Scorecards and dashboards seem to be everywhere in organizations these days. The concepts are typically associated with automated support for corporate performance management (CPM).2 Nevertheless, CPM and the supporting information and communication technology (ICT) cannot be reduced to these early warning systems. In this paper, we associate CPM and its high expectations with a generic application component representation and we elaborate on some underlying business intelligence (BI) technologies.

1.CPM

The terminology CPM undoubtedly owes much of its popularity to Gartner3, that describes it as follows:

"CPM is an umbrella term that describes the methodologies, metrics, processes and systems used to monitor and manage the business performance of an enterprise," (Buytendijk and Rayner, 2002).

This definition refers to a shift in scope from just registering the corporate performance for accounting purposes to effectively managing a rich and balanced set of organizational performance aspects, ultimately focusing on the continuity of the organization. ICT systems are mentioned on par with, or better, as enablers of adapted management methodologies and models, the adoption of performance measurement techniques, and integrated business process management. It should be clear that the kind of management maturity implied in Gartner's description of CPM requires an equally mature information and technology base.

2. Expectations

CPM clearly raises the bar for organizational management. The mandate for realizing the expectation of improved organizational management is reinforced by a recent set of company scandals (e.g. Enron, WorldCom) and resulting legislation and regulation (e.g. Sarbanes-Oxley, Basel II, Tabaksblatt) attesting to a definite need for more transparency in the way organizations are managed. These are some of the primary expectations set for CPM:

* The use of metrics - "You cannot manage what you cannot measure." This is equally true for the execution of organizational strategy. After the effective formulation of a strategy in clear objectives, the latter are to be translated into critical success factors that are then linked into well chosen performance metrics, so-called key performance indicators (KPI).

* The use of a balanced set of performance metrics - The call for balancing a variety of performance perspectives, fundamental to such frameworks as the Balanced Scorecard (Kaplan and Norton, 2000, 2004) and the Excellence Model of the European Foundation for Quality Management (EFQM, 2003), emerged in answer to the old-fashioned and overly narrow focus on financial and accounting metrics for assessing and benchmarking organizational performance. It is a call for balancing short term and long term objectives, identified objectives and their underlying drivers, and hard, objective metrics and softer, more subjective ones (Bruggeman, 2004). Figure 1 reflects the analysis pattern that is used by the Balanced Scorecard for linking performance metrics associated with the four traditional performance perspectives (i.e. shareholders, customers, internal processes, and innovation & learning) to the organizational strategy.

* The right-time delivery of actionable management information Management education is increasingly focusing on operating scenarios that stress the evolution towards the so-called "real-time enterprise." Although we prefer using the notion of "right-time enterprise," this does not alter the fundamental need for up-to-date management information in order to be able to compete within operating paradigms fraught with continuous environmental change. In view of the massively available potentially interesting information floating around, highly efficient and effective filtering mechanisms are essential for supporting contemporary organizational management.

* Horizontally integrated management - Aligning the value creating steps throughout the core enterprise processes, with a relentless focus on the customer - what is nowadays referred to as "business process management" (Smith and Fingar, 2003) - remains a fundamental organizational challenge. Organizational structures based on grouping functional competences, still a fundamental basis of most contemporary organizations, have a history of inducing silo management. The implementation of Enterprise Resource Planning (ERP) systems have, undoubtedly, provided some counterweight and contributed to functional integration of the enterprise. Still, their focus remained on automating repetitive transaction processes, rather than on supporting integrated, information-driven management, let alone strategic management. Moreover, since the introduction of ERP in the 1990s most of the organizations have changed operating models. Triggered by technological advances and globalization of trade we see more and more businesses shift from an intra-enterprise to an inter-enterprise operating model. A lot of the traditional integrated organizational value chains are reconfiguring into value networks of outsourced core competences. This does complicate streamlining the management of all the value adding puzzle pieces.

* Vertically integrated management - How do you make sure the top of your organization is synchronised, and more importantly, remains synchronised with its base? From a strategic management perspective this requires a clear articulation of the strategic objectives and underlying hypotheses. But this is just the start. In order to be able to execute the chosen strategy, the objectives and hypotheses need to be mapped onto the tactical and operational levels to align them. In Balanced Scorecard terminology (Kaplan and Norton, 2000, 2004) this is referred to as the "cascading process," which is notoriously hard to effectively implement.

* Closed loop management - Management, essentially, is about (re-) planning, organizing for execution, and control. The objective is to be and remain in control of the execution. Faced with constant, fast-paced change in the environment, mature management ought to be characterized by a capability for fast incremental learning as it continuously iterates through phases of planning, organization/ execution and control. This is valid for management at the strategic level as well as for management at the tactical and operational levels. Moreover, there is the need for continuous synchronisation of all three levels of management. All of this is supposed to streamline the value adding process components throughout all parts of the enterprise with an ultimate focus on the customer.

3. Application components

Automated CPM environments are composed of a number of application components. Figure 2 contains a generic depiction of the fundamental application components making up an automated CPM environment, based on Eckerson (2005).

The three views on performance information that make up the middle section of Figure 2 represent a natural sequence of information use for management purposes: monitor (e.g. early warning), analyze and report. Most users start off by monitoring KPIs for anomalies or exceptions. Then follows a phase of more complex multidimensional analysis, where the user is looking for causal relationships underlying the identified exceptions. Finally, the user validates and verifies the potential hypotheses he draws from the data with detailed data and reports this to all involved stakeholders before taking actions. Figure 3 presents a simple example of how the components in Figure 2 may interact in the context of operational fraud control when processing an insurance claim (Viaene et al., 2006).

The three views on performance information are equally relevant to all levels of management. It must be said, however, that the increase in the scope of the management information - from very local information to full-fledged enterprise-wide information - and depth of the management information - from very tangible operational information to aggregate and abstract strategic information - as we move from operational over tactical to strategic management coincides with a widening gap between the information views for monitoring, analysis and reporting. This is basically why practitioners seem to experience more trouble in designing good monitoring devices at the strategic level than at the operational level. Monitoring at the operational level of management tends to be much more directly driven by the detailed operational data. The success of monitoring devices at the strategic level tends to be highly dependent on the design of a much less straightforward to identify set of aggregate and more abstract KPIs.

We expect our CPM environments to be capable of linking up the more aggregated, abstract performance information with the more detailed performance numbers (and vice versa) in an efficient, consistent, transparent and user-friendly way. This is where the need for well integrated and well structured data at an enterprise level (see 4. Enterprise data warehousing) and for agile and efficient data navigation and analysis technologies (see 6. Multidimensional analysis and 7. Data mining) show up.

As information by itself is fundamentally worthless unless it is put to good use we should give special attention to the action component in Figure 2, in support of individual or group decision making and action. Integrating the information output from monitoring, analysis and reporting tools into the setting of meetings, workflows, knowledge work, etc. is itself a very specific challenge. How to effectively embed information into an organizational setting geared towards efficient decision making is one of the most neglected aspects when designing CPM environments.

The planning component in Figure 2 is fundamental to the pro-active capabilities of CPM environments. The ideal is for the plans, models and forecasts that are fundamental to the human-computer interaction in the monitoring, analysis, reporting and action components of the framework to be continuously updated and upgraded by information feedback mechanisms from these monitoring, analysis, reporting and action components. For example, the automated early claim screening monitor in Figure 3 will continuously have to be fed with up-to-date fraud indicator patterns extracted from the data that allow for detecting the latest types of insurance fraud. Data mining technology (see 7. Data mining) will help us recognise the patterns underlying new types of fraud in the available data.

What follows is a discussion of the data household that feeds the CPM application components presented in this section. Then we turn our attention to performance dashboards, multidimensional data analysis and data mining. These are the BI technologies that provide lifeblood to contemporary CPM environments.

4. Enterprise data warehousing

The centerpiece of an automated CPM environment, on top of which all CPM applications run, is the enterprise data warehouse. It represents the enterprise-wide consolidated and standardised data, grounded in agreed upon data definitions, business rules and data registration requirements and methods. The maturity of CPM implementations can be measured by the effective realization of this form of data household.

Bill Inmon describes a data warehouse as follows:

"A data warehouse is a subject-oriented, integrated, time-variant, non-volatile collection of data that is used primarily in organizational decision making," (Inmon, 1992).

This description is rooted in the distinction between transaction processing systems and BI systems, or as Inmon calls them, "organizational decision making" systems. Transaction processing systems are aimed at automating repetitive transactions and primarily focus on cost reduction and productivity, whereas BI systems are aimed at supporting managerial decision making and primarily focus on management effectiveness by providing managers with better information.4 The former systems are optimized for standard transaction throughput, the latter for response time in the context of complex data navigation and data querying.

The difference in finality between both types of system has its impact on the nature of the underlying data that are fuelling the system. The following four adjectives are used by Inmon to characterise data warehouse data as opposed to transactional data:

* "subject-oriented" - The organization of and the need for data is determined by the way decision makers look upon their decision domains (a.k.a. "subject areas"), i.e. as networks of linked data objects such as clients, suppliers, products, etc. This goes beyond the level of data linkage for the individual transaction and its processing. The single transaction process view underpinning the automation objective in transaction processing systems contrasts with the integrated subject area view underpinning the problem solving nature of BI systems. The data that is sufficient and necessary for automatically processing a transaction - nothing more, nothing less - gets enriched in a data warehouse with links to other data - from a transaction procession perspective redundant data - that managers optimizing the business deem interesting.

* "integrated" - In order to have a complete view of a subject area we typically have to integrate the relevant data from different source systems (usually transactional databases, possibly enriched with external data) into a data model that represents this reality (a.k.a. "the single version of the truth"). Beyond integrating data that spans multiple functional domains, this also entails catering to data redundancies, errors, inconsistencies, etc. during integration of the data sources. Don't forget that CPM is aiming for "corporate" integration.5 The data cleaning exercise may well take up an unexpectedly large part of the initial data warehousing budget. The objective is to come up with a set of transformation rules that allows for as much automation as possible for future source data loads into the warehouse.

* "time-variant " - All data in a data warehouse are accurate relative to a documented moment in time. From a practical perspective a data warehouse therefore can be looked upon as a series of periodically taken data snapshots, each registered with a time stamp. Depending on the volatility of the source data and the requirements of the managerial decision making process, source data will be loaded into the data warehouse and made available for analysis on an hourly, daily, weekly or monthly basis. "Real-time" data accuracy is in most cases not necessary - and also not preferable given the cost - as long as it is "right time." This way the data warehouse documents the complete history of events within the lifecycles of all relevant data objects for analysis purposes. This is different from what is typically documented in transactional databases, where one would merely expect accuracy of the stored data relative to the moment of access. In other words, we expect to see only the most recent status of a data object documented in a transactional database.

* "non-volatile " - Replacing outdated data with more recent data, i.e. overwriting older values with newer ones, is the standard in the world of transaction processing databases. In data warehouses, however, in principle no data are overwritten. Once data are correctly loaded into the data warehouse, they will not be changed. Newer data simply get appended to the old data after time stamping.

In order to be able to truly appreciate the challenge in building a high-quality enterprise data warehouse as a basis for CPM, we briefly look at how to, ideally, interact with this data household. Therefore we shall use the simplified (logical) warehouse architecture depicted in Figure 4. "Garbage in, garbage out" is the principle that underlies the rationale of this component structure.

Next to the actual warehouse data, it is the metadata that make the data warehouse function properly. Metadata stand for documentation on the actual warehouse data. This includes, among other things, the consolidated definitions of the warehoused data objects as well as the standardised set of rules and metrics used for performance management purposes. Metadata, in broad terms and from a warehouse user perspective, consist of all business and technical standards, rules and guidelines to guarantee the quality of the input into the data warehouse as well as the efficient, effective and consistent use of the warehoused data. The monitoring, analysis and reporting engines at the user side (top of Figure 4), as well as the integration engines at the input side of the data warehouse (bottom of Figure 4) will only approach the warehouse data by passing via the metadata to at all times ensure the quality of the stored data as well as the ensuing analysis and reporting.

It should be clear that investing in working with metadata - a bear necessity for ensuring high-quality data warehousing - does not just impose its share of discipline on the ICT side. The analysts and managers too shall be expected to abide by the rules and agreements captured in the metadata when they want to make use of the warehouse data. With reference to section 2, it is no less than seamless vertical and horizontal management integration of the enterprise that CPM is aiming for. Moreover, a significant part of the metadata, especially that metadata pertaining to the semantics and rules of the business, ought to be "owned" by the business itself, not ICT. Just think of the articulation of KPIs, their calculation and modes of use.

5. Performance dashboards and scorecards

Dashboards and scorecards are visually attractive monitoring mechanisms for information consumers. They are aimed at capturing the most critical performance information at a single glimpse. Though both notions are often used synonymously, and their difference may be labelled as futile by some, there seems to be a growing consensus in practice that associates dashboards with monitoring operational process performance and scorecards with monitoring the status and evolution of tactical and strategic objectives.

The way a performance dashboard is conceived is similar to the way you car's dashboard operates. the idea is to in "right time"-this may well be in "real time" - track the progress of events within the context of a process by means of the right set of KPIs. Figure 5 provides an example of what a performance dashboard may look like.

Scorecards on the other hand are monitoring devices of a more aggregate and periodical nature for tracking the status and evolution of a set of higher-level performance objectives, their underlying cause and effect relationships, critical success factors and KPIs. Ideally, the performance information contained in scorecards gets seamlessly linked into more detailed performance dashboards via interactive visualization mechanisms such as hyperlinks. Figure 6 provides an example of what a scorecard may look like.

6. Multidimensional analysis

During analysis navigating the enormity of data in the warehouse, in search of cause and effect relationships, is facilitated by online analytical processing (OLAP) technology. This technology complements the expert-level, time-honoured Structured Query Language (SQL)-based data query technology with the possibility for the non-ICT-sawy information consumers to get on-demand access to the performance data according to the real dimensionality of the data. Sarah Forsman of the OLAP Council defined OLAP as follows:

"OLAP is a category of software technology that enables analysts, managers and executives to gain insight into data through fast, consistent, interactive access to a wide variety of possible views of information that have been transformed from raw data to reflect the real dimensionality of the enterprise as understood by the user." (Forsman, 1997).

OLAP technology supports analyst-steered navigation of the warehouse data. The data are logically approached as a multidimensional cube of registered facts, just like the sales data cube in Figure 7. Cube operations supported by the OLAP software include adding and removing dimensions, pivoting cubes, re-specifying facts as frequencies, percentages, etc., and aggregating more detailed fact data along predefined dimensional concept hierarchies (e.g. day à quarter à year aggregation for the time dimension).

If you are a Microsoft Excel spreadsheet user then you have undoubtedly learned to appreciate OLAP data navigation capabilities in the form of the "pivot table" tool, Microsoft's equivalent to the multidimensional data cube. Today we also see more and more electronic reports and monitoring tools offer the capability to its users to interactively parameterise the view on the reported performance data with the use of embedded pivot table objects.

7. Data mining

Data mining stands for the algorithmic extraction of interesting patterns from huge amounts of (structured) data. Whereas with OLAP technology the analyst is still in the driver seat navigating the data in search for interesting insights, with data mining the navigation is, at least partially, set to auto pilot. The mined patterns are aimed at helping the analyst navigate the data more efficiently and effectively, at developing better early warning or analysis models, and at forecasting future performance.

By means of advanced data mining and statistical algorithms like neural networks, support vector machines, Bayesian networks, induction trees and linear regression we are capable of discovering patterns from the data that the human eye would not be able to spot, or at least way too late. The models (usually if-then-else rule sets) that are devised from the extracted patterns can then be applied to new situations to predict, classify, score, segment or make recommendations.

Experience has shown that designing a truly intelligent auto pilot requires specific analytical expert skills. These skills are mastered by trained statisticians and specialised analysts. One of the most fundamental challenges when mining for interesting patterns in data is to only dig up incrementally valuable information. In other words, one should make sure that the mining algorithm is able to capitalise on the existing business understanding in order to be able to go beyond what we already know. There is no point in investing in the design of an algorithm that only digs up prior knowledge. Still, since in many cases most of this prior knowledge resides only in the heads of certain domain experts our first goal should be to make these prior patterns of insight explicit, in order for us to in a next phase be able to, somehow, build them into the mining algorithm to guide its search through the data.

Bringing the mined patterns to the attention of human decision makers in such a way that the mined insight improves their analytical capabilities and properly guides their actions is no simple task either. It is obvious that having domain experts and information consumers in close proximity during the entire data mining exercise is a critical success factor of this kind of endeavour. In the end it's all about getting human creativity^ which has been numbed by masses of data, back online with somFsupport of so-called intelligent technology.

8. Conclusion

Early warning applications in the form of performance scorecards and dashboards are just the tip of the iceberg when aiming to comprehend the true nature of ICT support enabling contemporary CPM environments. We started by outlining some of the fundamental expectations of improved organizational management set out for CPM. Then we provided a generic depiction of the fundamental application components making up an automated CPM environment. Finally, we elaborated on some of the BI technologies that provide lifeblood to contemporary CPM environments, i.e. enterprise data warehousing, multidimensional analysis or OLAP, and data mining.

9. Checklist for practitioners

* Effective CPM revolves around stimulating and enabling human creativity and decision making competencies rather than just automated decision taking.

* For any CPM effort, ensure executive-level sponsorship that is committed to securing both short-term as well as long-term CPM development budgets.

* Get ready for an iterative, long-term learning process. Delivering on the expectations set out for CPM entails a journey of learning while doing. Value is essentially created over time.

* Start small, think big. Take an effective start with a highly-focused first project that is relatively narrow in scope and modest in its goals but is able to show clear benefits. Use this as a springboard for further investment. Focus on incrementally creating value by reusing previous realizations to help launch the next planned ones.

* Setting up a business CPM architecture and discipline focused on the proper use of information, is at least as important as putting in place a technical CPM architecture and discipline focusing on the design of technology systems.

* The maturity of CPM implementations can be measured by the effective realization of an enterprise data warehouse.

10. Acknowledgement

We gratefully acknowledge the support for writing this article from the K.U.Leuven Research Chair on Knowledge Discovery in Databases sponsored by the Dutch Police Region Amsterdam-Amstelland.