Ethical Decision
Softw Syst Model (2014) 13:1015–1041 DOI 10.1007/s10270-012-0290-8
T H E M E S E C T I O N PA P E R
Strategic business modeling: representation and reasoning
Jennifer Horkoff · Daniele Barone · Lei Jiang · Eric Yu · Daniel Amyot · Alex Borgida · John Mylopoulos
Received: 1 November 2011 / Revised: 7 August 2012 / Accepted: 13 September 2012 / Published online: 26 October 2012 © Springer-Verlag Berlin Heidelberg 2012
Abstract Business intelligence (BI) offers tremendous potential for business organizations to gain insights into their day-to-day operations, as well as longer term opportunities and threats. However, most of today’s BI tools are based on models that are too much data-oriented from the point of view of business decision makers. We propose an enterprise modeling approach to bridge the business-level understand- ing of the enterprise with its representations in databases and data warehouses. The business intelligence model (BIM) offers concepts familiar to business decision making—such as goals, strategies, processes, situations, influences, and indicators. Unlike many enterprise models which are meant to be used to derive, manage, or align with IT system imple-
Communicated by Dr. Tony Clark, Balbir Barn, Alan Brown, and Florian Matthes.
J. Horkoff (B) · D. Barone · L. Jiang · J. Mylopoulos Department of Computer Science, University of Toronto, Toronto, Canada e-mail: jenhork@cs.utoronto.ca; jenhork@cs.toronto.edu
D. Barone e-mail: barone@cs.toronto.edu
L. Jiang e-mail: leijiang@cs.toronto.edu
J. Mylopoulos e-mail: jm@cs.toronto.edu
E. Yu Faculty of Information, University of Toronto, Toronto, Canada e-mail: eric.yu@utoronto.ca
D. Amyot EECS, University of Ottawa, Ottawa, Canada e-mail: damyot@eecs.uottawa.ca
A. Borgida Department of Computer Science, Rutgers University, Camden, USA e-mail: borgida@cs.rutgers.edu
mentations, BIM aims to help business users organize and make sense of the vast amounts of data about the enterprise and its external environment. In this paper, we present core BIM concepts, focusing especially on reasoning about situ- ations, influences, and indicators. Such reasoning supports strategic analysis of business objectives in light of current enterprise data, allowing analysts to explore scenarios and find alternative strategies. We describe how goal reasoning techniques from conceptual modeling and requirements engi- neering have been applied to BIM. Techniques are also pro- vided to support reasoning with indicators linked to business metrics, including cases where specifications of indicators are incomplete. Evaluation of the proposed modeling and reasoning framework includes an on-going prototype imple- mentation, as well as case studies.
Keywords Business intelligence · Business model · Conceptual modeling languages · Influence diagrams · Goal · Situation · Key performance indicators · Strategic planning
1 Introduction
Business intelligence (BI) offers tremendous potential for business organizations to gain insights into their day-to-day operations, as well as longer term opportunities and threats. The past decade has seen unprecedented interest in BI tech- nologies and services, and a corresponding growth of the BI market. By now, most large organizations worldwide have a significant investment in BI. However, most BI systems are closely linked to the structure of available data, provid- ing detailed statistics that may be hard to understand with respect to overall business strategy. A recent survey indicates that BI system users are less satisfied with system flexibility
123
1016 J. Horkoff et al.
and the ability to manage risk than they are with the system as a whole [1]. However, the flexibility required to mod- ify BI queries, posing what-if analysis questions exploring risk factors, often requires technical knowledge concerning the structure of available data. Even state-of-the-art systems that raise the abstraction level of BI systems require detailed knowledge of queries and data dimensions (e.g., [2]). This need for technical knowledge poses a serious conceptual obstacle for most business users, who are interested in having their business data analyzed in their terms: strategic objec- tives, business models1 and strategies, business processes, markets, trends and risks. BI data and statistics often focus on specific measures of current operations, without providing a clear link to how these measures affect business strategies, or how they facilitate planning and decision making. Conse- quently, the gap between the world of business and the world of IT-supplied data remains one of the greatest barriers to the adoption of BI technologies [3].
As in the broader area of enterprise modeling, there is a need to provide modeling support at the strategic business level, so that data residing in databases and data warehouses can be interpreted and understood in business terms, thereby facilitating reasoning over strategic decisions. To bridge this business-data gap in BI, we have proposed a strategic busi- ness modeling approach that makes use of concepts from existing modeling approaches or methodologies familiar to business decision makers (e.g., strategy maps [4], SWOT analysis [5], the business motivation model [6]). The busi- ness intelligence model (BIM) [7, 8] is a business model- ing language that offers concepts such as goals, strategies, processes, situations, influences, and indicators, and tech- niques for reasoning about them. Unlike many enterprise models meant to be used to derive, manage, or align with IT system implementations, BIM aims to help business users organize and make sense of the vast amounts of data about the enterprise and its external environment. An enterprise model such as BIM may be viewed as the business-level counterpart to conceptual data models (e.g., entity-relationships mod- els), so that strategic objectives, business processes, risks and trends can all be represented in a BIM model, for purposes of analysis and monitoring.
Consider for example, a consumer electronics retailer with vast amounts of data from diverse sources available for analy- sis through BI tools. Business analysts and strategists want to be able to pose a variety of analysis questions, without being constrained by the structure of current BI reports and queries, which are often technically-oriented and difficult to modify. Instead, they want to pose questions at the business level,
1 We use the terms “enterprise model” and “business model” in a con- ceptual modeling sense, i.e., a collection of elements and relationships typically having a graphical representation, and not in the business sense of how an organization creates, delivers, and captures value.
facilitating reasoning over and comparison of potential busi- ness strategies. For example: Should we develop technology in-house or acquire technology through acquisition? Which option is better for maintaining revenue growth and reducing risks? Given the state of the business according to current data, will we be able to maintain revenue growth without new strategic partnerships or technology acquisitions?
In this paper, we build on existing work to show how con- struction and analysis of BIM models (also called schemas, by analogy to database schemas) could allow organizations to answer such questions. We capture the necessary business objectives, risks, and measures by focusing on three key con- cepts in BIM: situation, influence and indicator. BIM models can also be used to reason about strategic objectives, such as “increase sales volume” or “maintain revenue growth”, by estimating the degree to which they are achieved, or the probability that they will be achieved. Reasoning over BIM models can use current business data from indicators, or can use hypothetical data in “what if?” scenarios, facilitating both exploration and monitoring of business objectives. Although reasoning over BIM can allow analysts to formulate ques- tions at the level of business strategies, it is challenging to completely elicit or specify all necessary links to business data (indicators, business metrics). Accordingly, our pro- posed techniques support reasoning even if BIM models are incomplete, missing information regarding indicators, busi- ness metrics, or probabilistic information, especially during intermediate stages of model development. Supporting rea- soning over incomplete indicators requires use of qualitative reasoning techniques, which come with different information requirements.
Our previous work has introduced elements of the BIM language, describing key concepts and applications of rea- soning techniques [8–10]. Further work describes BIM rea- soning with indicator values from business data [11, 12]. The work presented in this paper is an extension, improvement and integration of our published work, combining existing BIM papers into one consistent description of language con- cepts and reasoning. Specifically, we build on existing work by: • Offering a more precise and detailed account of core BIM
concepts; • Using a consistent running example from real-world
analysis reports to demonstrate all concepts and reasoning techniques;
• Describing in more detail a methodology for constructing BIM models;
• Providing an overview of BIM reasoning techniques, summarizing information requirements and linking them to our proposed methodology;
• Providing more detail concerning the use of existing rea- soning approaches for BIM, including details concerning the mapping of BIM to existing languages and tools;
123
Strategic business modeling: representation and reasoning 1017
• Extending and revising the description of reasoning with indicators, including composite indicators;
• Describing how to reason with incomplete indicators using a hybrid reasoning approach;
• Consolidating the description of tool support for BIM, including use of existing model reasoning tools and custom-build BIM prototypes.
• Summarizing ongoing studies applying BIM to real- world cases, including a consideration of model scala- bility.
This research is conducted in the context of the Business Intelligence Network, a Canada-wide strategic research net- work with academic and industry partners.
The rest of the paper is organized as follows. Section 2 introduces BIM concepts. Section 3 presents alternative rea- soning techniques for BIM models. Some of these are based on existing proposals. Others are novel, such as the indicator reasoning techniques that depend on the availability of indi- cator and probabilistic information. Section 4 describes use of existing tool support as well as in-progress implementa- tions of prototype tools. We summarize the results of ongoing case studies using BIM in Sect. 5. Section 6 discusses related work, while Sect. 7 provides conclusions and outlines future work.
2 Strategic business models
In this section, we provide a description of the key compo- nents of the BIM, including goal, situation, influence, and indicator. A version of the metamodel linking these concepts can be found in [7, 8]. A running example is introduced to illustrate these concepts and the reasoning techniques pro- posed in later sections.
2.1 Running example: BestTech
The example presents the viewpoint of BestTech, a gen- eric company developing and selling consumer electronics. Model contents have been extracted from real-world analysis reports, published by DataMonitor, a company that special- izes in industry analysis for a number of industry sectors. The example, presented incrementally in the following sections (Figs. 1, 2, 3), contains an interrelated network of goals, situa- tions, processes, indicators, and domain assumptions relevant to BestTech.
2.2 BIM concepts and relationships
This section describes BIM’s key concepts and relation- ships—goal, situation, influence, and indicator—in detail,
using our running example for illustration. This section is an expansion and consolidation of the BIM description in [8–12].
2.2.1 Goal
The concept of goal has a long history as part of enterprise modeling (e.g. [13, 14]) and requirements analysis (e.g., [15– 18]). A goal represents an objective of a business. Goals may be (AND/OR) refined into sub-goals so that their satisfaction depends on that of their sub-goals. Moreover, a goal may be satisfied in multiple ways if it or its sub-goals are OR-refined, in which case a choice needs to be made among alternatives. In addition, a goal’s satisfaction may be affected by that of goals other than its sub-goals. Using typical goal model syn- tax such as in [15, 17, 18] goal-oriented elicitation within an enterprise produces a goal model consisting of an AND/OR refinement graph including positive/negative contributions.
Examples of goals in the BIM language are shown in Fig. 1. Notice how the “To increase sales” goal is AND- decomposed into sub-goals “To increase sales volume” and “To maintain gross margin”. Similarly, the “To increase sales volume” goal is OR-decomposed into two alterna- tive sub-goals, namely “To open sales channels”, and “To offer promotions”. Further concepts in this figure (e.g., situ- ation, influence, and indicator) are explained in the following sections.
The satisfaction of a goal can be inferred from the satisfac- tion level of other goals using label propagation algorithms [19, 20]. We describe how these algorithms can be applied to BIM in Sect. 3.
In addition to goals, the BIM language supports the notion of processes within an enterprise, as well as domain assump- tions describing properties required for goal satisfaction. Domain assumptions can be thought of as situations required to be true in order to achieve a goal. Domain assumptions can be part of a goal refinement alongside other sub-goals. We see an example of a domain assumption in Fig. 1 where “High demand” must be true in order for “Increase Sales” to be sat- isfied. A domain assumption may, in fact, be false (broken), in which case goal fulfillment is not possible.
Processes can be associated with a particular goal. We say that a process achieves a goal, represented by an achieves link. When used in this way, processes provide a “how” dimension to complement the intentional “why” dimen- sion of goals. We further connect these concepts to busi- ness processes with the notion of indicator, explored in later sections.
2.2.2 Situation
During strategic planning, SWOT analysis [5] is often used to identify internal and external factors that may influence
123
1018 J. Horkoff et al.
Legend
Goal
Indicator
Process
Situation
To open sales channels
To increase sales
Increased competition
High demand
To increase sales volume
To maintain gross margin
-
To offer promotions
+
To reduce costs
-
-
-
Total sales
Sales volume
Gross margin
Economic Slowdown
Low cost financing
- ++
AND
OR
Evaluates
Domain Assumption
AND
Refinement
OR
+
Influence
Achieves
Fig. 1 Example of goals, situations, and influences for BestTech
the fulfillment, favorable or unfavorable, of strategic goals. SWOT stands for Strengths (internal, favorable), Weaknesses (internal, unfavorable), Opportunities (external, favorable), and Threats (external, unfavorable). We propose to model these in terms of the notion of situation. Intuitively, a situation defines a partial state of affairs (partial model of the world) in terms of things that exist in that state, their properties, and interrelations [21]. Since we are interested in strategic busi- ness models, we focus on organizational situations. The same situation may be favorable for some organizational goals, represented via positive influence links on model concepts, but unfavorable for others, represented via negative influ- ence. In our example, the situation “Increased competition” constitutes a threat to the goal “To open sales channels”), while “low cost financing” is an opportunity for “Healthy bal- ance sheet” and therefore “Sufficient Funds”, and “economic slowdown” is a threat for “High Demand”.
Analogously to satisfaction levels for goals, we have occurrence levels for situations, which denote the degree to which a situation occurs in the current state-of-affairs.
2.2.3 Influence
To express the influence of situations on strategic goals and other situations, we extend the contribution relation from goals to situations. In order to support both reason- ing over goal satisfaction/denial and reasoning using con- ditional probabilities, BIM supports two types of influence
links: logical and probabilistic. Quantitative logical influence links are an estimation of the positive or negative influence of an object on another, while probabilistic influence links estimate the probability of an object being satisfied (denied) given the satisfaction (denial) of another object. In either case, influence links can have varying strengths specified in qualitative or quantitative terms.
Logical influence. As in many goal modeling approaches (e.g, [15, 17, 18]), one goal influences another if its sat- isfaction/denial implies (partial) satisfaction/denial of the other. Such relations also hold between situations and goals. We call this type of influence logical.
The influence strength is modeled using qualitative values: + (weak positive), ++ (strong positive), − (weak negative), and −− (strong negative)), as in existing goal mod- eling frameworks. When logical qualitative values are not expressive enough to capture the difference between influ- ence strengths, quantitative logical values in some standard range (e.g., [−1, 1]) can be used. For example, + may corre- spond to 0.5 and ++ to 1.0. If the modeler wants to express a strength in between, a number (e.g., 0.7) may be used, instead of further qualitative values.
Figure 1 shows some examples of influences from situ- ations to goals, e.g., the “Increased competition” situation, representing an external threat for the company, influences negatively the “To open sales channels” goal. Similarly, the situation “Economic slowdown” negatively influences the domain assumption “High demand”. An example of influ- ence among goals can be seen in the negative link from “To offer promotions” to “To maintain gross margins”.
Probabilistic influences. To support probabilistic reason- ing, we also support probabilistic influences among situa- tions, goals and domain assumptions. In this case influence links represent the conditional probability of satisfaction. For example, in Fig. 1, if the partial positive influence link between “To reduce costs” and “To maintain gross margins” had a probabilistic type with a strength of 0.7, it would mean that the probability of “To maintain gross margin” occurring given the satisfaction of “To reduce costs” is 0.7 ( P(T o r educe cost s | T o mai nt ai n gr oss mar gi n) = 0.7). These strengths can be quantitative (e.g., 0.7, 70% chance of satisfaction) or quantitative (e.g., high chance of satisfaction). In Sect. 3.3, we show how this type of influence is used to support decision-theoretic analysis. We summa- rize supported influence types in Table 1, showing example
Table 1 Example influence strengths for each combination of indicator and measurement type
Influence Types Logical Probabilistic
Qualitative ++, +, -, -- Quantitative +0.7, -0.7, +0.2, -0.2, 1.0
A B
123
Strategic business modeling: representation and reasoning 1019
To increase sales volume
Sales volume
Increase Sales
Evaluates Measures
Why What
Fig. 2 Detailed view of an example indicator
influence strengths for each combination of influence and measurement types.
2.2.4 Indicator
A successful business depends both on its initial strategic planning and subsequent business operations. Performance measures play an important role in helping businesses align their daily activities with their strategic objectives. Generally speaking, performance measures quantify various aspects of business activities, including their input, execution and output, for monitoring, control and improvement purposes [22]. We model performance measures through indicators. An indicator (or in some cases, a key performance indicator) is a metric that evaluates performance with respect to some objective, be it the degree of fulfillment for a strategic goal, or the quality of a business process or product. Such metrics can be directly derived from data, or can use a formula to combine values. In BIM, indicators constitute a conceptual bridge connecting a BIM model to enterprise data found in a variety of data sources.
In BIM, each indicator is associated with a particular model element (e.g., goal, situation). Indicators associated with a goal are also associated with a process which achieves the goal. Associating an indicator with a goal provides the “why” dimension, motivating the need for a specific measure, while association with a process provides the “how” dimen- sion, linking the indicator to a concrete business process. We say that an indicator evaluates a particular goal (or sit- uation) while it measures a process. A simple example is shown in Fig. 2, extracted and expanded from Fig. 1. Here the “Sales volume” indicator evaluates the “To increase sales volume” goal and measures the “Increase Sales” process. In order to simplify model presentation, we often omit the con- crete processes which may be associated with indicators.
Further, example indicators can be found in Fig. 1, where “Gross margin” evaluates “To maintain gross margin” and “Total sales” evaluates “Increase Sales”.
Performance measures employed in a business environ- ment often form an aggregation hierarchy—a higher-level measure is defined in terms of lower-level ones. Top level measures (e.g., “Total sales”, “Sales volume”) usually give a clear picture whether a business is moving towards fulfill-
ing its strategic objectives, while leaf level measures (e.g., “Number of competitors”, “Number of promotions”) are usu- ally tied to specific actions and responsibilities.
When eliciting or defining hierarchies of composite indi- cators, the value of an indicator measuring a model ele- ment should depend on the values of indicators measuring elements one level lower in the hierarchy. Unfortunately, there are no guidelines on how this dependency should be defined consistently for a given BIM model. We address these issues, including indicator measurement and propagation, in Sect. 3.4.
2.3 Construction of BIM models: sample methodology
In practice, BIM models can be built iteratively, either by starting with business goals and then working in a “top-down” fashion to derive required indicators and pro- cesses by asking “how” questions. Alternatively, one can start with indicators and processes and work “bottom-up” to elicit goals and situations by asking “why” questions. For illustration purposes, we describe here a hybrid method, per- forming “top-down” goal and situation elicitation based on the Tropos goal-modeling methodology [23], then matching existing indicators to elicited business goals and situations. This approach can reveal the need for indicators that do not yet exist in the current BI implementation. Such a hybrid approach is often applicable in practice, as many organiza- tions have an existing set of business indicators and data sources.
We start our example by identifying business goals and their interrelationships. Such a process can be viewed as part of strategic planning [24], usually starting with the defini- tion of an organization’s mission, followed by the specifica- tion of goals for fulfilling the mission, and the strategies to achieve these goals. In our Fig. 3 example, we have two root goals: “To maintain revenue growth” and “To reduce risks”. To achieve the first one, we need to achieve both “To increase sales” and “To maintain competitive advantage”. Influences are also identified among sub-goals. One of several alterna- tive ways to maintain competitive advantage is “To acquire technology through partnership”. This alternative helps to reduce financial risk, but increases the dependence on exter- nal partners. A goal can be decomposed into both sub-goals and domain assumptions. For example, in addition to achiev- ing the goals “To increase sales volume” and “To main gross margin”, the domain assumption “High demand” (for our products/services) needs to be true, in order fulfill the goal “To increase sales” (in dollar amount).
After goal modeling, we identify the internal and external factors that may influence fulfillment of the goals identified previously. Specifically, we start with domain assumptions in the model, and ask the question: what observable evidence could potentially support or challenge these assumptions.
123
1020 J. Horkoff et al.
To open sales channels
To increase sales
Increased competition
To acquire technology through
acquisition
To develop new technology in-house
To reduce risks
To reduce patent
infringement lawsuit risk
To reduce external
dependence
To establish strategic
partnership
-
To reduce financial risk
+
+
To invest in new technologies
-
To maintain competitive advantage
+
+
Acquire a competitor
Develop a technology
+
++
Sufficient funds
Strong R & D capability
High demand
To maintain revenue growth
To increase sales volume
To maintain gross margin
- To offer
promotions
-
+
To reduce costs
-
- -
Total sales
Sales volume
Gross margin
-
Economic Slowdown
Healthy balance sheet
High R&D expenditure Low cost
financing +
++
+
+
-
++
AND
ANDAND
AND
OR
OR
OR
Fig. 3 BestTech model example
In our example, “High R&D expenditure” (strength) posi- tively contributes to the domain assumption “Strong R&D capability”, while “Healthy balance sheet” (strength) sug- gests with a high degree of certainty that there are “Sufficient funds” available to make strategic investment. Situations may also influence goals directly. For example, the fact “Increased competition” (threat) may hinder the fulfillment of the goal “To open sales channels”. Influences may also occur among situations. For example, “Low cost financing” (opportunity), caused by “Economic slowdown”, positively contributes to “High R&D expenditure” and “Healthy balance sheet”.
To choose the right indicators for a given object, be it a goal, process or product, one must have a good understand- ing of what is important to the organization. Moreover, this importance is generally contextual. For instance, indicators useful to a finance team may be inappropriate for a sales force. Because of the need to develop a good understanding of what is important, performance indicators are closely associated with techniques for assessing the present state of the busi- ness. A very common method for choosing indicators is to apply a management framework such as the Balanced Score- card [25], whereby indicators measure a range of factors in a business, rather than a single one (e.g., profits). Approaches such as Basili’s Goal-Question-Metric [26] are also avail- able to help identify and validate indicators measuring goal satisfaction.
In our example, some indicators are associated with the goals under “To increase sales”. Note that these indicators
are composite indicators, and may be further decomposed. For example, “Gross margin” may be broken down by prod- uct/service categories, fiscal periods, or geographical loca- tions. Also notice that although not shown, “Total sales” (in dollar amount) can be mathematically determined by “Sales volume” and “Gross margin”, entailing a hierarchical relation among these indicators. We return to the topic of composite indicators, including further examples, in Sect. 3.4.
Elicitation of business goals, relevant situations, and cur- rent indicators can be accomplished through a series of interviews, focus groups, or a review of available strategy documentation. Ideally, model construction would be iter- ative and participatory, involving business stakeholders at varying levels of the organization (e.g., management, techni- cal personnel) in a process of model construction and valida- tion. In practice, the number of relevant business goals and indicators may be large. Our Fig. 3 example is kept relatively simple for illustrative purposes, not reflecting realistic com- plexity. Uses of BIM in practice, including suggestions for modularization of BIM models to allow for scalability, are described in Sect. 5.
3 Reasoning with BIM models
Although the construction of a BIM model is useful as a means to clarify and communicate business objectives,
123
Strategic business modeling: representation and reasoning 1021
Table 2 Reasoning techniques applied to BIM including required information and corresponding paper section
Reasoning technique Required information Described in section
Goal model reasoning Initial reasoning values 3.2 Probabilistic decision analysis Conditional probability tables, Utility functions 3.3 Reasoning with indicators Atomic indicator values, Business formulae, Unit
conversion factors 3.4
Hybrid reasoning (reasoning with incomplete indicators)
Atomic indicator values, (optional) business formulae, (optional) unit conversion factors, (optional) initial reasoning values
3.4.7
strategies, and organizational situations, much of the ben- efits of BIM models come from the capability to support reasoning. Reasoning with BIM allows an organization to answer strategic or monitoring questions. For example, Best- Tech may want to pose the following questions:
• Should we develop technology in-house or acquire tech- nology through acquisition? Which option is better for maintaining revenue growth and reducing risks?
• Is it possible to maintain revenue growth while reducing risks? What strategies can achieve these goals?
• Given business metrics and target values, what increase in sales volume can be expected from the current number of sales channels and new promotions?
• Given the state of the business according to current data, will we be able to maintain revenue growth without new strategic partnerships or technology acquisitions?
In order to support a variety of analysis questions over BIM models, several types of reasoning approaches can be applied, including existing reasoning approaches for similar types of models and approaches making using of business metrics and indicators. The selection of a reasoning approach depends on the types of analysis questions posed, the method- ological phase, and the availability of specific information. An end-user may prefer a reasoning approach over others depending on the quantity of domain information that she/he possesses, or on the available time she/he has for encoding such information into the model. Table 2 summarizes the types of reasoning described in this paper, including a sum- mary of the information required for each procedure, and the paper section in which the reasoning approach is described. Earlier descriptions of each of the first three types of reason- ing were provided in [10–12]. We provide an overview of each type of reasoning in the next section.
3.1 Overview of reasoning approaches
Goal model reasoning. If a BIM model is constructed in a top-down manner, eliciting relevant goals and strate- gies before deriving or eliciting indicators, reasoning must
operate in the absence of indicator values. Such models, used as part of strategic planning, often results in alternative strate- gies. It is important to be able to analyze and comparing strategies at a high-level. Techniques that facilitate strate- gic analysis using enterprise goals have long been used as part of goal-oriented analysis [15, 19, 27–30]. These proce- dures propagate either qualitative or quantitative evidence through links in the model in order to evaluate the satis- faction of goals in the model given a particular strategy or target. Although quantitative propagation is supported, most techniques for goal model reasoning operate in the absence of concrete business measures, making them appropriate for high-level, strategic analysis in the absence of indicators. These approaches are suitable when the user is interested in an early analysis of the domain, exploring and improving the model while it may not yet be sufficiently complete or correct [29].
In this work, we select a particular goal reasoning tech- nique (introduced by Giorgini et al. and described in [19, 20, 31]) and demonstrate how this technique can be used to analyze alternative strategies in BIM models. Other proce- dures, such as those described in [15, 27–29] could be simi- larly adapted.
Probabilistic decision analysis. In some cases it may be possible to derive probabilistic information concerning the likelihood of goal achievement given the achievement of other model elements. This information can be collected from business experts, depending on their level of expertise and confidence, or from past statistical data. If such information is available, probabilistic decision analysis can be applied to BIM models, providing an alternative method for choosing among business strategies. During strategic planning, a strat- egy is normally produced by making decisions at a number of decision points. At each point, a decision option is chosen from a pool of available options. Probabilistic decision analy- sis facilitates automated decision making, selecting strate- gies, using conditional probabilities and utility functions.
Reasoning with Indicators. If BIM model construction is performed in a bottom-up manner, indicators and their data sources will be derived or elicited from the business. In this case, reasoning techniques must support reasoning over
123
1022 J. Horkoff et al.
Fig. 4 Alternative view of Fig. 3 separating goal/strategy and indicator hierarchies showing application of reasoning techniques
indicators, using current values to calculate composite indi- cators, indicators whose values are obtained from those of their components. These components themselves may also be composite, leading to an aggregation hierarchy of indicators. In order to support indicator reasoning, we wish to propagate values of indicators from a lower level in a hierarchy to ones higher up. In some cases, a business metric or mathematical function may exist to combine composite indicator values. Such formulae may or may not include unit conversion fac- tors. We describe methods which calculate composite indi- cator values using either available unit conversion factors or unit normalization, allowing for optional indicator weights.
In other cases, there is no well-defined mathematical func- tion that relates atomic indicators to a composite one. This might simply be due to lack of knowledge about the indica- tors, or the intrinsic nature of the indicators at hand. We adapt existing techniques for goal model reasoning for reasoning with indicators, deriving values of composite indicators, even when the relationship between a composite indicator and its components cannot be fully described using well-defined mathematical functions.
Hybrid reasoning: reasoning with incomplete indicators. As described in our example methodology (Sect. 2.3), model construction may typically consist of a mix of strategic planning and goal identification combined with the elicita- tion and understanding of existing business indicators. As a result, models will often not have indicators correspond- ing to all model elements. By combining techniques for rea- soning with indicators with existing reasoning techniques,
we can create a hybrid technique which supports reason- ing with incomplete indicators. Figure 4 shows an abstract view of our BestTech example, separating the strategic goal hierarchy from the indicator hierarchy. Here we can see that the indicator hierarchy is not complete with respect to the BIM model, and yet we would like to reason over the entire model. Existing reasoning techniques can be applied to the hierarchy on the left hand side, while indicator reasoning techniques can be applied to the right. Hybrid techniques, reasoning with incomplete indicators, can be used to bridge the gap between these views, propagating values from right to left (forward in the direction of the links) using quantitative normalization.
3.2 Reasoning with BIM models using goal modeling techniques
In this section, we explore application of an existing goal model reasoning technique in order to evaluate specific strate- gies and discover alternative strategies.
3.2.1 Evaluation of specific strategies
Goal-oriented requirements engineering has studied the problem of systematic exploration of alternative designs for achieving specified goals. In some cases, a manager has spe- cific strategies and she wants to compare them relative to given root goals (in order to eventually select one). A bottom- up/forward reasoning algorithm (e.g., [15, 19, 27–30]) starts
123
Strategic business modeling: representation and reasoning 1023
Table 3 Qualitative propagation rules from Giorgini et al. [19] (the (OR), (+D), (−D), (++D), (− − D) cases are dual w.r.t. (AND), (+S), (−S), (++S), (− − S) respectively)
(G2, G3) and�→ G1 G2 +S�→ G1 G2 −S�→ G1 G2 ++S�→ G1 G2 − −S�→ G1
Sat(G1) min
{ Sat(G2), Sat(G3)
} min
{ Sat(G2), P
} N Sat(G2) N
Den(G1) max
{ Den(G2), Den(G3)
} N min
{ Sat(G2), P
} N Sat(G2)
with an assignment of satisfaction values to some goals in a goal model. Such an assignment corresponds to a particular strategy to fulfill root goals. It then forward propagates these input values to the root goals, according to a set of pre-defined propagation rules.
Most goal-oriented analysis procedures support either qualitative or quantitative reasoning. Qualitative or quantita- tive values or label are assigned to or computed for each con- nected goal in the model. These values represent the level of positive and/or negative evidence received via relationships from other goals, which themselves have positive and/or negative evidence. Goal model propagation specifies what level of evidence propagated through what relationships pro- duces what resulting level of evidence. Procedures described in [15, 19, 28, 29] support a qualitative scale of satisfied and denied levels. For example, the procedure in [19] supports two variables over each goal: goals have satisfiability (S) values but also deniability ( D) values. During label propaga- tion, a goal can be both “partially/fully satisfied” (PS/FS) and “partially/fully denied” (PD, FD). For a goal, G1, these values are recorded by functions Sat(g1) and Den(g1), respectively. Sat and Den values belong to the set {N, P, F} (none, partial, full). For simplicity, the same information can be recorded using predicates FS, PS, PD, and FD, over goals, to repre- sent their level of satisfaction or denial (e.g., PS(g) being true corresponds to Sat(g) = ’P’). Typically, predicate labels are written on models as shorthand representing the level of satisfaction or denial in analysis results.
In this procedure, qualitative satisfaction and denial is propagated through the model using the semantics of model links (decomposition, contribution, etc.) For example, in
Table 3, the rule (G2, G3) and�→ G1 states how labels are propa-
gated when there is an AND-decomposition relation between goal G1 and sub-goals G2 and G3.
Contribution links between goals, as used in this proce- dure, can be negative or positive, symmetric or asymmetric, and can have varying strengths. Polarity and strength are rep- resented by a number of +/− symbols, specifically +, ++, −, and −−. In this case, symmetry refers to whether the links propagates only positive evidence (S), negative evidence ( D)
or both. For example, G2 −S�→ G1 states how labels are prop-
agated when there exists a weak asymmetric negative rela- tion between goals G2 and G1. If G2 is satisfied, then there is so evidence that G1 is denied, but if G2 is denied, then
nothing is said about the satisfaction of G1 (see [19] for fur- ther details.)
In order to use this procedure with the BIM syntax, we require a mapping between BIM and goal model concepts. The languages contain many overlaps, for example, goals and operationalization into processes. Giorgini et al. describe propagation over goals, but such propagation can be applied to the BIM concepts of situation, goal, and domain assump- tion. For the purpose of BIM concepts, propagation through AND/OR refinement can be treated in the same way as in existing goal reasoning techniques. The achieves link, can be thought of as a more specific type of AND/OR refine- ment. When applying reasoning, it is indistinguishable from AND/OR.
For the purpose of applying existing goal reasoning tech- niques to BIM, if influence links are qualitative logical influ- ences as described in Table 1, we can apply contribution link propagation rules directly to influence links. Quantita- tive propagation is discussed later in this section.
Propagation using indicators, including measures and evaluates links, will be explored in Sect. 3.4, on reasoning with indicators.
Running example. In the context of BestTech, we have asked: Should we develop technology in-house or acquire technology through acquisition? Which option is better for maintaining revenue growth and reducing risks? Consider the input assignment depicted in Fig. 5, where we choose “To maintain competitive advantage” by partially fulfilling the goal “To develop new technology in-house”; we also assume that we fully satisfy the goal “To increase sales”. Notice that goals with input assignment are shaded. Bottom- up reasoning propagates these input labels up the goal hier- archy all the way to the two root goals (see Sect. 4 for a description of the various tools which can be used to facil- itate this type of reasoning). When propagating evidence in this example, we treat all influence links as symmetric (prop- agating both positive and negative evidence). As described, each model element has two values (Sat and Den). However, we use the common shortcut of only displaying values which are not none (N ).
As we can observe from the result (shown in Fig. 5), this strategy leads to “To maintain revenue growth” being par- tially satisfied, while “To reduce risks” is both partially sat- isfied and partially denied, producing a conflicting situation
123
1024 J. Horkoff et al.
Fig. 5 Bottom-up goal reasoning example
for this goal. The analyst can use these results to investi- gate the cause of the conflict, looking for strategies which effectively achieve the “To reduce risks” goal.
Quantitative propagation. Figure 5 demonstrates a partic- ular qualitative reasoning for goal modeling [19] adapted for BIM. Several existing procedures provide similar prop- agation using quantitative scales. For example, in Giorgini et al. [30], an expansion of the procedure in [19], satis- faction and denial values each range from 0 to 1. Here, positive or negative links (+, −) are assigned a quantita- tive strength or weight, also between 0 and 1. Propagation rules are adjusted to accommodate quantitative values, using link strengths and min/max operators over AND/OR links. This procedure intends for the resulting numbers to represent probabilistic values, as in a Bayesian network. If it were to be applied to BIM models, quantitative probabilistic indicators, as described in Table 1, would be used.
In the quantitative approaches to goal model reasoning, including the Giorgini et al. approach, the sources of ini- tial quantitative satisfaction and denial values, as well as the numeric link weights, is controversial. Giorgini et al. advo- cate collecting the numbers from domain experts, but the accuracy and meaningfulness of such numbers have been questioned [32, 33]. There is a need to ground these num- bers in realistic business measures. We address this issue by attaching indicators to goals and other model elements, grounding quantitative data in real data. We describe the use of indicators in BIM reasoning in Sect. 3.4.
3.2.2 Discovering alternative strategies
In other cases, a manager may be interested in finding possible viable alternatives within a model, given certain con- straints. Given a goal model and an assignment of desired sat- isfaction values (either qualitative or quantitative) to its root goals, a top-down/backward reasoning algorithm [30, 34] can look for an assignment (strategies) to leaf goals, processes and domain assumptions that lead to the desired satisfaction values of those root goals.
Running example. We have asked: Is it possible to main- tain revenue growth while reducing risks? What strategies can achieve these goals? We add as targets to our two root goals, FS(“To maintain revenue growth”) i.e., we strongly desire revenue growth, and PS(“To reduce risks”). The algo- rithm generates a possible strategy, shown in Fig. 6, which makes the required assignments true (here, the root goals with required assignments are shaded).
In this strategy, for example, the goal “To establish strate- gic partnership” is preferred to “To invest in new technolo- gies”, while the goal “To offer promotion” is preferred to “To open sales channels”. However, similar to our bottom- up reasoning, the selection of these alternatives produces conflicting results for “To reduce risks” and “To reduce external dependencies”. Had we not wanted such conflicts, we could have also tried to specify ¬ PD(“To reduce risk”), though there may be no solution that satisfies these conditions.
123
Strategic business modeling: representation and reasoning 1025
Fig. 6 Top-down reasoning
3.3 Probabilistic evaluation of strategies
Decision analysis techniques rest on an empirically verified assumption that humans are capable of framing a decision problem, listing possible decision options, and quantifying uncertainty, but are rather weak in combining information into a rational decision. An influence diagram [35] is a deci- sion model that supports decision-making by computing the expected utility value of each decision option; the option with the highest utility value is optimal and should be cho- sen by the decision maker. Influence diagrams include: deci- sion nodes for specifying decision options, chance nodes for quantifying uncertainty, and value nodes for quantifying user preferences using utility functions.
To carry out decision analysis, a BIM model needs to be projected into an influence diagram. Such projections require a set of translation rules. We summarize these rules here:
• Goals, processes, domain assumptions and situations are translated to Chance Nodes, meaning that they are they represent uncertain quantities relevant to the deci- sion problem; the uncertainties quantified by conditional probability distributions.
• Each AND-decomposition link is translated into an influ- ence link if the target node is a goal; it is removed if the target is a domain assumption (since its truth value is not influenced by whether the parent goal is been pursued or not). Also notice, an AND-decomposition is not really a
decision problem; however its translation is necessary in case it belongs to an OR-decomposed sub-tree.
• Each OR-decomposition link in BIM is translated into a decision node.
• Influence links among domain assumptions and situa- tions, as well as those having goals as the source or tar- get node, are translated into influence links. Note that the strength of the original influence link will not be used in the influence diagram; instead the user needs to define the conditional probability table for the source and target nodes.
Running example. We outline how probabilistic reasoning would work for a BIM model using a subset of our example. In Fig. 7, we focus on a subset of the goal model rooted at the goal “To increase sales”. The nodes directly involved in the decision analysis are shaded. In particular, to pursue the goal “To increase sales volume”, two decision options are avail- able (according to the OR decomposition in the original goal model): pursue either the goal “To open sales channels” or “To offer promotions”. We have previously asked: Should we develop technology in-house or acquire technology through acquisition? Here we show an alternative means to answer this analysis question. Notice that the goal “To maintain gross margin” and the domain assumption “High demand” are involved in this analysis through influence links. A deci- sion analysis amounts to answering the question: which of these two sub-goals should be pursued in order to maximize
123
1026 J. Horkoff et al.
Fig. 7 Decision analysis
expected gain for the indicators “Sales volume” and “Gross margin” (which could then be combined into “Total sales”)?
To reason with an influence diagram, we need to esti- mate conditional probabilities of various events, such as high demand (domain assumption) for a certain product given eco- nomic slowdown (its parent node), or high sales volume given that we open a sales channel vs. we offer a promotion; these events directly or indirectly affect the outcome of a decision. The possible outcome of a chance node is not restricted to be binary. For example, we can talk about “high demand” being true, partially true or false. These probabilities are rep- resented by a conditional probability tables (CPTs) for each event; the one for the event “High demand”, given its parent event “Economic slowdown”, is shown in Fig. 7. To support this form of reasoning, influence links must be probabilistic with strengths corresponding to the conditional probabilities in the probability tables.
In addition to CPTs, we also need to specify the utility function of each value node. A utility function introduces a measure of preference by mapping possible outcomes of a decision process on the set of real numbers. For exam- ple, the utility function for “Sales volume” given “To open sales channels”, “To offer promotions” and “High demand” is shown in Fig. 7. In particular, it maps {“High demand” = high, “To open sales channels” = FS; “To offer promo- tion” = FS} to the number 40,000, while {“High demand” = high, “To open sales channels” = FS, “To offer promotion” = PS} to 35,000. Notice it is the relative ordering of the utility function values that are important, not the absolute values. However, if values are very close (e.g., 35,000 and 34,950),
this may indicate that the models should be expanded to bet- ter differentiate between outcomes.
Given the CPTs for each event and the utility functions for each value node in the model, the algorithm produces utility values for the value nodes for all possible decision options (in our example, we have two decision options: “To open sales channels” and “To offer promotions”). The results show that the utility for “To open sales channels” is 11,100, which is higher than the 3,700 for “To offer promotions”. In this example, the option “To open sales channels” is prefer- able over “To offer promotions” as far as “Sales volume” is concerned.
3.4 Reasoning with indicators
In this section, we describe how indicators, linking to busi- ness data, can be used in analysis, including performance levels for indicators, composite indicators, and a variety of techniques for reasoning with indicators as part of BIM. Reasoning techniques include use of mathematical equations derived from business metrics to propagation rules derived from model structure, making use of unit conversion or nor- malization when needed. This section is an expansion and reorganization of material presented in [11, 12].
3.4.1 Indicator performance levels
As described in Sect. 2.2.4, an indicator is a measure, quan- titative or qualitative, of the progress or degree of fulfill- ment of organization goals. The subject of an indicator is
123
Strategic business modeling: representation and reasoning 1027
Fig. 8 Indicator goal example with performance region
a particular feature or quality of an element in the business environment, e.g., the workload of an employee, or the com- pliance of an internal process with respect to external reg- ulations. To express why an indicator is needed, we rely on the evaluates relation. For example, the indicator “Sales vol- ume” is needed (why) to evaluate the goal “To increase sales volume”.
Each indicator, has a current value (cv) which is evaluated against a set of parameters: target (value), threshold (value) and worst (value) [22]. The result of such an evaluation is a normalized value (ranging in [−1, 1] ⊂ R), which is often referred to as the performance level for an indicator.
Note that a current value can be assigned by: (i) extracting it at run-time from back-end data sources, (dimensions and levels [22] can be used to filter data from data warehouses); or (ii) supplied by users to explore “what-if” scenarios; or (iii) calculated by a metric expression in the case of composite indicators, as explored in Sect. 3.4.4.
An indicator can be positive, negative, or bidirectional, meaning that we want to maximize, minimize or balance its target. Performance regions are defined for each type of indicator by properly combining the indicator’s parameters. Figure 8 shows an example of performance region for a pos- itive indicator, i.e., we want to maximize sales volume, in which Target ≥ Threshold ≥ Worst value.
The relative position of indicator current values within such regions allows calculation of the performance level for an indicator, as shown in Fig. 9. Notice how the worst and target values are mapped respectively to −1 and +1, while the threshold value is mapped to 0. A linear interpolation is used to approximate performance levels, as also described by Eq. (1) [36]:
pl(current v.)
=
⎧⎪⎪⎨ ⎪⎪⎩
|current v. − threshold v.| |target v. − threshold v.| , if current v. ≥ threshold v. |current v. − threshold v.| |threshold v. − worst v.| , if current v. < threshold v.
(1)
Fig. 9 Example of interpolation [36] to calculate performance levels
Other forms of interpolation can be used, e.g., polynomial, spline, etc. For instance, the performance level (pl) for Fig. 9, given a threshold of 100 (thousands of dollars in sales) and a target of 210 is:
pl (155) = |155 − 100||210 − 100| = 0.5
Performance levels are, in turn, propagated to the corre- sponding goals to evaluate satisfaction levels. For example, in Fig. 8, the performance level 0.5 is propagated to the sat- isfaction level of the corresponding goal which, in turn, is mapped to a “partial satisfied” state (orange color).
Indicators can also be used to evaluate situations in a sim- ilar way as for goals, by propagating a performance level of an indicator to the occurrence level of the situation evalu- ated by the indicator. For example, the indicator “Number of competitors” can measures occurrence level of the situation “Increased competition”.
3.4.2 Composite indicators
In a BIM model, indicators are associated with various busi- ness elements (e.g., goals, situations). These elements are generally composite, consisting of hierarchies of elements. Such structure implies hierarchies for indicators. For exam- ple in Fig. 10, the goal hierarchy results in a hierarchy for associated indicators. More specifically, “Number of sales channels” and “Number of Promotions” are atomic indica- tors of “Sales volume”’, since they evaluate goals which are a sub-goal of “To increase sales volume”.
An alternative way to represent Fig. 10 is shown in Fig. 11, where the mirroring hierarchies of BIM concepts and indica- tors are shown separately. In this work, we follow the style of Fig. 10 in order to make the figures more visually compact.
123
1028 J. Horkoff et al.
To open sales channels (g2)
To increase sales volume
(g1)
To offer promotions (g3)
Sales volume
(i1)
OR
Number of New Sales
Channels (i2)
Number of Promotions
(i3)
Increased Competition
(s1)
-
Number of New
Competitors (i4)
Fig. 10 Example composite indicator and goal hierarchy
We have not yet considered how composite indicators should be combined (represented by “?”). Techniques to calculate the values for composite indicators are explored in the rest of this section and in the next one.
An alternative way to capture composite indicator val- ues would be to avoid composite values and apply com- bined atomic indicators values directly to parent elements (goals, situations, etc.) as input for reasoning. For example, in Fig. 10, the composite value derived from combining the indicators “Number of sales channels” and “Number of Pro- motions” could be applied as the input analysis value for the goal “To increase sales volume”. However, business met- rics often already form a hierarchy of compositions, and it is useful to retain this structure, keeping it separate from the conceptual business elements. We wish to explicitly differ- entiate between the goals and processes of the business and the metrics which evaluate the satisfaction levels of these elements; and we wish to make this differentiation not only at model leaves, but at all levels of the model. For example, “To increase sales volume” is a business goal which differs from the measure “Sales volume”. Keeping this separation emphasizes the links between the BI model and realistic data, differentiating the framework from existing intentional mod- eling frameworks.
In the rest of this section, we focus on algorithms that prop- agate values of indicators from a lower level in a hierarchy to higher-level indicators. In other words, we introduce a vari-
ety of ways to describe the combination of atomic indicators (the “?” in Fig. 11). These algorithms bear similarity to the label propagation in goal reasoning summarized in Sect. 3.
3.4.3 Indicator reasoning with varying levels of information
In an ideal case, all relations between atomic and composite indicators will be fully described using mathematical equa- tions derived from business metrics. We call such equations business formulae, formulae used to calculate desired quan- tities using available data. Such formulae make computation of composite indicators computationally simple, precise, and automatable. For example, profits can be calculated directly from revenues and costs.
Depending on the units of atomic indicators, business for- mulae may require unit conversion factors. For example, costs may be calculated from both unit costs and employee hours, which can be converted into a currency value. In other cases, if unit conversion factors are not available, val- ues can be converted to the same scale using normalization into a set range, producing performance levels as described in Sect. 3.4.1. We describe these methods in Sects. 3.4.4 and 3.4.5, respectively
Collecting precise domain formulae can require much effort, especially if business formulae do not already exist in the organization. In less ideal cases, when business formulae do not (yet) exist, indicator values have to be derived using estimation/approximation techniques based on the structure of a BIM model. Although such techniques may produce results which are less precise, their use allows reasoning with incompletely defined indicators. We describe such a tech- nique in Sect. 3.4.6. We can further consider the case when indicators are incomplete with respect to concepts in a BIM model, i.e., not all concepts are (yet) measured via an indi- cator. We describe a hybrid technique to allow for reasoning with incomplete indicators in Sect. 3.4.7.
We classify different levels of propagation into four cat- egories, as described in Table 4, based on the availability of unit conversion factors and the information required for reasoning. The first column describes the type of indicator reasoning. The second column describes the means of unit conversion, either not required, using unit conversion fac-
Fig. 11 Alternative representation of Fig. 10 showing separate hierarchies of goals and indicators
To open sales channels (g2)
To increase sales volume
(g1)
Sales volume (i1)
OR
Number of Sales
Channels (i2)
Number of Promotions
(i3)
Increased Competition
(s1)
-
Number of Competitors
(i4)
? ? To offer
promotions (g3)
123
Strategic business modeling: representation and reasoning 1029
Table 4 Classification of indicator reasoning technique based on unit conversion and required information
Indicator reasoning without Business Formula
Unit Normalization (Performance
Levels)
Atomic Indicator Values
3.4.6
Hybrid Reasoning
(with Incomplete Indicators)
Qualitative Normalization
Atomic Indicator Values, (Optional)
{Business Formulae, Unit conversion factors, Initial
Reasoning Values}
3.4.7
tors, or quantitative or qualitative normalization. The third column describes the type of information required from the organization in order to use this type of indicator reasoning. The last column indicates in what sub-section of the paper this type of reasoning is described.
3.4.4 Indicator reasoning using business formulae and unit conversion
When business formulae which combine atomic indicators exist or can be reasonably derived, these expressions should be used to derive the values of composite indicators. Such formulae should take into account values of atomic indica- tors associated with sub-goals (sub-elements). For example, in Fig. 10, the atomic indicators associated with the sub- goals “To open sales channels” and “To offer promotions” all contribute in some way to the composite indicator, “Sales volume”. In this case, our knowledge of the example enter- prise tells us that the atomic indicator values can be aggre- gated together, as each value can make a positive, additive contribution to the sales volume. Other indicator values may be combined using a variety of mathematical operators (e.g., division, multiplication). If, for example, all atomic indica- tors used the same unit (e.g., time in hours), then no unit conversion is necessary, and the values could be combined with an appropriate metric elicited from the enterprise. How- ever, in this case, each indicator is measured in a different unit (number of channels, number of promotions), and cannot be summed directly.
The definition or elicitation of equations calculating the values of composite indicator will often require some form of unit conversion. We account for this conversion via the elicitation or definition of a conversion factor for each atomic indicator having a different unit of measure.
Unit conversion example. For example, consider the two simple indicators “Employee cost” and “Working time” (not in our example model). Specifically, “Employee cost” can be defined as a composite indicator whose value relies on the atomic indicator “Working time”. In order to calculate the composite metric, we need to convert “Working time”
values measured in hours into “Employee cost” units. One possible conversion factor is to take the average of the wage per hour for all employees. Assuming that such an average is $20/h and that the current value for “Working time” is 160 h, we can calculate an approximated current value for Employee cost as:
1. 20 dollars = 1 h → $20h = 1, where 20 is the conversion factor (cf)
2. 160 h · $20 h = $3, 200
We will refer to a conversion factor using a function that maps the indicator being converted to the composite indica- tor using the conversion: cf(source, destination). Notice that in many cases a conversion factor is an estimate based on previous experience or statistics. For example, the average wage per hour could be $30 instead of $20 for a different company.
Running example. In our Fig. 10 example, elicitation in the particular enterprise could reveal that during a certain time period each sales channel produces, on average, an addi- tional $20,000 in sales, and that each promotion produces, on average, $7,000 in sales. Thus, if sales volume is measured in thousands of dollars, the conversion factors for each sub- goal are 20 and 7, respectively. Potentially, an indicator could have multiple conversion factors, if, for example, it influences or is a refinement in more than one expression measuring a composite indicator. The conversion factors (cf) for our example subgoals are cf(i2, i1) = 20 and cf(i3, i1) = 7, respectively.
When conversions are impossible, e.g., converting gal- lons to square feet, we have to fall back to a “normal- ized” approach or to a “qualitative” one; these are presented, respectively, in Sects. 3.4.5 and 3.4.6.
Once the conversion factors have been determined, the expression representing the business metric can be con- structed. In this case, “Sales volume” is calculated by:
cv(i2) cf (i2, i1) + cv(i3) cf (i3, i1) = 20cv(i2) + 7cv(i3) where cv(i) is the current value of an indicator, i , derived from data sources. In addition to factors from indicators associated with sub-goals, we must consider other business elements such as goals or situations that influence the goal. In the previ- ous example, we have the situation “Increased competition”, which influences negatively the “To open sales channels” goal. The “Increased competition” situation is evaluated by the indicator “Number of competitors”. We can use a math- ematic expression to capture the influence of the situation on the goal by expressing how the indicator associated with the situation (i4) affects the value for the indicator asso- ciated with the goal (i2). Elicitation within the enterprise could reveal that each competitor reduces the number of sales
123
1030 J. Horkoff et al.
Fig. 12 Example of reasoning with conversion factors and weights (cf = conversion factor, cv = current value, w = weight)
channels by, on average, two channels (cf(i4, i2) = 2). This parameter must be chosen accurately by the designer who must rely on her/his domain experience and/or estimates of historical data. We can express the combined effect of situa- tion s1 on goal g2 using the following expression:
cv(i2) − cv (i4) cf (i4, i2) = cv(i2) − 2cv(i4) In particular, i4 is the current value of the indicator “Num-
ber of competitors”, and cf(i4, i2) is the conversion factor (in this case, 2) used to convert the indicator value into a number of sales channels (the unit of i2).
Combing formulae, the value for the composite indicator “Sales volume” can be expressed via:
(cv(i2) − cv(i4)cf(i4, i2)) cf (i2, i1) + cv(i3) cf (i3, i1) = 20 (cv(i2) − 2cv(i4)) + 7cv(i3) This expression is used in Fig. 12 to calculate the cur-
rent value of the “To increase sales volume” indicator, cor- responding to our desired BestTech analysis question: Given business metrics and target values, what increase in sales volume can be expected from the current number of sales channels and new promotions? In this example, the corre- sponding current value for each indicator is extrapolated from the data sources. In particular, for the indicator “Number of sales channels” we have a current value of 10.
Indicator weights. The effect of goals or influencing situa- tions can be further refined by adding weights or importance to the value of each element. These optional values could be derived, for example, from domain experts using a prioritiza- tion elicitation method such as AHP [37]. As with conversion factors, indicators could have multiple weights if their values are used as part of the expression for more than on composite indicator. We express these values as w(source, destination);
for example the weight for “Number of sales channels” used to compute indicator “Sales volume” is w(i2, i1). The influ- ence from sub-goals and situations, including conversion fac- tors and weights, can be used to compute a final value for the composite indicator. In our Fig. 10 example, the final equa- tion is as follows:
(cv(i2) − cv(i4)w (i4, i2) cf (i4, i2)) w (i2, i1) cf (i2, i1) + cv(i3)w (i3, i1) cf (i3, i1) Although, in this case we sum indicator values, the
designer can customize each expression depending on the influence of situations and sub-goals as elicited from the domain. We allow users to express rich and flexible expres- sions capturing business formulae using the off-the-shelf grammar of the Jep Java Library (see Sect. 4).
3.4.5 Indicator reasoning using business formulae and performance levels
When conversion factors are not available, current values for composite indicators can be derived using range normaliza- tion, which takes values spanning a specific range and repre- sents them in another range. In the case of indicators, we take the current indicator value and convert this value to a perfor- mance level, as described in Sect. 3.4.1. Once performance levels have been produced for the required indicators, these values can be combined using business metrics, as described in the previous section. As with reasoning using conversion factors, other, more complex metrics could be used to com- bine results for composite indicators. After applying each formulae, results which lie outside the [1, −1] range must again be normalized.
Running example. In our previous example for calculating the “Sales volume” composite indicator, if the conversion factors were not available, the current values for “Number of sales channels” and “Number of Promotions” would be converted to a normalized scale, producing performance lev- els for each indicator. These values would then be combined using the appropriate business formula, in this case, summa- tion. When considering the effect of the situation “Increased competition”, the “Number of competitors” and “Number of Sales Channels” indicators would be converted into their cor- responding performance levels, with the former subtracted from the latter, as described in the business metric. We show the combined results for “To increase sales volume” using business formulae with performance levels in Fig. 13. In this example, performance levels are calculated using cur- rent indicator values, as used in Fig. 12, as well as associated target and threshold values. For example, the current value of 10 sales channels is converted to a performance level (pl) of 0.8 using the target and threshold values of 12 and 2, respectively.
123
Strategic business modeling: representation and reasoning 1031
Fig. 13 Example of reasoning with indicator normalization and business metrics
3.4.6 Indicator reasoning without business formulae
In cases where business formulae are not available to cal- culate the values for composite indicators, the structure of the BIM model mirroring the hierarchy of indicators can be used to derive formulae to calculate composite indicators following, for example, the propagation rules of Giorgini et al. [30]. In order to allow for this style of reasoning, which separates positive and negative evidence, we associate two variables to each indicator: positive performance ( per +) and negative performance ( per −), similar to the S (satis-
fied) and D (denied) values introduced previously. These variables represent the current evidence concerning perfor- mance or non-performance of an indicator i.The current val- ues of atomic indicators as read from the data sources are normalized (computing performance levels) as described in Sect. 3.4.5, producing values in the range [−1, +1]. These values are converted to the two variables per + and per − according to the mapping in Table 5. The inputs to this table are a target, current value, threshold value, and worst value for each indicator, while the outputs are per + and per − values for each indicator. For quantitative performance, the variable | pl| represents the absolute value of the performance level produced from the rules specified in equation (1) from Sect. 3.4.5. For each indicator, the per+ variable is mapped to a traffic light with a plus symbol on the top, while a minus symbol is used for the per − variable. Table 5 describes the mapping for positive indicators only. Similar mappings can be made for negative or bidirectional indicators. Values could also be mapped to the corresponding qualitative scale, as described in Giorgini et al. [30].
P er + /− values are propagated up the hierarchy of indi- cators using propagation rules from Giorgini et al., translated into indicator terminology in Table 6.
Our adopted goal reasoning technique propagates positive and negative evidence separately, allowing the notion of con- flicting evidence for goals. Applying this reasoning approach to indicators means that an indicator can be at the same time
Table 5 Mapping rules (t = target value, cv = current value, th = threshold value, w = worst value, pl = performance level)
Input mapping rule
Resulting performance variables ( per +, per −)
Resulting evidence Resulting indicator color ( per +, per −)
Quantitative Qualitative
cv ≥ t (1.0, 0.0) (full, none) Fully performant
th < cv < t (|pl|, 0.0) (partial, none) Partially performant
cv = th (0.0, 0.0) (partial, none) Partially performant
w < cv < th (0.0, |pl|) (none, partial) Partially non-performant
cv ≤ w (0.0, 1.0) (none, full) Fully non-performant
123
1032 J. Horkoff et al.
Table 6 Indicator propagation rules using model structure
(i2, i3) and�→ i1 i2 w+S�→ i1 i2 w−S�→ i1 i2 ++S�→ i1 i2 − −S�→ i1
per +(i1) per +(i3) ⊗ per +(i2) per +(i2) ⊗ w N per +(i2) N per −(i1) per −(i3) ⊕ per −(i2) N per +(i2) ⊗ w N per +(i2) The OR, (+D), (−D), (+ + D), (− − D) cases are dual w.r.t. AND, (+S), (−S), (+ + S), (− − S) respectively. See [30] for details
Fig. 14 Example of quantitative reasoning with indicators using model structure
“fully performant” and “fully non-performant”. Supporting such conflicts as part of our approach to reasoning with indi- cators allows the analyst to see the presence of conflicting evidence. This can be especially useful as part of initial BIM model development, with conflicting early analysis results potentially resolved by adding metrics derived from the enter- prise. In this way, such reasoning can direct the elicitation of business metrics for contentious areas of the model.
Running example. We return to our previous BestTech analysis question: Given business metrics and target values, what increase in sales volume can be expected from the cur- rent number of sales channels and new promotions? Current values and performance levels are taken from the scenarios in Figs. 12 and 13. For example, the current value of ten sales channels is mapped to a performance level of 0.8, meaning that per+ starts at 0.8 as per our mapping rules in Table 5. This indicator is “partially performant” (green-orange).
By applying the same procedure for all the atomic indica- tors, we obtain the per +/− values shown in Fig. 14. The next step is to rely on the propagation rules described in Table 6 to propagate and calculate the per +/− values of the “Sales volume” indicator.
First, we consider the propagation of the values for the “Number of competitors” (i4) indicator to the “Number of sales channels” (i2) indicator through the negative influ- ence link. In the quantitative procedure, such links require a numeric weight. In this case, the link can be treated as a quantitative logical link and assigned a value of 0.5.
Notice that, the influence from the “Increased competi- tion” situation has a minus (−) symbol. As described by Giorgini et al. [30], this is a symmetric relation and it is a shorthand for the combination of the two corresponding
asymmetric relationships i4 −S�→ i2 and i4 +D�→ i2 (the propaga-
tion rule for the latter is dual w.r.t. the former); this means that both satisfiability and deniability are propagated. Therefore, after propagation, we obtain:
per + (i2) = min (
per − (i4) ⊗ w, per + (i2) )
per − (i2) = max (
per + (i4) ⊗ w, per − (i2) )
Thus the i4 values of (per+ = 0.2, per − = 0.0) are propa- gated to the per + /− values of i2 by using the operator with the influence weight (0.5). This results in a value of (0.25, 0.0). Combining this with the initial value for i2 of per + /− of (0.8, 0.0) (using the min operator), we have a value of (0.25, 0.0) for i2.
We then propagate the two indicators associated with the corresponding sub-goals by relying on the AND rule in the first column of Table 6. As the OR rules are dual with respect to the AND rules, the per+ and per − variables are assigned the maximum and minimum values amongst the sub-indicators. The resulting propagation rules are:
per + (i1) = max (
per + (i2), per + (i3) )
per − (i1) = min (
per − (i2), per − (i3) )
Given the per +/− value of (0.0, 0.1) for i3, this produces a final value of (0.25, 0.1) for i1. Results are shown in Fig. 14.
In this example, a conflict is discovered for the composite indicator “Sales volume” when values are propagated fol- lowing the rules in Table 6. When such conflicts appear in a model, although undesirable, they do help to highlight par- ticular aspects of a business that need user attention because of possible inconsistencies.
3.4.7 Hybrid reasoning: reasoning with incomplete indicators
Users may want to reason over a model in its intermedi- ate stages, before complete indicators have been added. For this purpose, we introduce a forward reasoning procedure, as in Sect. 3.2.1, supporting analysis of current indicator data potentially combined with “what if?” alternative decisions. Future work can investigate how backward or probabilistic
123
Strategic business modeling: representation and reasoning 1033
Fig. 15 An example of reasoning with incomplete indicators
reasoning could be applied (i.e., what indicator values would support target satisfaction levels? or what decisions should be made given current indicator values?)
To support reasoning with incomplete indicators, we rea- son with the available indicator hierarchy, using business for- mulae, conversion factors, or performance levels depending on the availability of complete business metrics and con- version factors. When business formulae are not available, goal model propagation is applied between indicators, as described earlier (3.4.6). Generally, we use the most pre- cise source of information available, following the technique ordering in Table 4.
Once indicator reasoning has produced current values for all atomic and composite indicators, the values of the indicators are transferred onto the evaluated element (via the evaluates link) in the BIM model. At this point, goal model reasoning as described in Sect. 3.2.1 begins, using values transferred from indicators as initial reasoning values. If analysis results from indicator reasoning techniques differ from analysis results using goal reasoning techniques, prop- agation through indicators is favored.
In order to aim for reasoning that covers the entire BIM model, it may be necessary for the enterprise analyst to pro- vide “what if?” leaf-level values for model elements not directly or indirectly connected to indicators. If the analyst chooses to use qualitative values for such elements, incoming values will be both qualitative (manually set) and quantita- tive (from indicators). When propagation results in values
which are both qualitative and quantitative, quantitative val- ues can be converted to the qualitative scale (see Table 5), and reasoning can continue using qualitative values.
Running example. We provide a detailed example of this reasoning with partial indicators, with results shown in Fig. 15. Here, we have taken the reasoning with indicator results from Fig. 12 using business metrics and unit con- version. We transfer the results of this analysis to the cor- responding evaluated BIM model elements. Mappings from current indicator values to per + and per − values can be applied using the rules described in Table 5. When transfer- ring performance values from indicators to satisfaction levels of evaluated BIM elements, per +/− values are transferred to Sat/Den values, respectively. Values transferred onto the BIM model elements are then propagated through the model using goal reasoning techniques as described in Sect. 3.2.1. Such propagation can be either quantitative or qualitative. In our example model, when quantitative propagation is used, we assume that all influence weights are +/− 0.5 for + and − links, and +/− 1.0 for ++ and −− links, respectively.
In order for reasoning results to be complete, we place initial analysis values on leaf elements not connected to indicators, in this case we follow the bottom-up reasoning example in Fig. 5 and place initial values of FD, PS, and FD on “To establish strategic partnership”, “To develop new technology in-house”, and “To acquire technology through acquisition”, respectively (shaded in grey). We have also
123
1034 J. Horkoff et al.
added initial values to the “Gross margin” indicator and the “High demand” domain assumption. Through these values we answer our final example BestTech analysis question: Given the state of the business according to current data, will we be able to maintain revenue growth without new strategic partnerships or technology acquisitions?
In our example, qualitative and quantitative values are combined when evaluating “To maintain revenue growth” and “To reduce external defence”. In some cases, results from indicator reasoning and goal modeling techniques are not consistent.
For example, the OR relationship between “To open sales channels” and “To offer promotions” have received analy- sis values through indicator propagation using a business metric indicating summation of these values (producing a value of pl = 0.5), while the OR structure in the model leads to max/min propagation (producing a value of (0.25, 0.1). In this case the values from the indicator reasoning are retained, as these values are derived directly from business formulae.
Results in our example show conflicting values for the high-level goals of “To reduce risks” and “To maintain rev- enue growth”. This is in contrast to our Fig. 5 bottom-up reasoning example, where only “To reduce risks” had con- flicting values. In this case, use of indicator values reveals further conflicts and more detailed analysis results.
4 Tool support
Goal model reasoning and probabilistic decision analysis. In Sect. 3 we described how several existing goal and proba- bilistic reasoning techniques can be applied to BIM models. These techniques are supported by a variety of existing tools. Such tools can be used to approximately represent and then reason over BIM model. For example, the GR-Tool allows forward and backward qualitative and quantitative goal rea- soning as described in [19, 20, 30] and as used in the examples in Sect. 3.2. jUCMNav allows forward qualitative and quan- titative goal reasoning producing single analysis values that combines positive and negative evidence [38]. Indicators and indicator aggregation are also supported in jUCMNav. Prob- abilistic reasoning is supported by the GeNIe decision analy- sis tool, used to derive analysis results in Sect. 3.3. Figure 16 shows a view of several of these tools analyzing a model sim- ilar to the running example in Fig. 3. These screenshots are meant to give an idea of the types of interfaces and analyses that are possible with such tools. Details of the model can be seen more clearly in Figs. 1 and 3.
An early version of BIM has been implemented in the ADOxx metamodeling-based development and configura- tion environment [39]. Once a BIM model has been con- structed in ADOxx, it can be projected onto various analysis
models. Currently we support exporting a BIM model as a goal model (containing only intentions and their relations) and as a Bayesian network (containing only situations or indicators and their relations), so that formal goal reason- ing and probabilistic inference can be carried out using the GR-Tool [40], and GeNIe [41], respectively.
Reasoning with indicators. The ADOxx tool supports rea- soning with indicators by supporting queries over BIM mod- els via the AQL querying language provided with ADOxx [39].
Parallel to ADOxx development, we have implemented a visual editor prototype to draw BIM models and support indicator reasoning techniques described in Sect. 3.4. Our implementation uses Graphiti [42], an Eclipse-based graph- ics framework that enables easy development of state-of-the- art diagram editors for domain-specific modeling languages. The current version of the prototype implements the quantita- tive approach described in Sect. 3.4.4 by relying on Jep [43], a Java library for parsing and evaluating mathematical expres- sions. Jep supports strings, vectors, complex numbers and boolean expressions. We are working to expand the tool to support other types of reasoning, including a combination of indicator and goal reasoning techniques as described in Sect. 3.4.7.
Figure 17 provides a snapshot of the tool. Marker “A” highlights the BIM model and the toolbar containing business element constructs. Marker “B” highlights the property panel containing indicator parameters and current value. Marker “C” highlights the property panel containing the definition of the metric expression (notice available variables such as strengths, conversion factors, etc.).
5 Ongoing case studies
BIM concepts have been applied in several ongoing case studies. One study in the healthcare sector focuses on intro- ducing and improving business intelligence systems in a hospital, in order to use the wealth of data that the orga- nization produces to evaluate the quality and efficiency of its processes, the utilization of its resources, and the out- comes of its operations. The case focuses on a critical process within the hospital Emergency Department. From a research perspective, the study aims to evaluate the utility of BIM to design practitioners. Lessons learned include the ability of BIM concepts to enhance communication and collabora- tion between designers and domain experts, and to reduce common project risk factors that a BI solution may face dur- ing its lifecycle. BIM models functioned effectively both as communication tool and design blueprint. Case study out- puts include a methodology for designing dimensional BIM models, relating to existing or new data sources. The study
123
Strategic business modeling: representation and reasoning 1035
Fig. 16 Screenshots of several available reasoning tools (top GR-tool, middle jUCMNav, bottom GeNIe)
123
1036 J. Horkoff et al.
Fig. 17 Graphiti visual editor and reasoning tool
developed an approach to modular creation of BIM models in order to deal with the scale of a real-world problem. Specif- ically, BIM concepts were organized into separate views: Goal/Strategy Map, Indicator Map, Process Map, Process Workflow, Resource Map, Actor Map, and individual Actor Goal Indicator Object Graphs (AGIO). Views could be com- bined as needed. The AGIO graphs focused on individual combinations of a single goal, indicator and actor, and were created using corresponding AGIO sheets (tables). An in- depth description of intermediate case study results and meth- ods can be found in [44].
An additional study focused on using an earlier version of BIM in a large teaching hospital in order to manage an organizational transformation initiative to reduce antibi- otic resistant infections (RARI) by changing the use of antibiotics. This approach used BIM along with Conceptual Integration Models in order to more explicitly describe the mapping between BIM business concepts and the structure of existing data, providing examples of how current indicator values can be derived through queries over available data. A more detailed description of the method and study can be found in [9].
A further case study, currently in its early stages, is using BIM models as part of a framework to design, deploy, and build an infrastructure for organizational innovation in the Ericsson Corporation. In this context, BIM is intended to make the goals behind innovation efforts explicit, cap- ture barriers to innovation through situations, and moni- tor the readiness and success of innovation efforts using indicators.
6 Related work
BIM. Previous work has introduced early versions of the BIM language, describing concepts and usage scenarios [8–12]. We have consolidated and expanded this work, pro- viding a more detailed and consistent description of BIM concepts and relationships. We describe more explicitly an example model construction methodology. Existing work has described how reasoning can be performed with BIM models by mapping BIM to existing modeling languages for which reasoning techniques have been provided [10]. Further work has described reasoning using indicator values from business data [11, 12]. In this work, we have gone beyond previous work by introducing a hybrid reasoning procedure that combines indicator and goal modeling approaches in the presence of incomplete indicators (Sect. 3.4.7). We improve and expand on analysis descriptions, using a consistent run- ning example, and providing an overall view of analysis approaches, including information requirements and how each method may be used as part of a BIM modeling and analysis methodology (Sect. 3.1).
BIM is in the same spirit as other enterprise modeling languages (e.g., Archimate [45]), which bridges the business- data gap but with a different intent, e.g., for enterprise archi- tecture. A longer term objective is to reconcile or combine these enterprise modeling approaches.
BI surveys. Work in [3] surveys approaches for infor- mation requirements analysis, approaches specifically for the development of BI-type analytical information systems.
123
Strategic business modeling: representation and reasoning 1037
Conclusions include the potential to enhance existing app- roaches by models and documentation that can be easily understood by IT, without losing precision. BIM fits these requirements by using a broad set of concepts taken from business analysis approaches.
Surveys of BI users have shown that although users are generally satisfied with BI capabilities, they are less satis- fied with the flexibility of BI systems or their ability to pro- vide risk management [1]. BIM addresses the need for more flexibility by allowing users to ask analysis questions using business terms instead of more technical queries over specific data structures. Enhanced risk management can be facilitated both by allowing users to ask “what if?” analysis questions and by explicitly capturing risk using situations.
Commercial BI. Existing software tools, such as IBM Cog- nos [2], have begun to raise the level of abstraction of data schemas used in BI, adding a layer that combines exist- ing data sources. Although these advances are useful and necessary, they still only support concepts arising from the data, failing to capture top-down strategic planning aspects included such as objectives and situations, as included in BIM. In essence, BIM complements well these commercial products.
Business modeling. The use of business-level concepts, such as business objects, rules and processes, has been researched extensively for more than a decade [46–48], and is already practiced to some extent in both data engineering and software engineering. These efforts have more recently resulted in standards, such as the business process modeling notation (BPMN) [49]. These proposals focus on modeling business objects and processes, with little attention paid to business objectives. One exception is the business motiva- tion model (BMM) [6], which proposes an extensive vocab- ulary for modeling business objectives (among other things). BMM includes several intentional concepts, such as vision, goal and objective; in our case, all these are modeled as goals. Similarly, our BIM models can capture concepts com- monly found in strategy maps [4] and Balanced Scorecards [50] (e.g., strategic goals, performance measures, initiatives). However, BIM goes beyond these concepts by including the concept of situation, a fundamental concept for supporting SWOT analysis. BIM moves beyond the capabilities of strat- egy maps, balanced scorecards, and BMM by supporting rea- soning (goal, probabilistic, and indicator) over models.
Situation modeling. Modeling of situations, especially unfa- vorable ones (e.g., weaknesses or threats), has received much attention in security engineering under the topic of vulnera- bility. For example, Elahi et al. [51] proposed a vulnerability- centric modeling ontology. More specifically, it identified the basic concepts for modeling and analyzing vulnerabili- ties, and proposed criteria to compare and evaluate security
frameworks based on vulnerabilities. Inspired by SWOT analysis, our proposal supports a more comprehensive clas- sification of situations, covering both favorable and unfavor- able ones, also internal or external to an organization.
Goal models. Modeling of goals has a long tradition within requirements engineering (e.g., i ∗ [17], URN/GRL [18] and KAOS [16, 33]). The goal concept has also been used in enter- prise modeling, for example [13, 14]. From these approaches, we have adopted intentional and social concepts. However, these models lack primitive constructs for situation, influence and indicator which are important to Business Intelligence applications.
Indicators in RE. Several proposals have used indicators as part of requirements engineering (RE). For example, van Lamsweerde [33] uses indicators to evaluate the degree of goal fulfillment. Recent proposals have extended URN to include indicators [36, 52]. We share ideas with these works; however: (1) our approach pays special attention to methods for the construction of indicator hierarchies; (2) we provide more guidelines to distinguish “what” is measured and “why” it is measured; and (3) our indicators can be used to evalu- ate situations which, from our perspective, are fundamental for strategic reasoning. In [53], the authors propose a formal framework for modelling goals (and for evaluating their sat- isfaction) based on performance indicators. Our work shares similar intentions but focuses more on the concept of com- posite indicator, ways to define metric expressions to calcu- late their values, and reasoning with incomplete indicators.
Indicators have also been used in work focusing on self- adaptive software systems (e.g., [54]). Here they serve as monitored variables that determine whether a system is doing well relative to its mandate, or whether it should adapt its behavior.
A further approach has extended goal (i ∗) models for use in designing and monitoring data warehouse systems, defin- ing awareness requirements over KPIs [55]. This approach allows for queries to be mapped to decomposed awareness requirements using OCL expressions translated to a query language (MDX). Our work goes beyond this approach by including concepts such as situation and indicator, and spec- ifying how data query results captured by indicators can be linked to analysis of the entire goal model.
7 Conclusions and future work
As a first step towards bridging the gap between the worlds of business and data in the adoption of BI technologies, we have provided business modeling support so that business data can be interpreted and understood in business terms. In this paper, we have expanded and extended previous descriptions of BIM language and reasoning. Our expanded description of
123
1038 J. Horkoff et al.
BIM has focused on key concepts for building BIM models (goal, situation, influence, indicator), intended to capture the internal and external factors that affect the strategic goals of an organization, as well as the performance measures on their fulfillment. We have expanded previous narrative to provide a sample model creation methodology, combining “top-down” and “bottom-up” approaches.
We have presented a model-based approach to design and reason about an organization’s business environment and strategies, with a focus on indicators and indicator composi- tion in the context of the BIM language.
BIM concepts have been used to facilitate several forms of reasoning over business data and objectives, supporting dif- ferent analysis questions. We have shown how existing goal and probabilistic analysis techniques facilitating the analysis and discovery of alternative strategies can be applied to BIM models. We provided techniques to analyze the impact of strategies on organization goals, by relying on different types of knowledge measured through indicators. Using expres- sions based on the structure of the BIM model and perfor- mance levels, indicator reasoning approaches can be applied even when business metrics or unit conversions are incom- plete. Also new to this paper, we have combined indicator rea- soning techniques with existing goal reasoning techniques, allowing reasoning over an entire model even if the indi- cators do not yet have associated business metrics or are incomplete with respect to the BIM model. Facilitating rea- soning with complete information makes the approach more realistically applicable, and allows reasoning over incom- plete models as part of BIM model development. We argue that the indicators and composition mechanisms proposed here are more flexible and powerful than what is commonly found in related work, allowing reasoning with varying lev- els of specific information. All reasoning approaches have been illustrated using a running example derived from real analysis reports.
We have consolidated and expanded descriptions of pro- totype implementations and use of existing reasoning tools. Work in ongoing case studies applying BIM has been summa- rized, including mechanisms to scale BIM models to realistic levels of complexity.
We plan to expand our conceptual model to incorporate further aspects of the Balanced Scorecard and Strategy Map approaches, by allowing goals to be classified amongst four organizational perspectives (financial, customers, internal business processes, and learning and growth) [4, 25]. Such perspectives can help to ensure that a balanced set of objec- tives are considered. Similarly, we intend to use the language meta-properties to classify BIM objects using BMM con- cepts such as vision, objective, mission, strategy, and tactic [6].
Along the lines of [55], future work will investigate linking indicators to business data through queries, allowing users
to query a BIM model, much like conventional database schemas, but in terms of business concepts. Such queries are to be translated through model mappings into queries defined over databases and data warehouses, and the answers are to be translated back into business-level concepts [9]. This work will involve deeper investigation of reasoning with instance level BIM models. We are extending our prototype to con- nect indicators to Business Intelligence suites, such as IBM Cognos [2] and the open source Pentaho [56].
References
1. Isik, O., Jones, M.C., Sidorova, A.: Business intelligence (BI) suc- cess and the role of BI capabilities. Intell. Syst. Account. Finance Manag. 18(4), 161–176 (2011)
2. IBM: Cognos Express Reporter. http://www-01.ibm.com/ software/analytics/cognos/express/reporter/ (2012). Accessed Apr 2012
3. Stroh, F., Winter, R., Wortmann, F.: Method support of information requirements analysis for analytical information systems. Bus. Inf. Syst. Eng. 3(1), 33–43 (2011)
4. Kaplan, R.S., Norton, D.P.: Strategy Maps: Converting Intangible Assets into Tangible Outcomes. Harvard Business School Press, Cambridge (2004)
5. Dealtry, T.R.: Dynamic SWOT Analysis. Dynamic SWOT Asso- ciates, Birmingham (1994)
6. Object Management Group: Business Motivation Model, Version 1.0. Technical Report May, Object Management Group (2002). http://www.omg.org/spec/BMM/1.0/
7. Barone, D., Mylopoulos, J., Jiang, L., Amyot, D.: Business intel- ligence model, ver. 1.0. Technical Report CSRG-607. Department of Computer Science, University of Toronto (2010). ftp://ftp.cs. toronto.edu/csrg-technical-reports/607/
8. Barone, D., Yu, E., Won, J., Jiang, L., Mylopoulos, J.: Enterprise modeling for business intelligence. In: The Practice of Enterprise Modeling, 3rd IFIP WG8.1 (PoEM’10). LNBIP, vol. 68, pp. 31–45. Springer, Berlin (2010)
9. Barone, D., Peyton, L., Rizzolo, F., Amyot, D., Mylopoulos, J.: Towards model-based support for managing organizational trans- formation. In: E-Technologies: Transforma-tion in a Connected World. Lecture Notes in Business Information Processing, vol. 78, pp. 17–31. Springer, Berlin (2011)
10. Jiang, L., Barone, D., Amyot, D., Mylopoulos, J.: Strategic mod- els for business in-telligence: reasoning about opportunities and threats. In: Proceedings of the 30th International Conference on Conceptual Modeling (ER’11). LNCS, vol. 6998, pp. 429–443. Springer, Berlin (2011)
11. Barone, D., Jiang, L., Amyot, D., Mylopoulos, J.: Composite indi- cators for business in-telligence. In: Proceedings of the 30th Inter- national Conference on Conceptual Modeling (ER’11). LNCS, vol. 6998, pp. 448–454. Springer, Berlin (2011)
12. Barone, D., Jiang, L., Amyot, D., Mylopoulos, J.: Reasoning with key performance indicators. In: Proceedings of the 4th IFIP WG8.1 Working Conference on the Practice of Enterprise Model- ing (PoEM11), pp. 82–96 (2011)
13. Stirna, J., Persson, A.: Ten years plus with EKD: reflections from using an enterprise modeling method in practice. In: Proceedings of the 11th International Workshop on Exploring Modeling Meth- ods in Systems Analysis and Design (EMMSAD07), pp. 99–108. Springer, Berlin (2007)
14. Rolland, C., Prakash, N.: Bridging the gap between organisational needs and ERP functionality. Requir. Eng. 5(3), 180–193 (2000)
123
Strategic business modeling: representation and reasoning 1039
15. Chung, L., Nixon, B.A., Yu, E., Mylopoulos, J.: Non-Functional Requirements in Soft-ware Engineering. International Series in Software Engineering, vol. 5. Kluwer Academic Publishers, Dor- drecht (2000)
16. Dardenne, A., Lamsweerde, A.V., Fickas, S.: Goal-directed requirements acquisition. Sci. Comput. Program. 20(1–2), 3–50 (1993)
17. Yu, E.: Towards modelling and reasoning support for early-phase requirements en-gineering. In: Proceedings of 3rd IEEE Interna- tional Symposium on Requirements Engineering. pp. 226–235. IEEE Comput. Soc. Press, New York (1997)
18. ITU-T, Intern. Telecom. Union: Recommendation Z.151 (11/08): Recommendation Z.151 (11/08) User Requirements Notation (URN) Language definition. Technical Report, ITU-T International Telecommunications Union, Geneva (2008)
19. Giorgini, P., Mylopoulos, J., Nicchiarelli, E., Sebastiani, R.: Rea- soning with goal models. In: Proceedings of the 21st International Conference on Conceptual Model-ing (ER’02). LNCS, vol. 3084, pp. 167–181. Springer, Berlin (2002)
20. Giorgini, P., Mylopoulos, J., Sebastiani, R.: Simple and minimum- cost satisfiability for goal models. In: Proceedings of the 16th Conference On Advanced Information Systems Engineering (CAiSE’04). LNCS, vol. 3084, pp. 20–35. Springer, Berlin (2004)
21. Wetzel, T.: States of affairs. In: The Stanford Encyclopedia of Phi- losophy (Fall 2008 Edition), Stanford University, Stanford (2008)
22. Parmenter, D.: Key Performance Indicators. Wiley, New York (2007)
23. Bresciani, P., Perini, A., Giorgini, P., Giunchiglia, F., Mylopoulos, J.: Tropos: an agent-oriented software development methodology. Auton. Agents MultiAgent Syst. 8(3), 203–236 (2004)
24. Uchil, A.: Goals-Based Strategic Planning: A No-Nonsense Prac- tical Guide to Strategy. Outskirts Press, Denver (2009)
25. Kaplan, R.S., Norton, D.P.: The balanced scorecard: measures that drive performance. Harv. Bus. Rev. 70(1), 71–79 (1992)
26. Basili, V., Rombach, H.: The TAME project: towards improvement- oriented software environments. IEEE Trans. Softw. Eng. 14(6), 758–773 (1998)
27. Letier, E., Lamsweerde, A.V.: Reasoning about partial goal satis- faction for requirements and design engineering. ACM SIGSOFT Softw. Eng. Notes 29(6), 53–62 (2004)
28. Amyot, D., Ghanavati, S., Horkoff, J., Mussbacher, G., Peyton, L., Yu, E.: Evaluating goal models within the goal-oriented require- ments language. Int. J. Intell. Syst. 25(8), 841–877 (2010)
29. Horkoff, J., Yu, E.: Interactive analysis of agent-goal models in enterprise modeling. Int. J. Inf. Syst. Model. Des. (IJISMD) 1(4), 1–23 (2010)
30. Giorgini, P., Mylopoulos, J., Nicchiarelli, E., Sebastiani, R.: Formal reasoning techniques for goal models. J. Data Semant. 1, 1–20 (2004)
31. Giorgini, P., Mylopoulos, J., Sebastiani, R.: Goal-oriented require- ments analysis and reasoning in the Tropos methodology. Eng. Appl. Artif. Int. 18(2), 159–171 (2005)
32. Elahi, G., Yu, E.: Requirements trade-offs analysis in the absence of quantitative mea-sures: a heuristic method. In: Proceedings of the 2011 ACM Symposium on Applied Computing, SAC ’11, pp. 651–658. ACM, New York (2011)
33. Lamsweerde, A.V.: Reasoning about alternative requirements options. In: Conceptual Modeling Foundations and Applications. LNCS, vol. 5600, pp. 380–397. Springer, Berlin (2009)
34. Horkoff, J., Yu, E.: Finding solutions in goal models: an interactive backward reasoning approach. In: Proceedings of the 29th Interna- tional Conference on Conceptual Modeling (ER’10). LNCS, vol. 6412, pp. 59. Springer, Berlin (2010)
35. Howard, R.A., Matheson, J.E.: Influence diagrams. Decis. Anal. 2(3), 127–143 (2005)
36. Pourshahid, A., Chen, P., Amyot, D., Forster, A.J., Ghanavati, S., Peyton, L., Weiss, M.: Business process management with the user requirements notation. Electron. Commer. Res. 9(4), 269–316 (2009)
37. Saaty, T.L.: How to make a decision: the analytic hierarchy process. Eur. J. Oper. Res. 48, 9–26 (1990)
38. University of Ottawa: jUCMNav. http://softwareengineering.ca/ jucmnav/ (2011). Accessed Sept 2011
39. Open Model Initiative: Open Model Initiative. http://www. openmodels.at/web/omi/resources (2011). Accessed Sept 2011
40. GR-Tool: GR-Tool. http://troposproject.org/tools/grtool/ (2010). Accessed Dec 2010
41. Loboda, T.D., Voortman, M.: GeNIe. http://genie.sis.pitt.edu (2012). Accessed Aug 2012
42. Eclipse Foundation: Graphiti. http://www.eclipse.org/graphiti/ (2011). Accessed Sept 2011
43. Singular Systems: Jep Java. http://www.singularsys.com/jep/ (2011). Accessed Sept 2011
44. Barone, D., Topaloglou, T., Mylopoulos, J.: Business intelligence modeling in action: a hospital case study. In: Proceedings of the 24th Conference on Advanced Information Systems Engineer- ing (CAiSE’12). LNCS, vol. 7328, pp. 502–517. Springer, Berlin (2012)
45. The Open Group: ArchiMate 2.0: An Open Group Standard. http://pubs.opengroup.org/architecture/archimate2-doc/ (2012). Accessed Aug 2012
46. Sutherland, J.: Business objects in corporate information systems. ACM Comput. Surv. 27(2), 274–276 (1995)
47. Loucopoulos, P., Katsouli, E.: Modelling business rules in an office environment. SIGOIS Bull. 13(2), 28–37 (1992)
48. Jablonski, S.: On the complementarity of workflow management and business process modeling. SIGOIS Bull. 16(1), 33–38 (1995)
49. Object Management Group: Business Process Modeling Notation (BPMN) Version 2.1. Technical Report (2009). http://www.omg. org/spec/BPMN/1.2/
50. Kaplan, R.S., Norton, D.P.: Balanced Scorecard: Translating Strat- egy into Action. Harvard Business School Press, Cambridge (1996)
51. Elahi, G., Yu, E., Zannone, N.: A vulnerability-centric require- ments engineering frame-work: analyzing security attacks, coun- termeasures, and requirements based on vulnera-bilities. Requir. Eng. 15(1), 41–62 (2010)
52. Pourshahid, A., Richards, G., Amyot, D.: Toward a goal-oriented, business intelligence decision-making framework. In: Proceeding of E-Technologies: Transformation in a Connected World. LNBIP, vol. 78, pp. 100–115. Springer, Berlin (2011)
53. Popova, V., Sharpanskykh, A.: Formal modelling of organisational goals based on per-formance indicators. Data Knowl. Eng. 70(4), 335–364 (2011)
54. Souza, V., Lapouchnian, A., Mylopoulos, J.: System identifica- tion for adaptive software systems: a requirements-engineering perspective. In: Proceedings of the 30th International Conference on Conceptual Modeling (ER’11). LNCS, vol. 6998, pp. 346–361 (2011)
55. Souza, V., Mazón, J.N., Garrigós, I., Trujillo, J., Mylopoulos, J.: Monitoring strategic goals in data warehouses with awareness requirements. In: Proceedings of the 27th ACM Symposium on Applied, Computing, pp. 1075–1082 (2012)
56. Pentaho Corporation: Pentaho: Open Source Business Intelligence. http://www.pentaho.com/ (2011). Accessed Sept 2011
123
1040 J. Horkoff et al.
Author Biographies
Jennifer Horkoff University of Toronto, Canada. Dr. Jennifer Horkoff received her Ph.D. in December of 2011 from the Uni- versity of Toronto, Department of Computer Science. Her disser- tation, under the supervision of Dr. Eric Yu, focused on interac- tive analysis for agent-goal mod- els in early Requirements Engi- neering. Her recent post-doctoral work at the University of Toronto focused on the design and analy- sis of intentional business intel- ligence models as part of the
NSERC Business Intelligence Network, a Canada-wide strategic research network with academic and industry partners. Dr. Horkoff is currently a post-doc position at the Department of Information Engineering and Computer Science at the University of Trento, Italy, under the supervision of Dr. John Mylopoulos, working as part of the Lucretius: Foundations for Software Evolution project. She has been an author or coauthor of 20 papers in peer-reviewed journals, conferences, or workshops mainly focusing on the techniques and application of goal modeling and goal modeling analysis. Dr. Horkoff has been involved in the organization of several workshops and conferences involving goal modeling and requirements engineering.
Daniele Barone University of Toronto, Canada. Daniele Barone is a post-doctoral fellow at the Department of Computer Sci- ence, University of Toronto. He received his Ph.D. in Computer Science from the University of Milano-Bicocca (Italy) in 2009. His research interests are in the areas of information and data quality, knowledge man- agement, and business model- ing. His research concentrates on methodologies and techniques to assess and improve the quality of
data in Organizations, and on concepts and techniques for modelling and analyzing the business environment, such as, goals, strategics, events, processes, resources, etc.
Lei Jiang University of Toronto, Canada. Dr. Lei Jiang received his Ph.D. in 2010 from the Uni- versity of Toronto, Department of Computer Science. His disser- tation, under the supervision of Prof. John Mylopoulos, focused on goal-oriented data modeling. His recent post-doctoral work at the University of Toronto focused on the design of a lan- guage for building and analyzing strategic business models, as part of the NSERC Business Intelli- gence Network. Dr. Jiang is cur-
rently working at Canada Pension Plan Investment Board (CPPIB) on the modelling, design and implementation for financial data analytic applications.
Eric Yu University of Toronto, Canada. Eric Yu is Professor at the Faculty of Information, Uni- versity of Toronto, Canada. His research interests are in the areas of information systems mod- eling and design, requirements engineering, knowledge man- agement, and software engineer- ing. Books he has co-authored or co-edited include: Social Model- ing for Requirements Engineer- ing (MIT Press, 2011); Concep- tual Modeling: Foundations and Applications (Springer, 2009);
and Non-Functional Requirements in Software Engineering (Springer, 2000). He is an associate editor for the Int. Journal of Information Sys- tems Modeling and Design, and serves on the editorial boards of the Int. J. of Agent Oriented Software Engineering, IET Software, and the Journal of Data Semantics. He was Program Co-chair for the 27th Int. Conference on Conceptual Modeling (ER’08). He received his Ph.D. in Computer Science from the University of Toronto.
Daniel Amyot University of Ottawa, Canada. Daniel Amyot is Professor at the University of Ottawa, which he joined in 2002 after working for Mitel Networks as a senior researcher. His research interests include scenario-based software engi- neering, requirements engineer- ing, business process modeling, aspect-oriented modeling, and healthcare informatics. Daniel is Associate Rapporteur for requirements languages at the International Telecommunication
Union, where he leads the evolution of the User Requirements Notation. He has a Ph.D. and an M.Sc. from the University of Ottawa (2001 and 1994), as well as a B.Sc. from Laval University (1992).
Alexander Borgida Rutgers University, USA. Alexander Borgida is a Professor in the Department of Computer Sci- ence, Rutgers University, New Brunswick, USA, which he joined after obtaining his Ph.D. from the University of Toronto. His general research interests center around conceptual mod- eling and its applications to improved software development. He has done specific research in knowledge representation (esp. description logics), data and
knowledge bases (e.g., exceptions) software specification (esp. require- ments). He has (co)authored over 100 scientific papers, frequently serves on conference program committees and currently appears on the editorial board of 4 journals.
123
Strategic business modeling: representation and reasoning 1041
John Mylopoulos University of Trento, Italy. John Mylopoulos holds a distinguished professor position (chiara fama) at the Uni- versity of Trento, and a pro- fessor emeritus position at the University of Toronto. He earned a Ph.D. degree from Princeton University in 1970 and joined the Department of Computer Sci- ence at the University of Toronto that year. His research interests include conceptual modelling, requirements engineering, data semantics and knowledge man-
agement. Mylopoulos is a fellow of the Association for the Advance- ment of Artificial Intelligence (AAAI) and the Royal Society of Canada (Academy of Sciences). He has served as programme/general chair of international conferences in Artificial Intelligence, Databases and Soft- ware Engineering, including IJCAI (1991), Requirements Engineer- ing (1997), and VLDB (2004). Mylopoulos was recently awarded an advanced grant from the European Research Council for a project titled “Lucretius: Foundations for Software Evolution”.
123
Copyright of Software & Systems Modeling is the property of Springer Science & Business Media B.V. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
- Strategic business modeling: representation and reasoning
- Abstract
- 1 Introduction
- 2 Strategic business models
- 2.1 Running example: BestTech
- 2.2 BIM concepts and relationships
- 2.2.1 Goal
- 2.2.2 Situation
- 2.2.3 Influence
- 2.2.4 Indicator
- 2.3 Construction of BIM models: sample methodology
- 3 Reasoning with BIM models
- 3.1 Overview of reasoning approaches
- 3.2 Reasoning with BIM models using goal modeling techniques
- 3.2.1 Evaluation of specific strategies
- 3.2.2 Discovering alternative strategies
- 3.3 Probabilistic evaluation of strategies
- 3.4 Reasoning with indicators
- 3.4.1 Indicator performance levels
- 3.4.2 Composite indicators
- 3.4.3 Indicator reasoning with varying levels of information
- 3.4.4 Indicator reasoning using business formulae and unit conversion
- 3.4.5 Indicator reasoning using business formulae and performance levels
- 3.4.6 Indicator reasoning without business formulae
- 3.4.7 Hybrid reasoning: reasoning with incomplete indicators
- 4 Tool support
- 5 Ongoing case studies
- 6 Related work
- 7 Conclusions and future work
- References