Goal-Oriented BI

BI is typically thought of as a category of applications specifically related to the provision of information for management decision-making. In some cases, Data Warehouses (DW) are seen as components of BI, but many standalone BI tools exist for the delivery of on-line analytic processing, dashboards, and reporting. These tools capture data from a variety of different sources including the DW.

Currently, BI information is consumed in a variety of ways. Managers either interact directly with the system viewing data presented in graphical or tabular reports, or they rely on analysts to prepare such reports. In some cases, dashboards are used to provide a quick review of performance against targets. Alternatively or in conjunction, on-line analytic processing (OLAP) might be employed to explore performance across a range of different dimensions within the data set such as customer, location, or product.

The use of BI to actually make a decision, however, still requires a fair bit of data manipulation even after the data has been delivered in the formats mentioned above because another series of steps are needed to arrive at a decision after the data has been viewed. Although the fundamental goal of BI is to enable informed decision-making that results in improved organizational performance, it has been argued that, perhaps due to the lack of integration of BI into the decision making process, more than 50% of BI implementations fail to influence the decision-making process in any meaningful way.

The dichotomy between data delivery and the use of data in decision making is evidenced by the categorization of BI tools into data-driven support and decision support. Data support is related to the delivery of accurate, up-to-date data while decision support refers to the assistance provided to the user in actually making decisions based on the available data. As March and Hevner point out, simply loading transactional data into a Data Warehouse is not necessarily the answer if the objective is to enable managerial decision-making.

Korhonen et al. argue that one of the key challenges faced in institutionalizing decision aids is validating decision models used by the decision maker. The implication is that such a model would help to provide information more closely linked to the decision to be made. One problem associated with the integration of data in this way is the fact that the need for analytic information differs from the need for transactional information, and Information System professionals lack adequate models to clearly define analytic information needs.

This notion of delivering information based on "decision analysis", however, has a long history in management information systems (MIS). For example, over 30 years ago, Ackoff called for the use of decision analysis as the key to designing a useful MIS. King and Cleland, as well as Henderson and West, explored the use of "decision inventories", and King and Cleland proposed the concept of "decision models" as a means of specifying the decision environment in more detail.

More recent decision-centric approaches to information delivery include Watson and Frolick's "strategic business objective" technique. In this approach, the strategic business objectives and the processes that influence their accomplishment are defined and serve as the basis for information delivery. Similarly, goal-oriented techniques are based on mapping organizational goals and attendant sub-goals as a means of defining information needs of managers describe the goal-decision-information (GDI) approach for DW development, which defines goals and associated decisions leading to a better understanding of the data required in a DW. Finally, Liew and Sundaram define a complex "flexible object-oriented decision system" that leverages information about organizational objectives as a starting point for defining information requirements.

Although decision or goal analysis is the foundation for many of the above approaches, most are applied at the enterprise level and thus are not necessarily in line with current BI technologies. Furthermore, while they do focus on delivery of information linked to goals and outcomes, many are static in that they seek to design the goal system as a "business architecture" that informs the information architecture.

In addition, while many of these models do explore linkages between goals and information required for the decisions, the actual KPIs and the interrelationships between KPIs and goals are not defined. According to Popova and Sharpanskykh, even when relationships can be defined, such as in the ARIS model where users articulate cause-effect relationships using Balanced Scorecards and connect KPIs to strategic goals, the analysis options are inadequate due to a lack of formal modelling foundations and proper representations.

In order to shorten the gap between the delivery of information and decision making, we propose an approach that borrows from the decision analysis tradition. Specifically, we seek to create a BI-supported decision methodology by providing a means of defining  decision models that illustrate cause-effect relationships among variables relevant to the decision for specific roles in the organization. As in the GDI and strategic business objective approaches, the model includes definition of goals and the presumed drivers of these goals. To this approach we add Key Performance Indicators, which are organized according to the decision model and are "dimensionally aware", allowing users to compare KPIs from different store locations for example. Finally, to allow for adaptability in the process, we take a prototyping, iterative approach to model development, which allows for adaptation as business needs change or as managers become clearer on what types of information they need for various decisions.

It is conceivable that not all types of decisions in an organization would be enabled by such an approach. The taxonomy of management decisions from Koutsoukis and Mitra is depicted in Table 1. Executive decisions are thought to be longer term and relatively unstructured. Operational decisions are structured but are amenable to automation. Mid-level decisions by contrast, are semi-structured in that they relate to management control: ensuring that certain objectives meet their targets by managing the drivers of these objectives. Accordingly, our proposed methodology is expected to be particularly suited to mid-level management decision making.

Table 1 A taxonomy of management decision-making