These stimuli correspond to two kinds of decisions, managerial and imperative. Managerial decisions are used to ‘initialize’ the IS where as the latter work within the initialized IS to operate the system. For example, in a railway reservation system IS administrators initialize train data whereas IS operators invoke functionality to make reservations and cancellations using information set up by the IS administrator.
Information System
Stimulus
Invoked function Decisional Environment: rationale for stimulus
Fig. 1. Embedded IS in a Decisional Environment.
Let there be a manager who has to perform extra work and needs to allot it to an employee. He can decide on the employee from the choice set {Transfer employee, Recruit employee, Overload employee}. The manager needs information to decide which alternative to pick and, also which individual employee shall be transferred, recruited, or overloaded respectively. There are two decision making problems here, to select from the choice set and to identify the individual, respectively. We shall use the notions of tactical decisions and operational decisions to classify these.
Fig. 2 shows the interplay of tactical and operational decisions. The tactical decision to Transfer an employee enters the operational decision making environment where the employee is identified and the stimulus to be sent to the information system is completely formulated. The information system performs the desired function and this information is now available to be sent to the DW at refresh time.
Looking from the information system outside, the decision making layers surrounding it formulate the stimulus to which the IS responds. This stimulus must identify IS functionality and the data. The former is done in the tactical environment whereas the latter is done in the operational decision making environment. 2.2
There are two kinds of managerial decisions, those that follow a business policy, enforce it or create exceptions to it, and those that formulate the policy. We refer to the former as administrative decisions, since they are concerned with administering the system and to the latter as policy decisions. The latter provide the context for the former.
DW To-Be
What to do with policy Choice set = {Modify, Stay, Delete}
Modify policy Choice set = {First class, Second class}
Add first class bogey
Information System
Policy Decision Making
Environment Administrative Decision Making Environment Information system Let us be given a policy decision that the ratio of first class bogies in a train to second class bogies is 1:2. This policy is to be enforced as an administrative decision. Policy decisions may define the norms and standards that are used by administrative decisions or business rules used by imperative decisions. A policy decision requires knowledge of the state of the organization. For example deciding the 1:2 norm above requires the knowledge of patterns of bookings made, revenue targets, revenue receipts etc. Out of the many choices available to fix the ratio, the policy decision maker uses this knowledge to fix the desired one.
Fig. 3 shows that the policy decision to modify the ratio of first to second class bogeys in a train leads to the administrative decision to add a first class bogey, and the information system is stimulated to reflect the change. This information is now available for train reservation purposes and is also available to be sent to the DW.
We have seen that in order to make a decision reference to the information in the data warehouse needs to be made. We represent this as a pair <decision, information> and refer to it as a decision requirement. Here, we elaborate on the notion of decision requirement.
The Decision Requirement, DR, meta-model is shown in Fig. 4. As shown it is modeled as an aggregate of information and decision.
Complex DR
N Composed of
1 required information, Unsatisfied Orders for 1-tonners and Unsatisfied Orders for 13tonners.. Each of these is an ISA relationship with Set up New Assembly Line.
Set up New Assembly Line
Unsatisfied
Orders
AND Decide Resources Capacity Available
Choose Land
Location Availability AND
Create Separate
Profit Centre
Staff Availability
Merge with Volume of
eCPxiersontiftnirteg HWanodrlked OR Now let us consider composition. The Decision Requirement <Set up New Assembly Line, Unsatisfied Orders> is a complex one having two component decision requirements, <Decide Capacity, Resources Available> and <Choose Location, Land Availability>. An AND link connects these two components so as to define the complex decision requirement, <Set up New Assembly Line, Unsatisfied Orders> (see Fig. 5).
The foregoing shows that a DR can be decomposed to reflect the decomposition of its decision component. It is also possible to do DR decomposition through information decomposition. In this case, the decision part is held constant whereas information components are elaborated. The Choose Location decision of Fig. 5 is shown as associated with the information, Land Availability. Land availability can be decomposed into two pieces of information, Land site and Land size Then the complex DR <Choose location, Land availability> can be decomposed into <Choose Location, Land site> and <Choose Location, Land size> respectively. 3.2
The key concept underlying the decision meta model of Fig. 6 is that of a decision parameter. Decision parameters reveal the factors that must be taken into consideration before a decision can be selected by the decision maker.
The decision to decision parameter relationship is M:N. A decision parameter must be associated with at least one decision. Similarly a decision must be associated with at least one decision parameter. Dependent decision parameters depend on other parameters for their existence whereas independent decision parameters determine a completely new aspect of a decision. Independent parameters may have dependent parameters but are themselves not dependent on any other decision parameter for their existence.
Decision
Cardinality
M has
N
N Decision Parameters
Depends on
M Independent Consider the decision Set_Up_New_Assembly_Line(Product Type, Location, Line Capacity). Here, the parameters, Product Type and Location are independent of one another. In contrast, Line capacity is dependent on Product Type since it is determined by the type of the product built by the line. The information model in Fig. 7, shows three kinds of information, detailed, summarized or aggregates, and historical. Aggregate information is obtained as a summary by computing from simpler information. This is shown in Fig. 7, by the specialization of information into Simple and Aggregate as well as by the ‘Is computed from’ relationship between Aggregate and Information.
Historical information is represented by the relationship ‘History of’ between Information and Temporal unit. The cardinality of this relationship shows that it is possible for information to have no temporal unit associated with it. In such a case, only current information is to be maintained. However, when a temporal unit is associated with information then we must also know the number of years of history to be maintained. This is captured, as shown in the figure, by the attribute Period.
Value-set Temporal Unit
Property
1 Period
Takes value from History of
N
N
Information N
M
Simple Aggregate Fig. 7. Information Model in Data Warehouses showing three kinds of information. Is computed from M Information is also associated with a value-set and takes on values from it. In Fig. 7 this association is called “Takes value from”.
In traditional goal oriented requirements engineering, the aim is to specify system functionality. No support is provided in determining which of the many actions is to be performed. In our proposals, however, the focus is on the latter.
Our approach does not attempt to directly reach facts and dimensions unlike the database and ER driven approaches. Additionally, unlike these approaches, we can identify the required aggregate and historical information.
Goal oriented data warehouse development approaches of [
Decision classification on the basis of time and planning horizon was proposed n
GRAI grid [
Finally, our decisional environment is similar to the work system proposed in [
The notion of decision making implies the existence of a choice set from which the alternative that best meets organizational goals, is selected. These alternatives can be (a) managerial, for setting up the environment and (b) imperative, for providing the right service. Our emphasis is on modeling the set of decisions and associated information in an organization. It is only thereafter that one can proceed to subsequent stages of star schema design.
The ideas presented here have been tried out in a health scheme operating in India. Details can be obtained from the authors. Future work is centred round elicitation of imperative and managerial decisions.