- Email: [email protected]
COMEX: A cost management expert system
Expert Systems with Applications 37 (2010) 7684–7695
Contents lists available at ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
COMEX: A cost management expert system Danijela Grahovac *, Vladan Devedzic 1 Department of Software Engineering, FON – School of Business Administration, University of Belgrade, P.O. Box 52, Jove Ilica 154, 11000 Belgrade, Serbia
a r t i c l e
i n f o
Keywords: Expert system Cost management system ABC method
a b s t r a c t Understanding and using cost management is a key to success in any type of organization, because it helps managers to collect important data about the costs assigned to products, services and customers and use it for planning and control function. This paper discusses development and application of an expert system for reasonable cost related decision making and continuous cost management process improvement. The system is called COMEX (cost management expert system). This paper also shows how COMEX can be used for improvement of the billing department business processes. COMEX relies on information generated using the activity-based costing (ABC) method, accounting method based on activities, created to improve cost management system by focusing on the driver costs of individual activities. Still, this paper focuses on design and implementation of the COMEX system, and only briefly surveys the ABC method. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction Beware of little expenses. A small leak will sink a great ship. Benjamin Franklin A cost management system (CMS) consists of a set of formal methods developed for planning and controlling an organization’s cost-generating activities related to its short-term objectives and long-term strategies. Business entities face two major challenges: achieving profitability in the short run, and maintaining a competitive position in the long run. An effective CMS must provide the managers with the information needed to meet both of these challenges. Cost management is not practiced in isolation. It is an integral part of general management strategies and their implementation. Examples include programs that enhance customer satisfaction and quality assurance, as well as programs that promote new product development. Cost management has a broad focus. It covers – but is not limited to – the entire process of continuous control of costs. Planning and control of costs is usually inextricably linked with revenue and profit planning. For instance, to enhance revenues and profits, managers often deliberately incur additional costs for advertising and product modifications. * Corresponding author. Present address: 50 Jerome Cres. #503, Hamilton, Ontario, Canada L8E1K6. Tel.: +1 905 869 4020; fax: +1 905 561 5320. E-mail addresses: [email protected] (D. Grahovac), [email protected] (V. Devedzic). URLs: http://goodoldai.org/danijela_grahovac (D. Grahovac), http://devedzic.fon.rs/ (V. Devedzic). 1 Tel.: +381 11 3950853; fax: +381 11 2461221. 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.04.073
The information generated from a CMS should help managers measure and evaluate performance. The measurements may be used to evaluate human or equipment performance or to evaluate future investment opportunities. However, although information from accounting systems helps managers to make decisions, the information and the accounting systems themselves are not cost management. The decision process, as all other business processes, includes a lot of problem solving actions/activities, experience, and heuristics. An expert consultancy may be necessary when a particular organization has to make a decision or solve some problem. These experts possess a specific knowledge and an experience in certain areas. When the company has to decide on the matter of something like a big investment, integration or advertising strategy, as a rule they will hire an expert for advice. Expert systems are able to support a decision-making process, and sometimes they can completely replace an expert. We have developed COMEX, a cost management expert system that supports the work of billing departments of utility companies. The purpose of COMEX is to satisfy the customer’s needs and stay concurrent with lower costs. For example, customer inquiries can consume a large percentage of a company’s resources, and reducing costs for this activity brings savings and customer’s satisfaction. This is applicable to all other activities in the company. COMEX recommends how to achieve these savings. COMEX can be applied to any kind of billing (e.g., electricity, water, or waste). The rest of this paper is organized as follows. Section 2 states the problem covered in this paper precisely. Section 3 briefly surveys relevant facts and methods from the domain of cost management, as well as JavaDON, the ES shell we used to develop COMEX.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Section 4 is the central section in this paper. It discusses the architecture and design of COMEX in detail. Section 5 shows a system evaluation, whereas Section 6 discusses other relevant aspects of COMEX. 2. Problem statement The purpose of this paper is to discuss how the expert system technology can be successfully applied in the domain of cost management. To this end, this paper covers the COMEX project in detail. The COMEX project objectives and the requirements include the following: the system must be well engineered from the knowledge engineering perspective; the system must be user friendly and easily understandable for the end user; the system should display questions in a logical, natural sequence, in a similar way that a human expert is thinking when solving related problems; the system must be properly designed to support the work of a billing department of any utility company; the system should suggest logical and useful solutions; the system should be open for further development and extensions. The completed COMEX system is now being constantly tested, evaluated, and explored for possibilities to be successfully applied on all organizations with a billing department (BD). The project revealed as useful and consultative, and by those facts, COMEX succeeded as the cost management system. 3. Related work 3.1. Cost management systems There have been significant criticisms regarding the lack of efficiency and capability of traditional cost management and accounting practices in the literature during the last two decades leading to a call for new techniques (Askarany & Smith, 2000). Designed for application under different business circumstances and for the other epoch, traditional cost management appeared to be a system with some major weaknesses that demonstrate a lack of overall competency. Traditional cost management systems are inaccurate in the way that they allocate costs, and managers armed with this system were not able to reduce costs, identify opportunities for improvement, and determine a more profitable product mix (Cooper & Kaplan, 1991). The lack of informational support was an obstacle for managers. Cost accounting had traditionally allocated overhead to products or services using only one volume-sensitive driver, typically direct labor. For organizations with high overhead and a mix of products or services, using a single cost driver may distort cost estimates (Cooper & Kaplan, 1991). Traditional methods of allocating overhead were therefore believed to be deficient in terms of improving global competitiveness (Johnson, 1990). Expert systems are useless with data gathered from traditional cost management systems. Information provided by these systems is inaccurate, and will result with inadequate solutions and advice. Hence, companies have to switch from a traditional costing system to a new method for cost management, and then interpret the data by the expert system. Modern cost management systems are adequate for the dynamic environment in which business operates. A setting for this approach requires including different types of experts (e.g., indus-
7685
trial, accounting or management) who are going to work on the preparation of a plan and control programs, as well as on resources, processes, cost, performance and risk management. However, there is a large number of different cost management methods and techniques (see http://www.valuebasedmanagement.net/). Some of the most popular ones are briefly covered in the Appendix. All such methods have certain advantages and disadvantages. In developing COMEX, we have used the ABC method for cost management (Kaplan & Cooper, 1998) as an underlying domain knowledge and methodology. The reason we opted for ABC is the fact that this method links company resources (inputs) and products/services in an innovative way, which is a precondition for any business process improvement (according to the ISO 9001 standards series requirements (ISO 9000, 2009)). Business process improvement in a company/enterprise/organization, in turn, brings improvement to the organization’s overall business improvement. 3.2. ABC method ABC is a powerful tool for measuring a company’s performance (Kaplan & Cooper, 1998). It recognizes cause–effect relationships between assigned objects costs. Once activity costs are identified, ABC assigns costs to products/services based on the resources they consume. ABC is a new method of cost assignment, more compatible with the increased high technology environment in which business operates. ABC assigns costs to products on the basis of the types and quantities of activities that must be performed to create those products. This costing system accumulates costs for activity centers in multiple cost groups at a variety of levels (unit, batch, product, and organizational) and then allocates these costs using multiple cost drivers (Barfield, Raiborn, & Kinney, 1998). Thus, the costs are assigned more accurately, and managers can focus on controlling activities that cause costs rather than trying to control the costs that result from the activities. The use of activitybased costing should provide a more realistic picture of actual production costs than has traditionally been available. Managers have more success in understanding how organization is using its own capital by assigning costs for defined activities using the ABC method. As a more accurate CMS than traditional cost accounting, ABC identifies opportunities to improve business process effectiveness and efficiency by determining the ‘‘true” cost of a product or a service (APCG, 2002). It is implemented in many companies around the world. The way of applying ABC systems differs from one organization to another. Some organizations use ABC system as the main CMS, but many of the ABC issues are selective and can be applied on subunits or separate functions of organizations. Note, however, that although ABC typically provides better cost estimates than the traditional overhead allocation process, it is not a panacea for all managerial concerns. This is the point where an expert system can take a part (Grahovac & Devedzic, 2009). 3.3. JavaDON expert system shell COMEX is implemented using the JavaDON ES shell (Tomic, Jovanovic, & Devedzic, 2006). It is an academic open source shell, currently used as a developing and teaching tool; its current version is a shareware (http://iis.fon.rs/additional%20materials.html). JavaDON is well suited for building complex expert system applications based on techniques such as rules and frames. The shell is based on the OBOA framework, whose partial conceptual model is shown in Fig. 1, illustrated in the UML notation. OBOA is a systematic approach to design and development of software for intel-
7686
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
ligent systems (Devedzic & Radovic, 1999). It is based on a multilevel, general object-oriented model of intelligent systems, and has been developed starting from some observations regarding difficulties in making extensions to methods and tools for building intelligent systems – scalability efforts – as well as in reusability and embedding-oriented efforts (Nedovic & Devedzic, 2002). Some of the most important elements of the OBOA framework are (Tomic et al., 2006): Knowledge element – all of the primitives and the units (no matter the complexity) are considered to be knowledge elements, since they all share common properties. Knowledge chunk – a basic logical statement. It can be used, e.g., to relate a frame slot with a value or with another slot. It is typically used as a building block for rules. Domain – determines the basic type and range of a value. The basic types are: string, integer, floating point and boolean.
4. COMEX architecture and design COMEX interprets the data resulting from applying the ABC method. The ABC method studies a company’s activities to classify them into activity centers, than traces costs to those activities and determines the cost drivers. For example, the maintenance activity has the number of machine hours as its a cost driver, and the correspondence activity has the number of letters as the a cost driver. The map of the activities supplied by the ABC method is essential for a knowledge engineer during the design phase of the COMEX development. 4.1. Knowledge representation COMEX is essentially a frame-and-rule-based system. There are nine frames in its knowledge base:
The COMEX knowledge base (system level) and knowledge elements (integration level) are domain-depended (e.g., cost management), and all other elements are domain-independent.
Customer Activity BillInquiry WebSite
SYSTEM
INTEGRATION
COMEX KNOWLEDGE BASE
BLOCKS
0..* KNOWLEDGE ELEMENTS
0..* KNOWLEDGE ELEMENT
UNITS
RULE IF
FRAME THEN
1 0..*
0..*
1..* 0..*
0..*
CLAUSE
SLOT 0..* 1 0..*
0..*
KNOWLEDGE CHUNK 0..* 1
0..*
0..1
1
ACTION
PRIMITIVES
1
0..*
1..*
ATRIBUTE
0..*
1
DOMAIN
0..*
MEDIA
0..* RELATION
0..1
VALUE
Fig. 1. COMEX in the context of OBOA framework/architecture (adapted according to Tomic et al. (2006)).
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
AccountVerification CustomerInquiry Correspondence Expert ExpertSolution
The frames are, of course, inter-related. The UML class diagram in Fig. 2 explains the relations among the frames.
7687
As we mentioned before, COMEX is created to support the work of the billing department (BD). BD has two types of customers – commercial and residential. The activities that have to be done in order to generate the bill for these types of customers are different. All activities are grouped into four possible categories (see Fig. 2). In order to improve the business process, the user can select the customer type and then the activity type. COMEX will then lead the user though an appropriate dialog (a series of questions). For
Fig. 2. Class diagram relating the major concepts of the COMEX system.
Fig. 3. COMEX frames (screenshot from JavaDON).
7688
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
example, if the BillInquiry activity type is selected, the dialog will be related to the data from the specific type of bill (e.g., in the case of utility billing, the number of printed lines or information redundancy). After a certain number of the system’s questions, a solution may be reached; it is the job of the Expert frame, which tightly collaborates with the activity frames. Note that a solution could be offered after each step in the dialog. In case of the BillInquiry activity, if no solution is reached during the dialog defined by the questions related to the BillInquiry frame, the WebSite frame will take over and will come up with a new set of questions. These are related to the option of billing the customer through a WebSite, or sending bills as email (e.g., ‘‘Do you have your own website?”, ‘‘Do you have printed notification about website on each bill?”). They will be asked until COMEX comes up with a solution (such as ‘‘Offer the customer to switch to online billing, with the bonus of **$”). The WebSite frame refers to the BillInquiry activity, and it is separated because it offers a new approach to BillInquiry process improvement. 4.2. Frames Fig. 3 shows the nine frames of COMEX as they are implemented in JavaDON. In JavaDON, each frame can have zero or
more subframes and parent frames. For example, there are two slots in the Customer frame but no subframes and parent frames, Fig. 3 (SubFrames and ParentFrames of the Customer frame have no children nodes). Although it appears in Fig. 3 that there is no inheritance among the COMEX frames, it is not so (see Fig. 2). It just appears to be so due to the fact that JavaDON shows all frames as branching directly from the root of the knowledge base. JavaDON frames have slots that represent the corresponding frame’s attributes. Each slot contains the definition of the attribute it represents, a question to be presented to the end user in dialogs with COMEX whenever the system asks the user to supply a value for the attribute, possible answers, and a description (canned text used for providing explanations in the user’s sessions with COMEX). A slot’s attribute is shown in its node in the expanded tree. For example, the typeOfCustomer slot from the Customer frame, Fig. 4, is defined to ask the user ‘‘What is the type of the customer?”, and the user is supposed to select between two possible types of customers: Commercial and Residential. 4.3. Rules The structure of COMEX rules is shown in Figs. 5 and 6. For example, the BillCostComGreat rule in plain English says: If the customer is a commercial customer, and the customer’s bill cost is greater than or equal to 8.5 Then ask the user about the type of activity The same rule in COMEX includes the following knowledge chunks: a) Customer:typeOfCustomer=(Commercial;1.0) b) Customer:billCost>=(8.5;1.0) c) ask_question(Activity:typeOfActivity)
Fig. 4. The typeOfCustomer slot.
1
RULE
1
0..*
0..* THEN CLAUSE
IF CLAUSE
0..*
0..*
1
KNOWLEDGE CHUNK
1
Fig. 5. Rules and chunks relationship.
Fig. 6. The structure of rules in JavaDON/COMEX (a screenshot from JavaDON).
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
The rule (i.e., its knowledge chunks) states that if the slot typeOfCustomer from the Customers frame has the value Commercial and the billCost slot from the same frame has a value greater than or equal to 8.5 (IF clause), then the system will ask the question (the ask_question action) related to the typeOfActivity slot of the Activity frame. This means that if the selected customer class is Commercial and the bill is greater than a certain limit, COMEX will proceed with the process improvement. The value 1.0 in the knowledge chunks denotes the corresponding certainty factor. In addition to IF and THEN parts, each rule has its TYPE (which can be either AND or OR) and IMPORTANCE (the rule’s priority, used to resolve conflicts when multiple rules can fire). IMPORTANCE is a rule feature provided by JavaDON and is used to control the order of rule firing. In each inference cycle, the rule with the highest importance is selected to fire if more than one rule is satisfied. By a more detailed analysis of some of the COMEX rules, one can get a closer insight into the way COMEX solves problems. Rule 1 initiates each dialog with the user. It has the highest importance (10) and always fires first. Its If clause is empty, so it will always fire unconditionally at the beginning of the session. Rule 1: INIT (Importance=10) If Then ask_question(Customer:typeOfCustomer)
The first question that COMEX will ask the user during his/her session with the system is ‘‘What is the type of the customer?”, and the user is supposed to select between two possible types of customers: Commercial and Residential (see the Then clause of Rule 1). This is a very important partitioning, since the billing process is different for the two types of customers. The Then clause executes the action ask_question(Customer:typeOfCustomer) related to the typeOfCustomer slot of the Customer frame, and the user is asked to enter the appropriate answer. Rule 2 has Importance=9, and always fires second, since all other rules do not have such a high importance. This rule’s If clause is also empty, so Then clause will always fire unconditionally. The Then clause refers to the ask_question(Customer:billCost) chunk that executes the action related to the billCost slot of the Customer frame with the question ‘‘What is your cost per account?”. The user enters the cost that the customer has made by a certain bill. Rule 2: BillCostCommRes (Importance=9) If Then ask_question(Customer:billCost)
7689
Rule 2.1: BillCostCommLess (Importance=0) If Customer:typeOfCustomer=(Commercial;1.0) AND Customer:billCost<(8.5;1.0) Then Expert:finalSolution=(‘‘You are under market average limit!”) ExpertSolutions:solution=(‘‘You don’t have to do process improvement.”) Rule 2.2: BillCostCommGreat (Importance=0) If Customer:typeOfCustomer=(Commercial;1.0) AND Customer:billCost>=(8.5;1.0) Then ask_question(Activity: typeOfActivity) Rule 3 has Importance= 10, which means it always fires last. Its If clause is empty, so it will always fire unconditionally. The Then clause contains the chunks show_value(Expert:finalSolution) and show_value(ExpertSolutions:solution), which execute the show value action. In this case, these are the values of the finalSolution and solution attributes, where the COMEX solutions and advice are stored. Rule 3: ShowSolution (Importance= 10) If Then show_value(Expert:finalSolution) show_value(ExpertSolutions:solution) If the selected customer class is Residential, the inference process proceeds in a similar way, this time working with Rule 2.3 and Rule 2.4. Rule 2.3: BillCostResLess (Importance=0) If Customer:typeOfCustomer=(Residential;1.0) AND Customer:billCost<(3.5;1.0) Then Expert:finalSolution=(‘‘You are under market average limit!”) ExpertSolutions:solution=(‘‘You don’t have to do process improvement.”) Rule 2.4: BillCostResGreat (Importance=0) If Customer:typeOfCustomer=(Residential;1.0) AND Customer:billCost>=(3.5;1.0) Then ask_question(Activity:typeOfActivity) Fig. 7 shows the sequence diagram of the COMEX inference process from the beginning to the point where the type of activity is chosen.
If the selected customer class is Commercial, then Rule 2.1 or Rule 2.2 is fired. Both of these rules have two If clauses. The first one checks the type of the customer, and the second one checks whether or not the entered value is greater than, less than, or equal to the market average value (8.5). If the current value is less, than Rule 2.1 is fired. This results in ending a cycle, and a rule presents the solution, stating that the organization is under the market average limit and there is no real need for process improvement. In order to show this solution, the system fires Rule 3 (see description below). If the value of billCost is greater than or equal to the market average value, the Then clause will fire the chunk ask_question(Activity:typeOfActivity). The chunk executes the action of asking a question. The question located in the typeOfActivity slot is ‘‘Which activity do you want to improve?”. When this rule is fired, the user is supposed to select one of four possible types of activities for further improvement.
4.4. Dialogs Another very important piece of knowledge in the COMEX knowledge base is the typeOfActivity slot in the Activity frame. The related question that COMEX will ask the user during his/her session with the system is ‘‘Which activity do you want to improve?” There are four possible answers: BillInquiry, CustomerInquiry, AccountVerification and Correspondence (Fig. 8). These activities origin from the ABC method applied to BD. After the user selects an activity, COMEX will typically ask a question about the average percentage of the total cost allocated to the chosen activity. The average slot in the Activity frame stores the value of the total cost allocated in the activity of the billing department (BD). Depending on the activity selected, the value entered is compared to the actual market average percentage. For
7690
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 7. Start COMEX – sequence diagram.
example, the market average percentage for the AccountVerification activity is 16%. Anything under this value does not need improvement; in case of a greater value, COMEX goes forward into improvement and asks question Q11: Question: Q11 What percent of the total cost is allocated for this activity? The related rules in the knowledge base are: Rule 17: IF A=‘‘Commercial” AND 16
Rule 18: IF A=‘‘Commercial” AND 16>=average THEN ask-question Q11
Fig. 9 shows the sequence diagram for one of the four branches, i.e., CustomerInquiry improvement. There are five different endings that can happen, depending on the user’s answers. 4.5. User modeling COMEX users are knowledge engineers and end users. A knowledge engineer is the one who creates the knowledge base and tests the system. Speaking in terms of UML, the use cases related to knowledge engineers are illustrated in Fig. 10. They include building the knowledge base, creating a new project, opening and updating the knowledge base, and the like. During the design of the COMEX system, the knowledge engineer communicated with a cost management expert, an ABC expert, and accounting and system quality experts. All the important knowledge was entered in the knowledge base through multiple iterations (Grahovac & Devedzic, 2009). COMEX end users are ABC experts, accountants, managers, business analysts, and students. End users do not have the permission to update knowledge base elements. They can open a project (open project COMEX), enter the session data (enter data) and start the forward-chaining inferencing (forward chaining) (Fig. 11) (Tomic´, 2005). 4.6. Reasoning and conflict resolution
Fig. 8. Selecting an activity to improve.
COMEX reasons using forward chaining with its rules. The inference mechanism of COMEX is presented by the example shown in Fig. 12 and Table 1. The example presents one COMEX cycle, and it shows how automatic inferencing starting from the knowledge base and the working memory data brings up new solutions. The figure shows that variables A and B are stored in the working memory, which means that the user has answered questions about the type of customer (Residential) and entered a cost per account greater than the current market average (3.5). Since COMEX is a forward-chaining system, in each cycle of its reasoning process a set of candidate rules to fire is created (the
7691
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 9. CustomerInquiry activity – sequence diagram.
Create Project COMEX <>
<> <>
<> <>
Knowledge Engineer
Build Knowledge Base
Edit COMEX
Save COMEX
<><> <>
Close COMEX
<>
Open Project COMEX
Fig. 10. Knowledge engineer’s use cases with COMEX.
rules that have their IF clauses satisfied). Conflict resolution and selection of the rule to fire is done by selecting the highest priority rule. Each rule can be fired only once. 4.7. A session with COMEX The reasoning process is initiated by having three rules automatically satisfied when the session starts: INIT (Fig. 13), BillCostCommRes (Fig. 15) and ShowSolution (Fig. 18). The INIT rule’s importance is set to 10 and thus it will be fired first. As a result, the user is presented the question about the customer type as in Fig. 14.
Upon selecting the customer type and clicking Next, the next cycle of forward chaining is executed and the BillCostCommRes rule (with the IMPORTANCE set to 9.0, Fig. 15) is fired. Its THEN clause executes the action ask_question(Customer:billCost) and the user is asked to enter the appropriate bill cost, Fig. 16. In this example, it is assumed that the user is a Commercial customer and that he has entered the value 6.5 that corresponds to this customer type. This value, as same as all others, is a result of implementing the ABC method in the BD. This results in COMEX activating the BillCostComLess rule (Fig. 17), because the bill cost entered is less than the market average. The rule’s THEN clauses are added to the working memory and
7692
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Open Project COMEX <>
Inference
User
<> <>
Forward Chaining
Enter Data
Fig. 11. End user use cases.
Fig. 12. COMEX – forward chaining.
Table 1 Facts. Symbol
Variable
Value
Rule
A B C D E F
typeOfCustomer billCost typeOfActivity Average Letter finalSolution and solution
Residential >3.5 Correspondence >6.0 True ‘‘Conclusion”
Rule Rule Rule Rule Rule Rule
1 and Rule 2 2.4 4.4 4.4.1 4.4.1.1 3
Fig. 14. The start of a session.
Fig. 13. The INIT rule.
the slots Expert:finalSolution and ExpertSolution:solution are assigned the appropriate values. The ShowSolution rule (Fig. 18) will use these values when presenting the solution that COMEX displays to the user at the end of the session, Fig. 19. Note that ShowSolution has an empty IF clause, so it will certainly fire. However, it has the lowest IMPORTANCE ( 10), hence it will fire only at the end of the session.
Fig. 15. The BillCostCommRes rule.
Of course, this simple example only demonstrates how COMEX works. In real-world examples, its reasoning takes many more steps and can get quite complex.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 16. The result of executing the ask_question(Customer:billCost) action.
Fig. 17. The BillCostComLess rule.
Fig. 18. The ShowSolution rule.
7693
to business intelligence (BI) applications, as well as to some expert systems in the more general domain of business management. The first comparison is made between COMEX and BI applications. BI applications provide an integrated ABC method based modules for cost management. The calculated data are the real picture of an organization’s costs, but the important question therefore is exactly what to do with obtained data and what kind of concrete steps to take in order to improve the business process and make some savings. COMEX is able to reasons over these data to provide practical advice and recommendations, if the calculated costs are for the billing department. Further development of COMEX will include a completely automated input of the calculated costs. BusinessObjects ERP profitability and cost management application (Weismantel et al., 2009) gives a picture of real costs and profits by using the ABC method, and makes possible to ‘‘dig” a source of the costs. At the end, attractive and functional reports require an expert for advice. The analytic module compares a company’s strategic goals with current data and results show where the company is now, and where it is supposed to be. How? It requires a human expert again. SYSPRO Enterprise Software,2 IBM Cognos 8,3 Microsoft Dynamics4 and NetSuite,5 operates on similar principles as BusinessObjects, with more or less analytics, data mining, forecast and optimization included. However, after all these creative reports conclusion is essentially the same, it needs someone to suggest what to do. That is exactly the right place for deploying an ES, and in the case of utility companies, the place for COMEX. The second comparison is made between COMEX and some expert systems in the more general domain of business management (Table 2). Regarding reporting, Vanguard automatically identifies the most influential company’s financial risk and shows results as graph presentation, similar as Agena. Agena has an option to present data using tables and risk maps. Cost Advantage shows comparative tables of estimated and actual costs. Thus, the user is able to come up with desirable costs, by simulating different models of product design. Contrary to that, the results of COMEX are advice and solutions, but it does not validate results with some$”. This additional functionality is not thing like ‘‘saving is fully developed yet because the Formula function in the JavaDON shell is not implemented yet. Regarding the usage of this kind of systems, currently on the highest place is Vanguard, followed by Agena. These commercial software packages are more general and target various types of industries, whereas Cost Advantage and COMEX target more specific ones. Cost Advantage is developed to support the work of companies that design products, and COMEX to support the work of utility companies. Tools and applications are quite different for all systems. Furthermore, the described systems are completely incompatible and simply it is not possible to connect them. This is a typical case of the software diversification, and it does not merge with the globalization concept that is initiated in a previous decade. Again, regarding knowledge representation and collection, there is an entire set of different techniques. The Vanguard is based on the production rules (just like COMEX), and decision trees. In
Fig. 19. The solution presented by COMEX.
5. Evaluation We are not aware of another expert system in the area of cost management, especially not of any such a system that builds upon the ABC method. Hence we have evaluated COMEX by comparing it
2 URL: http://www.sysprosolutions.com/files/fact_sheets_financial/Activity_Based_Costing.pdf, last accessed: April, 2009. 3 URL: http://www.cognos.com/products/cognos8businessintelligence/index.html, last accessed: April, 2009. 4 URL: http://www.microsoft.com/dynamics/using/default.mspx, last accessed: April, 2009. 5 URL: http://www.netsuite.com/portal/products/netsuite/main.shtml, last accessed: April, 2009.
7694
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Table 2 Summary of the expert systems. System
Vanguard, risk management module
Agena
Cost advantage
COMEX
Domain
Risk analysis
Risk analysis
Cost management
Output (reports)
List of identified risks
Tables, graphs and risk maps
Production cost management Costs table
Usage
More than 60 companies around the worlda Vanguard Studio, DScript Decision trees and production rules
RBC Bank, Philips, Simens, Motorola, Bocsh, etc. Java API Risk maps, Bayesian networks
Hierarchical decision models Currently used Vanguard (2008)
Probability-based decision making, Bayesian inference engine Currently used Agena (2005)
Tools/applications Knowledge representation and collection Reasoning Status Reference a
Allegro CL’s common Lisp Production rules written in natural language
Advice and solution for BD’s processes improvement Power-utility of Republic of Srpska JavaDON Production rules, frames, O–A–V triplets
N/A
Forward chaining
Currently used Cognition (2008)
Testing and limited use Grahovac and Devedzic (2009)
Motorola, Cummins,. . .
URL: http://www.vanguardsw.com/customers, last accessed: April, 2009.
addition, COMEX represents knowledge with frames and O–A–V triplets. Production rules are the basis for Cost Advantages’ knowledge representation as well, but in Cost Advantage they are written in natural language. Agena has an embedded Bayesian Networks technique as well as risk maps. What is better? An answer for this particular question can be only subjective. The reasoning or inferencing is different from one application to another. Forward chaining is a very popular technique, and is a part of COMEX. All of the reasoning in JavaDON is done by using the forward chaining inference technique (Tomic et al., 2006), thus it is a part of COMEX, too. The JavaDON team is currently working on the full implementation of the backward chaining technique, which apparently is not going to affect the current COMEX configuration. Vanguard’s decision trees are necessarily associated with a hierarchical decision model, and likewise Agena is with Bayesian inference engine. Vanguard’s decision about this kind of reasoning is, likely, because of the system complexity, since risk management is only one of the modules. Agena is oriented only on risk management associated with probability-based inference that Bayesian technique is based on.
During the development of COMEX we were asked the following questions: Why an expert system? Is this technology ‘‘in” or it was dismissed in 1990s? The answers to these questions reflect the authors’ inspiration to research the subject. If we called COMEX a business rule management system (BRMS), it would certainly be ‘‘in”. And what is a BRMS? The little secret is that a BRMS was originally called ‘‘expert system” and its rule engine was called ‘‘inference engine”. Then it was called ‘‘business rule engine”, and today this kind of system is known as ‘‘business rule management system”. Hence COMEX may change its name/categorization over time. But not yet. Expert systems deserve all due respect. Expert systems didn’t really go away. They went undercover (Rolo, 2007). This statement can be justified by, e.g., AMEX’s Authorizer’s Assistant (ES for credit cards transactions management) (BizRules, 2008). This magnificent ES is working perfectly since 1988. COMEX is developed using JavaDON’s current capabilities. Simultaneously with JavaDON development, COMEX will develop, too. Currently, the team is working on developing JavaDON’s interoperability with an external database; in case of COMEX, a relationship with Excel tables would be very helpful.
7. Conclusions 6. Discussion Despite the fact that COMEX supports the work of billing departments of utility companies, it can be applied to any kind of billing (e.g., electricity, water, or waste). The central idea of the COMEX project was to illustrate how business effectiveness could be significantly improved by applying a modern method for cost management and expert system technology to interpret the data. COMEX demonstrates that synergy between ES and modern economic tools (i.e., ABC) results in remarkable improvement of cost management. Besides the assumption about using the ABC method, COMEX is based on activities, too. One of the most essential branches is BD’s activity selection. The ABC method provides these activities. COMEX solutions are in fact recommendations for improvement of BD’s processes. Naturally, cost savings come along with process improvement. Cost management is generally very popular, mainly because it has a direct impact on profitability. Thus, COMEX can find its place in practical application, both as manager’s decisionsupport aid, and as a training tool for employees. As an application, COMEX is easy and friendly to use. It is made for users without any knowledge about expert systems, but in need for sound advice about BD’s process improvement and costs saving.
COMEX is a cost management expert system that supports the work of billing departments of utility companies. It relies on using the ABC method for cost management. ABC can calculate cost data for specific activities in utility companies, and COMEX interprets the data. COMEX demonstrates how business effectiveness can be improved by applying a new method for cost management and expert system technology to interpret the data resulting from applying the method. In other words, the ABC method calculates the real costs and COMEX reasons over the data to provide useful advice and recommendations. Together, they bring significant improvement to cost management. Although specifically developed for a power utility company (Power Utility of the Republic of Srpska), COMEX has an option to be extended to other organizations that incorporate a billing department and print out a large amount of bills. Examples of such other companies include telephone companies, TV companies, Internet providers and the like. Our current and future work is oriented towards putting the COMEX demo online and commercialization of the developed full version. As expert systems developed with JavaDON can be either desktop or Web applications, one more COMEX improvement could be an application for the Web.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Appendix A. Popular cost management methods Popular methods used today in various cost management systems include: Target Costing, Kaizen Costing and Life Cycle Costing. A.1. Target Costing Target Costing (TC) – (Kucˇinar, 2005) – originated in Japan, as a result of requirements for creating a market-oriented concept under global economic terms, growing competition, new technologies expansion, innovations and reduced product life cycle. From the objectives and methodology point of view, TC is differing from traditional systems based on ‘‘cost-plus” accounting. TC applies to the design phase with a specific focus on research, development and product design activities. This approach seeks to listen, understand and anticipate consumers’ needs and aspirations, and emphasizes their willingness to purchase and pay for the quality and proper functionality of the product. It presumes integration of the overall business and management knowledge, and within the design phase, routes realization of the strategic goals, as cost reduction, performance and profit increasing. A disadvantage of Target Costing is the danger of ‘‘killing” potentially good ideas early, since getting the real costs is difficult. It is almost impossible to develop an ‘‘IPhone” or a new ‘‘Pampers” under the pure ‘‘Target Costing” thinking (Sudmann, 2006). A.2. Kaizen Costing Kaizen Costing (KC) is a costing technique that reflects continuous efforts to reduce product costs, improve product quality, and/ or improve the production process after manufacturing (Barfield et al., 1998). Yashihuro Monden from Japan, who developed Kaizen Costing, defines it as the maintenance of present cost levels for products currently being manufactured via systematic efforts to achieve desired cost level (IFS, 2001). Contrary to TC, which is applied to a particular product (new or an existing one) that is currently under (re)design, KC is applied to a product that is already in production. The primary goal of this method is to permanently reduce specific product costs, using improvements in small steps. One of the related characteristics for the continuous process improvement is that cost reduction authority and responsibility delegates from high managerial level to operative levels. A disadvantage of the KC technique is that the seek for cost savings results in a reduced quality of the product. A.3. Life Cycle Costing Life Cycle Costing (LCC) is the accumulation of costs for activities that occur over the entire life cycle of a product from inception to abandonment by the manufacturer and consumer (Barfield et al., 1998). LCC is a successful answer to the criticism to other
7695
methods, especially traditional ones, that states that cost management and products performance ‘‘start too late and end too soon”. However, the LCC system has a small value by itself. It is most useful when it can be compared to the LCC of other design alternatives, which can perform the same function, in order to determine which alternative is the most cost effective one for this purpose (Fuller & Peterson, 1995). References Agena Limited (2005). Getting started with AgenaRisk. (last visited April, 2009). APCG (2002). Activity-based costing (white paper). (last visited April, 2009). Askarany, D., & Smith, M. (2000). A critical evaluation of the diffusion of cost and management accounting innovations. Management of Innovation and Technology, 1, 59–64. Barfield, J. T., Raiborn, C. A., & Kinney, M. R. (1998). Student solutions manual: Cost accounting: Traditions and innovations (3rd ed.). Cincinnati, Ohio, USA: SouthWestern Pub.. BizRules (2008). American express authorizer’s assistant (news bulletin). (last visited April, 2009). Cooper, R., & Kaplan, R. S. (1991). Profit priorities from activity-based costing. Harvard Business Review, 69(3), 130–135. Devedzic, V., & Radovic, D. (1999). A framework for building intelligent manufacturing systems. IEEE Transactions on Systems, Man, and Cybernetics, Part C – Applications and Reviews, 29(3), 402–419. Fuller, S., & Peterson, S. R. (1995). Life-cycle costing manual. National Institute of Standards and Technology Handbook 135, USA, Gaithersburg. Grahovac, D., & Devedzic, V. (2009). COMEX: cost management expert system. In Proceeding, the fourth international conference on information technology, Jordan, June 2009. Industrial and Financial Systems (IFS) (2001). Kaizen costing and value analysis (pdf document). . ISO 9000/ISO 14000 (2009). Quality management principles (online catalogue). (last visited April, 2009). Johnson, H. T. (1990). Activity management: Reviewing the past and future of cost management. Journal of Cost Management, 3(4), 4–7. Kaplan, R., & Cooper, R. (1998). Cost and effect: Using integrated cost systems to drive profitability and performance. Boston, MA: Harvard Business School Press. Kucˇinar, R. (2005). Prilog razvoju sistema za menadzˇment troškovima procesa. Master Thesis. University of Kragujevac (Department in Trebinje). Trebinje, B&H. Nedovic, Lj., & Devedzic, V. (2002). Expert systems in finance – A cross-section of the field. Expert Systems with Applications, 23(1), 49–66. Rolo Ask (2007). Why business rules? Why not expert systems? (web discussion). (last visited March, 2009). Sudmann, L. (2006). Upstream initiative analysis: Target costing, global strategic planning –Finance. P&G (pdf document). . Tomic´, B. (2005). JavaDON: Shell za razvoj ekspertnih sistema zasnovan na modelu OBOA. BSc Thesis. FON, University of Belgrade. Tomic, B., Jovanovic, J., & Devedzic, V. (2006). JavaDON: An open-source expert system shell. Expert Systems with Applications, 3, 595–606. Vanguard Software Corporation (2008). Knowledge automation overview (web page). (last visited April, 2009).
Contents lists available at ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
COMEX: A cost management expert system Danijela Grahovac *, Vladan Devedzic 1 Department of Software Engineering, FON – School of Business Administration, University of Belgrade, P.O. Box 52, Jove Ilica 154, 11000 Belgrade, Serbia
a r t i c l e
i n f o
Keywords: Expert system Cost management system ABC method
a b s t r a c t Understanding and using cost management is a key to success in any type of organization, because it helps managers to collect important data about the costs assigned to products, services and customers and use it for planning and control function. This paper discusses development and application of an expert system for reasonable cost related decision making and continuous cost management process improvement. The system is called COMEX (cost management expert system). This paper also shows how COMEX can be used for improvement of the billing department business processes. COMEX relies on information generated using the activity-based costing (ABC) method, accounting method based on activities, created to improve cost management system by focusing on the driver costs of individual activities. Still, this paper focuses on design and implementation of the COMEX system, and only briefly surveys the ABC method. Ó 2010 Elsevier Ltd. All rights reserved.
1. Introduction Beware of little expenses. A small leak will sink a great ship. Benjamin Franklin A cost management system (CMS) consists of a set of formal methods developed for planning and controlling an organization’s cost-generating activities related to its short-term objectives and long-term strategies. Business entities face two major challenges: achieving profitability in the short run, and maintaining a competitive position in the long run. An effective CMS must provide the managers with the information needed to meet both of these challenges. Cost management is not practiced in isolation. It is an integral part of general management strategies and their implementation. Examples include programs that enhance customer satisfaction and quality assurance, as well as programs that promote new product development. Cost management has a broad focus. It covers – but is not limited to – the entire process of continuous control of costs. Planning and control of costs is usually inextricably linked with revenue and profit planning. For instance, to enhance revenues and profits, managers often deliberately incur additional costs for advertising and product modifications. * Corresponding author. Present address: 50 Jerome Cres. #503, Hamilton, Ontario, Canada L8E1K6. Tel.: +1 905 869 4020; fax: +1 905 561 5320. E-mail addresses: [email protected] (D. Grahovac), [email protected] (V. Devedzic). URLs: http://goodoldai.org/danijela_grahovac (D. Grahovac), http://devedzic.fon.rs/ (V. Devedzic). 1 Tel.: +381 11 3950853; fax: +381 11 2461221. 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.04.073
The information generated from a CMS should help managers measure and evaluate performance. The measurements may be used to evaluate human or equipment performance or to evaluate future investment opportunities. However, although information from accounting systems helps managers to make decisions, the information and the accounting systems themselves are not cost management. The decision process, as all other business processes, includes a lot of problem solving actions/activities, experience, and heuristics. An expert consultancy may be necessary when a particular organization has to make a decision or solve some problem. These experts possess a specific knowledge and an experience in certain areas. When the company has to decide on the matter of something like a big investment, integration or advertising strategy, as a rule they will hire an expert for advice. Expert systems are able to support a decision-making process, and sometimes they can completely replace an expert. We have developed COMEX, a cost management expert system that supports the work of billing departments of utility companies. The purpose of COMEX is to satisfy the customer’s needs and stay concurrent with lower costs. For example, customer inquiries can consume a large percentage of a company’s resources, and reducing costs for this activity brings savings and customer’s satisfaction. This is applicable to all other activities in the company. COMEX recommends how to achieve these savings. COMEX can be applied to any kind of billing (e.g., electricity, water, or waste). The rest of this paper is organized as follows. Section 2 states the problem covered in this paper precisely. Section 3 briefly surveys relevant facts and methods from the domain of cost management, as well as JavaDON, the ES shell we used to develop COMEX.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Section 4 is the central section in this paper. It discusses the architecture and design of COMEX in detail. Section 5 shows a system evaluation, whereas Section 6 discusses other relevant aspects of COMEX. 2. Problem statement The purpose of this paper is to discuss how the expert system technology can be successfully applied in the domain of cost management. To this end, this paper covers the COMEX project in detail. The COMEX project objectives and the requirements include the following: the system must be well engineered from the knowledge engineering perspective; the system must be user friendly and easily understandable for the end user; the system should display questions in a logical, natural sequence, in a similar way that a human expert is thinking when solving related problems; the system must be properly designed to support the work of a billing department of any utility company; the system should suggest logical and useful solutions; the system should be open for further development and extensions. The completed COMEX system is now being constantly tested, evaluated, and explored for possibilities to be successfully applied on all organizations with a billing department (BD). The project revealed as useful and consultative, and by those facts, COMEX succeeded as the cost management system. 3. Related work 3.1. Cost management systems There have been significant criticisms regarding the lack of efficiency and capability of traditional cost management and accounting practices in the literature during the last two decades leading to a call for new techniques (Askarany & Smith, 2000). Designed for application under different business circumstances and for the other epoch, traditional cost management appeared to be a system with some major weaknesses that demonstrate a lack of overall competency. Traditional cost management systems are inaccurate in the way that they allocate costs, and managers armed with this system were not able to reduce costs, identify opportunities for improvement, and determine a more profitable product mix (Cooper & Kaplan, 1991). The lack of informational support was an obstacle for managers. Cost accounting had traditionally allocated overhead to products or services using only one volume-sensitive driver, typically direct labor. For organizations with high overhead and a mix of products or services, using a single cost driver may distort cost estimates (Cooper & Kaplan, 1991). Traditional methods of allocating overhead were therefore believed to be deficient in terms of improving global competitiveness (Johnson, 1990). Expert systems are useless with data gathered from traditional cost management systems. Information provided by these systems is inaccurate, and will result with inadequate solutions and advice. Hence, companies have to switch from a traditional costing system to a new method for cost management, and then interpret the data by the expert system. Modern cost management systems are adequate for the dynamic environment in which business operates. A setting for this approach requires including different types of experts (e.g., indus-
7685
trial, accounting or management) who are going to work on the preparation of a plan and control programs, as well as on resources, processes, cost, performance and risk management. However, there is a large number of different cost management methods and techniques (see http://www.valuebasedmanagement.net/). Some of the most popular ones are briefly covered in the Appendix. All such methods have certain advantages and disadvantages. In developing COMEX, we have used the ABC method for cost management (Kaplan & Cooper, 1998) as an underlying domain knowledge and methodology. The reason we opted for ABC is the fact that this method links company resources (inputs) and products/services in an innovative way, which is a precondition for any business process improvement (according to the ISO 9001 standards series requirements (ISO 9000, 2009)). Business process improvement in a company/enterprise/organization, in turn, brings improvement to the organization’s overall business improvement. 3.2. ABC method ABC is a powerful tool for measuring a company’s performance (Kaplan & Cooper, 1998). It recognizes cause–effect relationships between assigned objects costs. Once activity costs are identified, ABC assigns costs to products/services based on the resources they consume. ABC is a new method of cost assignment, more compatible with the increased high technology environment in which business operates. ABC assigns costs to products on the basis of the types and quantities of activities that must be performed to create those products. This costing system accumulates costs for activity centers in multiple cost groups at a variety of levels (unit, batch, product, and organizational) and then allocates these costs using multiple cost drivers (Barfield, Raiborn, & Kinney, 1998). Thus, the costs are assigned more accurately, and managers can focus on controlling activities that cause costs rather than trying to control the costs that result from the activities. The use of activitybased costing should provide a more realistic picture of actual production costs than has traditionally been available. Managers have more success in understanding how organization is using its own capital by assigning costs for defined activities using the ABC method. As a more accurate CMS than traditional cost accounting, ABC identifies opportunities to improve business process effectiveness and efficiency by determining the ‘‘true” cost of a product or a service (APCG, 2002). It is implemented in many companies around the world. The way of applying ABC systems differs from one organization to another. Some organizations use ABC system as the main CMS, but many of the ABC issues are selective and can be applied on subunits or separate functions of organizations. Note, however, that although ABC typically provides better cost estimates than the traditional overhead allocation process, it is not a panacea for all managerial concerns. This is the point where an expert system can take a part (Grahovac & Devedzic, 2009). 3.3. JavaDON expert system shell COMEX is implemented using the JavaDON ES shell (Tomic, Jovanovic, & Devedzic, 2006). It is an academic open source shell, currently used as a developing and teaching tool; its current version is a shareware (http://iis.fon.rs/additional%20materials.html). JavaDON is well suited for building complex expert system applications based on techniques such as rules and frames. The shell is based on the OBOA framework, whose partial conceptual model is shown in Fig. 1, illustrated in the UML notation. OBOA is a systematic approach to design and development of software for intel-
7686
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
ligent systems (Devedzic & Radovic, 1999). It is based on a multilevel, general object-oriented model of intelligent systems, and has been developed starting from some observations regarding difficulties in making extensions to methods and tools for building intelligent systems – scalability efforts – as well as in reusability and embedding-oriented efforts (Nedovic & Devedzic, 2002). Some of the most important elements of the OBOA framework are (Tomic et al., 2006): Knowledge element – all of the primitives and the units (no matter the complexity) are considered to be knowledge elements, since they all share common properties. Knowledge chunk – a basic logical statement. It can be used, e.g., to relate a frame slot with a value or with another slot. It is typically used as a building block for rules. Domain – determines the basic type and range of a value. The basic types are: string, integer, floating point and boolean.
4. COMEX architecture and design COMEX interprets the data resulting from applying the ABC method. The ABC method studies a company’s activities to classify them into activity centers, than traces costs to those activities and determines the cost drivers. For example, the maintenance activity has the number of machine hours as its a cost driver, and the correspondence activity has the number of letters as the a cost driver. The map of the activities supplied by the ABC method is essential for a knowledge engineer during the design phase of the COMEX development. 4.1. Knowledge representation COMEX is essentially a frame-and-rule-based system. There are nine frames in its knowledge base:
The COMEX knowledge base (system level) and knowledge elements (integration level) are domain-depended (e.g., cost management), and all other elements are domain-independent.
Customer Activity BillInquiry WebSite
SYSTEM
INTEGRATION
COMEX KNOWLEDGE BASE
BLOCKS
0..* KNOWLEDGE ELEMENTS
0..* KNOWLEDGE ELEMENT
UNITS
RULE IF
FRAME THEN
1 0..*
0..*
1..* 0..*
0..*
CLAUSE
SLOT 0..* 1 0..*
0..*
KNOWLEDGE CHUNK 0..* 1
0..*
0..1
1
ACTION
PRIMITIVES
1
0..*
1..*
ATRIBUTE
0..*
1
DOMAIN
0..*
MEDIA
0..* RELATION
0..1
VALUE
Fig. 1. COMEX in the context of OBOA framework/architecture (adapted according to Tomic et al. (2006)).
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
AccountVerification CustomerInquiry Correspondence Expert ExpertSolution
The frames are, of course, inter-related. The UML class diagram in Fig. 2 explains the relations among the frames.
7687
As we mentioned before, COMEX is created to support the work of the billing department (BD). BD has two types of customers – commercial and residential. The activities that have to be done in order to generate the bill for these types of customers are different. All activities are grouped into four possible categories (see Fig. 2). In order to improve the business process, the user can select the customer type and then the activity type. COMEX will then lead the user though an appropriate dialog (a series of questions). For
Fig. 2. Class diagram relating the major concepts of the COMEX system.
Fig. 3. COMEX frames (screenshot from JavaDON).
7688
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
example, if the BillInquiry activity type is selected, the dialog will be related to the data from the specific type of bill (e.g., in the case of utility billing, the number of printed lines or information redundancy). After a certain number of the system’s questions, a solution may be reached; it is the job of the Expert frame, which tightly collaborates with the activity frames. Note that a solution could be offered after each step in the dialog. In case of the BillInquiry activity, if no solution is reached during the dialog defined by the questions related to the BillInquiry frame, the WebSite frame will take over and will come up with a new set of questions. These are related to the option of billing the customer through a WebSite, or sending bills as email (e.g., ‘‘Do you have your own website?”, ‘‘Do you have printed notification about website on each bill?”). They will be asked until COMEX comes up with a solution (such as ‘‘Offer the customer to switch to online billing, with the bonus of **$”). The WebSite frame refers to the BillInquiry activity, and it is separated because it offers a new approach to BillInquiry process improvement. 4.2. Frames Fig. 3 shows the nine frames of COMEX as they are implemented in JavaDON. In JavaDON, each frame can have zero or
more subframes and parent frames. For example, there are two slots in the Customer frame but no subframes and parent frames, Fig. 3 (SubFrames and ParentFrames of the Customer frame have no children nodes). Although it appears in Fig. 3 that there is no inheritance among the COMEX frames, it is not so (see Fig. 2). It just appears to be so due to the fact that JavaDON shows all frames as branching directly from the root of the knowledge base. JavaDON frames have slots that represent the corresponding frame’s attributes. Each slot contains the definition of the attribute it represents, a question to be presented to the end user in dialogs with COMEX whenever the system asks the user to supply a value for the attribute, possible answers, and a description (canned text used for providing explanations in the user’s sessions with COMEX). A slot’s attribute is shown in its node in the expanded tree. For example, the typeOfCustomer slot from the Customer frame, Fig. 4, is defined to ask the user ‘‘What is the type of the customer?”, and the user is supposed to select between two possible types of customers: Commercial and Residential. 4.3. Rules The structure of COMEX rules is shown in Figs. 5 and 6. For example, the BillCostComGreat rule in plain English says: If the customer is a commercial customer, and the customer’s bill cost is greater than or equal to 8.5 Then ask the user about the type of activity The same rule in COMEX includes the following knowledge chunks: a) Customer:typeOfCustomer=(Commercial;1.0) b) Customer:billCost>=(8.5;1.0) c) ask_question(Activity:typeOfActivity)
Fig. 4. The typeOfCustomer slot.
1
RULE
1
0..*
0..* THEN CLAUSE
IF CLAUSE
0..*
0..*
1
KNOWLEDGE CHUNK
1
Fig. 5. Rules and chunks relationship.
Fig. 6. The structure of rules in JavaDON/COMEX (a screenshot from JavaDON).
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
The rule (i.e., its knowledge chunks) states that if the slot typeOfCustomer from the Customers frame has the value Commercial and the billCost slot from the same frame has a value greater than or equal to 8.5 (IF clause), then the system will ask the question (the ask_question action) related to the typeOfActivity slot of the Activity frame. This means that if the selected customer class is Commercial and the bill is greater than a certain limit, COMEX will proceed with the process improvement. The value 1.0 in the knowledge chunks denotes the corresponding certainty factor. In addition to IF and THEN parts, each rule has its TYPE (which can be either AND or OR) and IMPORTANCE (the rule’s priority, used to resolve conflicts when multiple rules can fire). IMPORTANCE is a rule feature provided by JavaDON and is used to control the order of rule firing. In each inference cycle, the rule with the highest importance is selected to fire if more than one rule is satisfied. By a more detailed analysis of some of the COMEX rules, one can get a closer insight into the way COMEX solves problems. Rule 1 initiates each dialog with the user. It has the highest importance (10) and always fires first. Its If clause is empty, so it will always fire unconditionally at the beginning of the session. Rule 1: INIT (Importance=10) If Then ask_question(Customer:typeOfCustomer)
The first question that COMEX will ask the user during his/her session with the system is ‘‘What is the type of the customer?”, and the user is supposed to select between two possible types of customers: Commercial and Residential (see the Then clause of Rule 1). This is a very important partitioning, since the billing process is different for the two types of customers. The Then clause executes the action ask_question(Customer:typeOfCustomer) related to the typeOfCustomer slot of the Customer frame, and the user is asked to enter the appropriate answer. Rule 2 has Importance=9, and always fires second, since all other rules do not have such a high importance. This rule’s If clause is also empty, so Then clause will always fire unconditionally. The Then clause refers to the ask_question(Customer:billCost) chunk that executes the action related to the billCost slot of the Customer frame with the question ‘‘What is your cost per account?”. The user enters the cost that the customer has made by a certain bill. Rule 2: BillCostCommRes (Importance=9) If Then ask_question(Customer:billCost)
7689
Rule 2.1: BillCostCommLess (Importance=0) If Customer:typeOfCustomer=(Commercial;1.0) AND Customer:billCost<(8.5;1.0) Then Expert:finalSolution=(‘‘You are under market average limit!”) ExpertSolutions:solution=(‘‘You don’t have to do process improvement.”) Rule 2.2: BillCostCommGreat (Importance=0) If Customer:typeOfCustomer=(Commercial;1.0) AND Customer:billCost>=(8.5;1.0) Then ask_question(Activity: typeOfActivity) Rule 3 has Importance= 10, which means it always fires last. Its If clause is empty, so it will always fire unconditionally. The Then clause contains the chunks show_value(Expert:finalSolution) and show_value(ExpertSolutions:solution), which execute the show value action. In this case, these are the values of the finalSolution and solution attributes, where the COMEX solutions and advice are stored. Rule 3: ShowSolution (Importance= 10) If Then show_value(Expert:finalSolution) show_value(ExpertSolutions:solution) If the selected customer class is Residential, the inference process proceeds in a similar way, this time working with Rule 2.3 and Rule 2.4. Rule 2.3: BillCostResLess (Importance=0) If Customer:typeOfCustomer=(Residential;1.0) AND Customer:billCost<(3.5;1.0) Then Expert:finalSolution=(‘‘You are under market average limit!”) ExpertSolutions:solution=(‘‘You don’t have to do process improvement.”) Rule 2.4: BillCostResGreat (Importance=0) If Customer:typeOfCustomer=(Residential;1.0) AND Customer:billCost>=(3.5;1.0) Then ask_question(Activity:typeOfActivity) Fig. 7 shows the sequence diagram of the COMEX inference process from the beginning to the point where the type of activity is chosen.
If the selected customer class is Commercial, then Rule 2.1 or Rule 2.2 is fired. Both of these rules have two If clauses. The first one checks the type of the customer, and the second one checks whether or not the entered value is greater than, less than, or equal to the market average value (8.5). If the current value is less, than Rule 2.1 is fired. This results in ending a cycle, and a rule presents the solution, stating that the organization is under the market average limit and there is no real need for process improvement. In order to show this solution, the system fires Rule 3 (see description below). If the value of billCost is greater than or equal to the market average value, the Then clause will fire the chunk ask_question(Activity:typeOfActivity). The chunk executes the action of asking a question. The question located in the typeOfActivity slot is ‘‘Which activity do you want to improve?”. When this rule is fired, the user is supposed to select one of four possible types of activities for further improvement.
4.4. Dialogs Another very important piece of knowledge in the COMEX knowledge base is the typeOfActivity slot in the Activity frame. The related question that COMEX will ask the user during his/her session with the system is ‘‘Which activity do you want to improve?” There are four possible answers: BillInquiry, CustomerInquiry, AccountVerification and Correspondence (Fig. 8). These activities origin from the ABC method applied to BD. After the user selects an activity, COMEX will typically ask a question about the average percentage of the total cost allocated to the chosen activity. The average slot in the Activity frame stores the value of the total cost allocated in the activity of the billing department (BD). Depending on the activity selected, the value entered is compared to the actual market average percentage. For
7690
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 7. Start COMEX – sequence diagram.
example, the market average percentage for the AccountVerification activity is 16%. Anything under this value does not need improvement; in case of a greater value, COMEX goes forward into improvement and asks question Q11: Question: Q11 What percent of the total cost is allocated for this activity? The related rules in the knowledge base are: Rule 17: IF A=‘‘Commercial” AND 16
Rule 18: IF A=‘‘Commercial” AND 16>=average THEN ask-question Q11
Fig. 9 shows the sequence diagram for one of the four branches, i.e., CustomerInquiry improvement. There are five different endings that can happen, depending on the user’s answers. 4.5. User modeling COMEX users are knowledge engineers and end users. A knowledge engineer is the one who creates the knowledge base and tests the system. Speaking in terms of UML, the use cases related to knowledge engineers are illustrated in Fig. 10. They include building the knowledge base, creating a new project, opening and updating the knowledge base, and the like. During the design of the COMEX system, the knowledge engineer communicated with a cost management expert, an ABC expert, and accounting and system quality experts. All the important knowledge was entered in the knowledge base through multiple iterations (Grahovac & Devedzic, 2009). COMEX end users are ABC experts, accountants, managers, business analysts, and students. End users do not have the permission to update knowledge base elements. They can open a project (open project COMEX), enter the session data (enter data) and start the forward-chaining inferencing (forward chaining) (Fig. 11) (Tomic´, 2005). 4.6. Reasoning and conflict resolution
Fig. 8. Selecting an activity to improve.
COMEX reasons using forward chaining with its rules. The inference mechanism of COMEX is presented by the example shown in Fig. 12 and Table 1. The example presents one COMEX cycle, and it shows how automatic inferencing starting from the knowledge base and the working memory data brings up new solutions. The figure shows that variables A and B are stored in the working memory, which means that the user has answered questions about the type of customer (Residential) and entered a cost per account greater than the current market average (3.5). Since COMEX is a forward-chaining system, in each cycle of its reasoning process a set of candidate rules to fire is created (the
7691
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 9. CustomerInquiry activity – sequence diagram.
Create Project COMEX <
<
<
Knowledge Engineer
Build Knowledge Base
Edit COMEX
Save COMEX
<
Close COMEX
<
Open Project COMEX
Fig. 10. Knowledge engineer’s use cases with COMEX.
rules that have their IF clauses satisfied). Conflict resolution and selection of the rule to fire is done by selecting the highest priority rule. Each rule can be fired only once. 4.7. A session with COMEX The reasoning process is initiated by having three rules automatically satisfied when the session starts: INIT (Fig. 13), BillCostCommRes (Fig. 15) and ShowSolution (Fig. 18). The INIT rule’s importance is set to 10 and thus it will be fired first. As a result, the user is presented the question about the customer type as in Fig. 14.
Upon selecting the customer type and clicking Next, the next cycle of forward chaining is executed and the BillCostCommRes rule (with the IMPORTANCE set to 9.0, Fig. 15) is fired. Its THEN clause executes the action ask_question(Customer:billCost) and the user is asked to enter the appropriate bill cost, Fig. 16. In this example, it is assumed that the user is a Commercial customer and that he has entered the value 6.5 that corresponds to this customer type. This value, as same as all others, is a result of implementing the ABC method in the BD. This results in COMEX activating the BillCostComLess rule (Fig. 17), because the bill cost entered is less than the market average. The rule’s THEN clauses are added to the working memory and
7692
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Open Project COMEX <
Inference
User
<
Forward Chaining
Enter Data
Fig. 11. End user use cases.
Fig. 12. COMEX – forward chaining.
Table 1 Facts. Symbol
Variable
Value
Rule
A B C D E F
typeOfCustomer billCost typeOfActivity Average Letter finalSolution and solution
Residential >3.5 Correspondence >6.0 True ‘‘Conclusion”
Rule Rule Rule Rule Rule Rule
1 and Rule 2 2.4 4.4 4.4.1 4.4.1.1 3
Fig. 14. The start of a session.
Fig. 13. The INIT rule.
the slots Expert:finalSolution and ExpertSolution:solution are assigned the appropriate values. The ShowSolution rule (Fig. 18) will use these values when presenting the solution that COMEX displays to the user at the end of the session, Fig. 19. Note that ShowSolution has an empty IF clause, so it will certainly fire. However, it has the lowest IMPORTANCE ( 10), hence it will fire only at the end of the session.
Fig. 15. The BillCostCommRes rule.
Of course, this simple example only demonstrates how COMEX works. In real-world examples, its reasoning takes many more steps and can get quite complex.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Fig. 16. The result of executing the ask_question(Customer:billCost) action.
Fig. 17. The BillCostComLess rule.
Fig. 18. The ShowSolution rule.
7693
to business intelligence (BI) applications, as well as to some expert systems in the more general domain of business management. The first comparison is made between COMEX and BI applications. BI applications provide an integrated ABC method based modules for cost management. The calculated data are the real picture of an organization’s costs, but the important question therefore is exactly what to do with obtained data and what kind of concrete steps to take in order to improve the business process and make some savings. COMEX is able to reasons over these data to provide practical advice and recommendations, if the calculated costs are for the billing department. Further development of COMEX will include a completely automated input of the calculated costs. BusinessObjects ERP profitability and cost management application (Weismantel et al., 2009) gives a picture of real costs and profits by using the ABC method, and makes possible to ‘‘dig” a source of the costs. At the end, attractive and functional reports require an expert for advice. The analytic module compares a company’s strategic goals with current data and results show where the company is now, and where it is supposed to be. How? It requires a human expert again. SYSPRO Enterprise Software,2 IBM Cognos 8,3 Microsoft Dynamics4 and NetSuite,5 operates on similar principles as BusinessObjects, with more or less analytics, data mining, forecast and optimization included. However, after all these creative reports conclusion is essentially the same, it needs someone to suggest what to do. That is exactly the right place for deploying an ES, and in the case of utility companies, the place for COMEX. The second comparison is made between COMEX and some expert systems in the more general domain of business management (Table 2). Regarding reporting, Vanguard automatically identifies the most influential company’s financial risk and shows results as graph presentation, similar as Agena. Agena has an option to present data using tables and risk maps. Cost Advantage shows comparative tables of estimated and actual costs. Thus, the user is able to come up with desirable costs, by simulating different models of product design. Contrary to that, the results of COMEX are advice and solutions, but it does not validate results with some$”. This additional functionality is not thing like ‘‘saving is fully developed yet because the Formula function in the JavaDON shell is not implemented yet. Regarding the usage of this kind of systems, currently on the highest place is Vanguard, followed by Agena. These commercial software packages are more general and target various types of industries, whereas Cost Advantage and COMEX target more specific ones. Cost Advantage is developed to support the work of companies that design products, and COMEX to support the work of utility companies. Tools and applications are quite different for all systems. Furthermore, the described systems are completely incompatible and simply it is not possible to connect them. This is a typical case of the software diversification, and it does not merge with the globalization concept that is initiated in a previous decade. Again, regarding knowledge representation and collection, there is an entire set of different techniques. The Vanguard is based on the production rules (just like COMEX), and decision trees. In
Fig. 19. The solution presented by COMEX.
5. Evaluation We are not aware of another expert system in the area of cost management, especially not of any such a system that builds upon the ABC method. Hence we have evaluated COMEX by comparing it
2 URL: http://www.sysprosolutions.com/files/fact_sheets_financial/Activity_Based_Costing.pdf, last accessed: April, 2009. 3 URL: http://www.cognos.com/products/cognos8businessintelligence/index.html, last accessed: April, 2009. 4 URL: http://www.microsoft.com/dynamics/using/default.mspx, last accessed: April, 2009. 5 URL: http://www.netsuite.com/portal/products/netsuite/main.shtml, last accessed: April, 2009.
7694
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Table 2 Summary of the expert systems. System
Vanguard, risk management module
Agena
Cost advantage
COMEX
Domain
Risk analysis
Risk analysis
Cost management
Output (reports)
List of identified risks
Tables, graphs and risk maps
Production cost management Costs table
Usage
More than 60 companies around the worlda Vanguard Studio, DScript Decision trees and production rules
RBC Bank, Philips, Simens, Motorola, Bocsh, etc. Java API Risk maps, Bayesian networks
Hierarchical decision models Currently used Vanguard (2008)
Probability-based decision making, Bayesian inference engine Currently used Agena (2005)
Tools/applications Knowledge representation and collection Reasoning Status Reference a
Allegro CL’s common Lisp Production rules written in natural language
Advice and solution for BD’s processes improvement Power-utility of Republic of Srpska JavaDON Production rules, frames, O–A–V triplets
N/A
Forward chaining
Currently used Cognition (2008)
Testing and limited use Grahovac and Devedzic (2009)
Motorola, Cummins,. . .
URL: http://www.vanguardsw.com/customers, last accessed: April, 2009.
addition, COMEX represents knowledge with frames and O–A–V triplets. Production rules are the basis for Cost Advantages’ knowledge representation as well, but in Cost Advantage they are written in natural language. Agena has an embedded Bayesian Networks technique as well as risk maps. What is better? An answer for this particular question can be only subjective. The reasoning or inferencing is different from one application to another. Forward chaining is a very popular technique, and is a part of COMEX. All of the reasoning in JavaDON is done by using the forward chaining inference technique (Tomic et al., 2006), thus it is a part of COMEX, too. The JavaDON team is currently working on the full implementation of the backward chaining technique, which apparently is not going to affect the current COMEX configuration. Vanguard’s decision trees are necessarily associated with a hierarchical decision model, and likewise Agena is with Bayesian inference engine. Vanguard’s decision about this kind of reasoning is, likely, because of the system complexity, since risk management is only one of the modules. Agena is oriented only on risk management associated with probability-based inference that Bayesian technique is based on.
During the development of COMEX we were asked the following questions: Why an expert system? Is this technology ‘‘in” or it was dismissed in 1990s? The answers to these questions reflect the authors’ inspiration to research the subject. If we called COMEX a business rule management system (BRMS), it would certainly be ‘‘in”. And what is a BRMS? The little secret is that a BRMS was originally called ‘‘expert system” and its rule engine was called ‘‘inference engine”. Then it was called ‘‘business rule engine”, and today this kind of system is known as ‘‘business rule management system”. Hence COMEX may change its name/categorization over time. But not yet. Expert systems deserve all due respect. Expert systems didn’t really go away. They went undercover (Rolo, 2007). This statement can be justified by, e.g., AMEX’s Authorizer’s Assistant (ES for credit cards transactions management) (BizRules, 2008). This magnificent ES is working perfectly since 1988. COMEX is developed using JavaDON’s current capabilities. Simultaneously with JavaDON development, COMEX will develop, too. Currently, the team is working on developing JavaDON’s interoperability with an external database; in case of COMEX, a relationship with Excel tables would be very helpful.
7. Conclusions 6. Discussion Despite the fact that COMEX supports the work of billing departments of utility companies, it can be applied to any kind of billing (e.g., electricity, water, or waste). The central idea of the COMEX project was to illustrate how business effectiveness could be significantly improved by applying a modern method for cost management and expert system technology to interpret the data. COMEX demonstrates that synergy between ES and modern economic tools (i.e., ABC) results in remarkable improvement of cost management. Besides the assumption about using the ABC method, COMEX is based on activities, too. One of the most essential branches is BD’s activity selection. The ABC method provides these activities. COMEX solutions are in fact recommendations for improvement of BD’s processes. Naturally, cost savings come along with process improvement. Cost management is generally very popular, mainly because it has a direct impact on profitability. Thus, COMEX can find its place in practical application, both as manager’s decisionsupport aid, and as a training tool for employees. As an application, COMEX is easy and friendly to use. It is made for users without any knowledge about expert systems, but in need for sound advice about BD’s process improvement and costs saving.
COMEX is a cost management expert system that supports the work of billing departments of utility companies. It relies on using the ABC method for cost management. ABC can calculate cost data for specific activities in utility companies, and COMEX interprets the data. COMEX demonstrates how business effectiveness can be improved by applying a new method for cost management and expert system technology to interpret the data resulting from applying the method. In other words, the ABC method calculates the real costs and COMEX reasons over the data to provide useful advice and recommendations. Together, they bring significant improvement to cost management. Although specifically developed for a power utility company (Power Utility of the Republic of Srpska), COMEX has an option to be extended to other organizations that incorporate a billing department and print out a large amount of bills. Examples of such other companies include telephone companies, TV companies, Internet providers and the like. Our current and future work is oriented towards putting the COMEX demo online and commercialization of the developed full version. As expert systems developed with JavaDON can be either desktop or Web applications, one more COMEX improvement could be an application for the Web.
D. Grahovac, V. Devedzic / Expert Systems with Applications 37 (2010) 7684–7695
Appendix A. Popular cost management methods Popular methods used today in various cost management systems include: Target Costing, Kaizen Costing and Life Cycle Costing. A.1. Target Costing Target Costing (TC) – (Kucˇinar, 2005) – originated in Japan, as a result of requirements for creating a market-oriented concept under global economic terms, growing competition, new technologies expansion, innovations and reduced product life cycle. From the objectives and methodology point of view, TC is differing from traditional systems based on ‘‘cost-plus” accounting. TC applies to the design phase with a specific focus on research, development and product design activities. This approach seeks to listen, understand and anticipate consumers’ needs and aspirations, and emphasizes their willingness to purchase and pay for the quality and proper functionality of the product. It presumes integration of the overall business and management knowledge, and within the design phase, routes realization of the strategic goals, as cost reduction, performance and profit increasing. A disadvantage of Target Costing is the danger of ‘‘killing” potentially good ideas early, since getting the real costs is difficult. It is almost impossible to develop an ‘‘IPhone” or a new ‘‘Pampers” under the pure ‘‘Target Costing” thinking (Sudmann, 2006). A.2. Kaizen Costing Kaizen Costing (KC) is a costing technique that reflects continuous efforts to reduce product costs, improve product quality, and/ or improve the production process after manufacturing (Barfield et al., 1998). Yashihuro Monden from Japan, who developed Kaizen Costing, defines it as the maintenance of present cost levels for products currently being manufactured via systematic efforts to achieve desired cost level (IFS, 2001). Contrary to TC, which is applied to a particular product (new or an existing one) that is currently under (re)design, KC is applied to a product that is already in production. The primary goal of this method is to permanently reduce specific product costs, using improvements in small steps. One of the related characteristics for the continuous process improvement is that cost reduction authority and responsibility delegates from high managerial level to operative levels. A disadvantage of the KC technique is that the seek for cost savings results in a reduced quality of the product. A.3. Life Cycle Costing Life Cycle Costing (LCC) is the accumulation of costs for activities that occur over the entire life cycle of a product from inception to abandonment by the manufacturer and consumer (Barfield et al., 1998). LCC is a successful answer to the criticism to other
7695
methods, especially traditional ones, that states that cost management and products performance ‘‘start too late and end too soon”. However, the LCC system has a small value by itself. It is most useful when it can be compared to the LCC of other design alternatives, which can perform the same function, in order to determine which alternative is the most cost effective one for this purpose (Fuller & Peterson, 1995). References Agena Limited (2005). Getting started with AgenaRisk.