Electronic Business: Concepts, Methodologies, Tools, and Applications (4-Volumes) P47 docx

394 Dynamic Pricing for E-Commerce places, products traded in posted-price markets are no-niche items and exhibit continuous demand over time. The Web site of online book merchant Amazon (http://www.amazon.com) is an example of a posted-price market. A buyer interested in a particular book enters the necessary information through a form on Amazon’s Web site to request the price of the book and receives the price in response. Modern seller Web sites employ automated techniques for the different stages of e-commerce. Intermediaries called intelligent agents are used to automate trading processes by implementing different algorithms for selling products. For ex- ample, Web sites such as MySimon (http://www. mysimon.com) and PriceGrabber (http://www. pricegrabber.com) automate the search stage by employing the services of intelligent agents called shopbots. Shopbots enable buyers to make an informed purchase decision by comparing the prices and other attributes of products from thousands of online sellers. Automated price comparison by buyers has resulted in increased competition among sellers. Sellers have responded to this challenge by using intelligent agents called pricebots that dynamically determine the price of a product in response to varying market conditions and buyers’ preferences. Intelligent agents are also used to enable other e-commerce processes, such as supply-chain management and automated negotiation. In this article, we focus on the different algorithms that sellers’ pricebots can use for the dynamic pricing of goods in posted-price markets. BACKGROUND Over the past few years, online dynamic pricing has stimulated considerable interest in both the commercial and research communities. Increased SUR¿WVDQGUDSLGO\FOHDULQJLQYHQWRULHVUHVXOW- LQJIURPHI¿FLHQWSULFLQJKDYHHQFRXUDJHGWKH development of software pricing tools includ- ing Azerity (http://www.azerity.com) and Live Exchange (http://www.moai.com). Automated dynamic pricing for posted-price markets has been implemented and analyzed using simulated market models (Brooks, Gazzale, MacKie-Mason, & Durfee, 2003; Dasgupta & Melliar-Smith, 2003; Kephart, Hanson, & Greenwald, 2000). Most of these models consider the price of a product as the only attribute affecting a buyer’s purchase decision. Surveys of consumers who purchase products online, reported in Brown and Gools- bee (2000) and by ResellerRatings (http://www. resellerratings.com), reveal that online buyers are frequently willing to pay an elevated price for particular product attributes such as delivery time, seller reputation, and service. Moreover, the preferences of buyers vary over time depending on exogenous factors such as sales promotions, aggressive advertising, and the time of year. Therefore, it is important for an online seller to differentiate a product using multiple attributes and to determine the purchase preferences of a potential buyer over those attributes so that the seller can tailor its offer to the buyer’s require- PHQWVDQGLPSURYHLWVSUR¿WV In online markets, a seller must determine the prices that its competitors charge for a product so that it can place its price at a competitive advan- WDJH7KHUDSLGÀXFWXDWLRQRIPDUNHWSULFHVFDQ leave a seller with outdated competitor price infor- mation that can cause the seller’s dynamic-pricing algorithm to function incorrectly. However, it is GLI¿FXOW IRUVHOOHUVWR REWDLQ SULRU LQIRUPDWLRQ about buyers’ parameters. Therefore, it is desirable if online sellers do not assume prior knowledge about market parameters, but rather use a learn- ing algorithm (Brooks et al., 2003; Dasgupta & Hashimoto, 2004) to determine changing market parameters dynamically. 395 Dynamic Pricing for E-Commerce DYNAMIC PRICING USING INTELLIGENT AGENTS In an automated posted-price market, a seller em- ploys the services of a pricebot that dynamically FDOFXODWHVDSUR¿WPD[LPL]LQJSULFHRIDSURGXFW LQUHVSRQVHWRÀXFWXDWLRQVLQPDUNHWSDUDPHWHUV VXFKDVWKHSULFHVDQGSUR¿WVRIFRPSHWLQJVHOO- ers and the reservation prices of buyers. The seller posts the updated product price at regular intervals to attract buyers while maintaining a competitive edge. The market model we consider is based on the shopbot economy model of Kephart, Hanson, and Greenwald (2000), which makes simplify- ing assumptions about the online economy that facilitate analysis while retaining the essential features of the market. It consists of S sellers who compete with each other for B buyers (BS). Only one type of commodity is traded in the PDUNHW$VHOOHUEHKDYHVDVDSUR¿WPD[LPL]HU DQGKDVDVXI¿FLHQWVXSSO\RIWKHFRPPRGLW\IRU the lifetimes of the buyers. Buyers return to the market repeatedly to purchase the commodity. Examples of such markets include telephone and Internet services. A product is characterized by multiple attri- butes. A seller offers a slightly different price for the product along each of its attributes. As shown LQ)LJ XUHDEX\HU¿ UVWUHTXH VW VDTXRWHI URPWKH sellers for the price based on his or her preferred product attribute, and then selects the seller that makes the best offer. The buyer’s preferred at- tribute is not revealed to a seller when the buyer PDNHVDTXRWHUHTXHVW7KHUHIRUHDSUR¿WPD[L- mizing seller must determine a buyer’s preferred attribute in response to the buyer’s quote request. The seller then calculates a competitive price for the product along the buyer’s preferred attribute and makes an offer to the buyer. Dynamic-Pricing Algorithms %HFDXVHRQOLQHVHOOHUVDUHSUR¿WPD[LPL]HUVWKH objective of a seller is to determine a price for each attribute of the product that maximizes the VHOOHU¶VSUR¿W+RZHYHUWKHSULFLQJIXQFWLRQRID seller cannot be stationary as there are other com- peting sellers who revise their prices to improve their offers and attract buyers away from each other. Therefore, the seller updates the prices it charges on different product attributes at intervals in response to competitors’ pricing strategies and changes in the buyers’ preferred attributes. We describe in the following sections some pricing algorithms used by an online seller’s pricebot to determine the price of a product. We omit the subscript for attribute a i in the price and SUR¿WQRWDWLRQIRUWKHVDNHRIFODULW\:HLOOXVWUDWH the algorithms for a single seller assuming that it Figure 1. A hypothetical market showing two buy- ers, B1 and B2, with preferred attributes a1 and a3, respectively, making a quote request to four sellers, S1, S2 , S3, and S4, and then selecting the seller that offers the best price for the product on their respective attributes. The four-tuple below each seller denotes the normalized price on the different product attributes offered by that seller. B 1 B 2 S 1 S 3 S 2 S 4 < 0.8, 0.4, 0.3, 0.5 > < 0.9, 0.3, 0.6, 0.7 > < 0.7, 0.2, 0.8, 0.1 > < 0.6, 0.1, 0.7, 0.4 > Preferred attribute is a Selects seller S 1 4 Preferred attribute is a Selects seller S 3 1 Get current offer from sellers for attribute a 1 Get current offer from sellers for attribute a 3 396 Dynamic Pricing for E-Commerce is competing with other sellers in the market. We use p t to denote the price charged by the seller during interval t. Derivative-Following Algorithm In the derivative-following (DF) algorithm, a VHOOHUXVHVLWVSUR¿WLQIRUPDWLRQVLQFHWKHODVWSULFH update to adjust its price in the next interval. If WKHSUR¿WLQWKHODVWLQWHUYDOKDVLQFUHDVHGIURP its previous value, the price for the next interval continues to move in the same direction as in the ODVWLQWHUYDO2QWKHRWKHUKDQGLIWKHSUR¿WLQ the last interval has decreased from its previous value, the direction of the price movement is the reverse of the direction in the last interval. The equation for updating the price during interval W1 using the DF technique is given by: p W1 = p t G t sign(S t - S t -1 )sign(p t - p t -1 ) Here, S t UHSUHVHQWVWKHSUR¿WPDGHE\WKHVHOOHU during interval t, and G t represents the amplitude of the price change and is drawn randomly from the uniform distribution U[l, u], where l > 0 and u > 0. In the DF algorithm, the price of the product LVXSGDWHGEDVHGRQWKHSUR¿WLQIRUPDWLRQIURP only the last interval. Therefore, the DF strategy LVQRWYHU\HI¿FLHQWLQG\QDPLFDOO\WUDFNLQJWKH SULFHRIDSURGXFWLQDUDSLGO\ÀXFWXDWLQJPDUNHW $PRUHHI¿FLHQWWHFKQLTXHWKHmodel-optimizer (MO) algorithm described next, employs the his- WRULFDOSULFHDQGSUR¿WLQIRUPDWLRQRIWKHVHOOHUWR update the price during the next interval. Model-Optimizer Algorithm A seller using the MO algorithm maintains its SULFHYVSUR¿WSUR¿OHRYHUWKHODVWh intervals, where h denotes the size of the history window of the seller, as shown in Figure 2 for h = 5. The MO algorithm works as follows: 1. Assign weights to the last h points in the SULFHYVSUR¿W SUR¿OH RI WKH VHOOHU 7KH weight of a point expresses its relevance to current market conditions. Older points are less relevant and are assigned lower weights; more recent points are more relevant and are assigned higher weights. 2. Fit a polynomial over the h points in the his- tory window of the seller using a nonlinear regression approach. 3. Use a nonlinear optimization scheme, like the Nelder-Mead algorithm (Nelder & Mead, 1965), to determine the price that FRUUHVSRQGVWRWKHPD[LPXPSUR¿W Although a large history window h might yield accurate results, it increases the time required for calculating the price for the next interval. If the seller is relatively slow in dynamically updat- ing its price, its competitors might outperform it. Therefore, the value of h should be selected carefully to balance accuracy with rapidity in price calculation. Figure 3 shows the variation in prices over time for three competing sellers in a market using the 02DOJRULWKP$VVKRZ QLQWKH¿J X UHWKHVHOOHUV using the MO algorithm engage in repeated cycles of price wars with each other. The reason for the price wars is that there are buyers with different Figure 2. The operation of the MO algorithm with h = 5 397 Dynamic Pricing for E-Commerce preferences in the market. For simplicity, we as- sume that there are only two types of buyers. • A-type buyers that do not have price as the preferred attribute. Such buyers select a seller using other undetermined criteria, which we model as selection at random. • B - t y p e b u y e r s t h a t h a v e p r ic e a s t h e p r e f e r r e d attribute. Such buyers shop for the lowest price in the market and select the seller of- fering the lowest price. Based on a survey of online markets (Clark, 2000), we assume that the ratio between A-type and B-type buyers in the market is 1:3. Because B-type buyers are greater in number, they gen- erate the majority of the revenue for the sellers. Therefore, the sellers reduce the price of the com- modity in successive intervals so that they can attract the maximum number of B-type buyers by offering the lowest price among competitors, thereby undercutting each other. This price war continues until each seller’s price reaches the production cost p co of the commodity. Each seller KDV]HURPDUJLQDOSUR¿WLQVXFKDVFHQDULR$WWKLV point, the sellers realize that they can make more SUR¿WE\LQFUHDVLQJWKHSULFHRIWKHFRPPRGLW\WR attract A-type buyers instead of charging p co to attract B-type buyers. Therefore, the sellers reset their prices to a high value and another cycle of the price war ensues. The drawback of the MO algorithm is that it charges a uniform price to all buyers irrespec- tive of the buyers’ preferences. However, this uniform pricing results in a loss of revenue from A-type buyers who are willing to pay a much higher price for a commodity than B-type buy- ers. Thus, the buyer population can be segmented into different clusters depending on the buyers’ preferences, and a different price can be charged for each segment. Although some online merchants such as Ama- zon have implemented dynamic pricing, it is yet to be adopted widely in e-commerce. The principal drawback of the dynamic pricing mechanism that those merchants have employed is that it offers identical products to different buyers at different prices, resulting in discontented buyers. A better pricing strategy would be to identify the preferred attribute of different buyers and charge a slightly different price for the product based on a buyer’s preferred attribute as described below for the multiattribute dynamic-pricing algorithm. Multiattribute Dynamic Pricing As shown in Figure 1, a buyer compares the prices offered by different sellers based on his or her preferred product attribute. To make a competitive offer in response to a buyer’s purchase request, DVHOOHULGHQWL¿HVWKHEX\HU¶VSUHIHUUHGDWWULEXWH to offer a competitive price to the buyer on that attribute. The seller estimates the distribution f pa of a buyer’s preferences over the product at- tributes and then uses it to predict the preferred attribute of a buyer in response to the buyer’s purchase request. The algorithm for multiattribute dynamic pricing is based on FROODERUDWLYH¿OWHULQJ&) which enables a seller to predict a buyer’s pre- IHUUHGDWWULEXWH&ROODERUDWLYH¿OWHULQJDOJRULWKPV (Kleinberg & Sandler, 2003; Sarwar, Karypis, )LJXUH3ULFHYVWLPHSUR¿OH RI WKUHHVHOOHUV XVLQJWKH02DOJRULWKP 398 Dynamic Pricing for E-Commerce Konstan, & Reidl, 2001) collect potential buyers’ opinions or preferences on products of interest, and recommend possible products to new or returning buyers. A seller’s attribute-prediction algorithm for a potential buyer must adaptively respond to changes in the buyer’s preferences. The buyer attribute-prediction algorithm described below achieves this adaptive response by dynamically updating the seller’s model of the buyer’s attribute preferences. Buyer Attribute-Prediction Algorithm In the buyer attribute-prediction algorithm, a seller constructs one buyer cluster for each product at- tribute. Suppose the seller maintains C clusters. A buyer with preferred attribute a i is placed into cluster c i with probability w i,t during interval t. These probabilities are updated dynamically in response to the buyer’s accepting or rejecting offers made by the seller. When a buyer makes a purchase request, the prediction algorithm takes the history of w i,t – s and outputs the pre- dicted cluster (preferred attribute) for the buyer. Sophisticated, rather complex algorithms have been developed for assigning buyers to clusters, determining appropriate prices for buyers within clusters, and revising assignments and prices in response to decisions by buyers to purchase or not (Dasgupta & Hashimoto, 2004). FUTURE TRENDS 7KH FROODERUDWLYH¿OWHULQJDOJRULWKP GHVFULEHG above enables online sellers to determine a buyer’s preferences over multiple product attributes and WRXSGDWHWKHSRVWHGSURGXFWSULFHVHI¿FLHQWO\LQ a competitive market. More powerful learning techniques such as Q-learning (Mitchell, 1997) and multi-objective, evolutionary algorithms (Coello, Veldhuizen, & Lamont, 2002) offer PHFKDQLVPVWRHQDEOHVHOOHUVWRVHDUFKWKHSUR¿W ODQGVFDSH PRUH HI¿FLHQWO\ 7KHUH DUH YDULRXV trade-offs between the rapidity and accuracy of such learning algorithms. A naive but fast learn- ing algorithm might compare favorably against a complex and accurate but slow learning algorithm in a dynamic environment like a competitive online market. An interesting scenario arises when buyers’ purchase preferences are dependent on the prices being charged by sellers. In such a scenario, a seller can attempt to learn not only the temporally varying buyer purchase-preference distribution, but also the variation in that distribution. Proba- bilistic algorithms such as hidden Markov models and moving-target functions that estimate the de- pendence between temporally varying functions might be applied in such an environment. CONCLUSION We have described different algorithms that an online seller can use for the dynamic pricing of products in a posted-price market, where the seller announces the price of a product on its Web site. We have also described techniques that an online seller can use to determine the price of a product, including multiattribute dynamic pricing and adaptive response, in which the seller’s model of the buyers’ attribute preferences is updated dynamically. REFERENCES Brooks, C., Gazzale, R., MacKie-Mason, J., & Durfee, E. (2003). Improving learning per- formance by applying economic knowledge. Proceedings of the Third ACM Conference on Electronic Commerce (pp. 252-253). Brown, J., & Goolsbee, A. (2000). Does the In- ternet make markets more competitive (NBER Working Paper No. 7996)? National Bureau of Economic Research, Massachusetts. 399 Dynamic Pricing for E-Commerce Chavez, A., & Maes, P. (1996). Kasbah: An agent marketplace for buying and selling goods. Pro- ceedings of the First International Conference on the Practical Application of Intelligent Agents and Multi-Agent Technology (pp. 75-90). Clark, D. (2000). Shopbots become agents for business change. IEEE Computer, 33, 18-21. Coello, C., Veldhuizen, D., & Lamont, G. (2002). Evolutionary algorithms for solving multi-ob- jective problems. New York: Kluwer Academic Publishers. Dasgupta, P., & Hashimoto, Y. (2004). Multi-at- tribute dynamic pricing for online markets using intelligent agents. Proceedings of the Third Au- tonomous Agents and Multi-Agents Conference (pp. 277-284). Dasgupta, P., & Melliar-Smith, P. M. (2003). Dy- QDPLFFRQVXPHUSUR¿OLQJDQGWLHUHGSULFLQJXVLQJ software agents. Journal of Electronic Commerce Research, 3(3-4), 277-296. Kephart, J., Hanson, J., & Greenwald, A. (2000). Dynamic pricing by software agents. Computer Networks, 32(6), 731-752. Kleinberg, J., & Sandler, M. (2003). Convergent D O J R U L W K P V I R U F RO O D E R U D W L Y H ¿ O W H U L Q J  Proceedings of the Fourth ACM Conference on E-Commerce (pp. 1-10). Mitchell, T. (1997). Machine learning. McGraw Hill. Nelder, J., & Mead, R. (1965). A simplex method for function minimization. Computer Journal, 7, 308-313. Sandholm, T., Suri, S., Gilpin, A., & Levine, D. (2002). Winner determination in combinatorial auction generalizations. Proceedings of the First International Conference on Autonomous Agents and Multi-Agent Systems (pp. 69-76). Sarwar, B., Karypis, G., Konstan, J., & Reidl, J. ,WHPEDVHGFROODERUDWLYH¿OWHULQJUHFRP- mendation algorithms. Proceedings of the Tenth International WWW Conference (pp. 285-295). KEY TERMS Auction: A type of market in which sellers post an initial price for the item being offered and a deadline by which the item needs to be sold. Buyers make bids on the offered item. The auc- tion mechanism determines the dynamics of the prices bid by the buyers, the winner-determination strategy, and the bid-disclosure strategy. Common auction mechanisms include the English auction, Dutch auction, and Vickrey auction. Buyer’s Reservation Price: The reservation price of an item for a buyer is the maximum unit price that the buyer is willing to pay for an item. The buyer’s reservation price is typically drawn from a uniform or normal distribution. Collaborative Filtering: A technique that is used to collect user opinions or preferences for items of interest. A CF algorithm employs a cor- relation method to predict and recommend items to new or returning users based on the similarity of their interests with those of other users. E-Commerce: Consists of techniques and algorithms used to conduct business over the Internet. Trading processes such as supply-chain management, strategic purchase planning, and market mechanisms for trading commodities online are implemented using e-commerce. Intelligent Agent: Performs tasks that are given to it without continuous supervision. An agent can perceive changes in its environment and can perform actions to accomplish its tasks. 400 Dynamic Pricing for E-Commerce Marketplace: A type of a market that corre- sponds to a central location that enables buyers and sellers to rendezvous. A marketplace is typically implemented as a blackboard where sellers post information about items being offered. Buyers make offers to sellers, and sellers respond with counteroffers. Pricebot: An intelligent agent that is used by DQRQOLQHVHOOHUWRGHWHUPLQHDSUR¿WPD[LPL]LQJ price for a product that it sells. A pricebot encap- sulates the pricing algorithm used by an online seller and enables a seller to maintain an edge over its competitors in a dynamically changing market scenario. Seller’s Production Cost: The production cost of an item for a seller includes the manufac- turing and procurement costs for the item, and corresponds to the minimum price that the seller can charge for the item. Shopbot: An intelligent agent that enables online buyers to determine and compare prices and other attributes of products from different online sellers. This work was previously published in Encyclopedia of E-Commerce, E-Government, and Mobile Commerce, edited by M. Khosrow-Pour, pp. 247-252, copyright 2006 by Information Science Reference (an imprint of IGI Global). 401 Copyright © 2009, IGI Global, distributing in print or electronic forms without written permission of IGI Global is prohibited. Chapter 2.6 Planning and Designing an Enterprise Wide Database Systems for E-Business Alexander Y. Yap Elon University, USA ABSTRACT The planning and development of an enterprise- wide electronic database system for e-business usually calls for the re-engineering of information processes coupled with a push toward data content standardization across the entire organization. In this chapter, the case study involves a multi- national conglomerate that is in the process of integrating and Web-enabling their enterprise database systems. The objective of the system was to help engineers sift through millions of components offered by various suppliers and component manufacturers, where the end-result ZDVWRLPSURYHGWKHLQWHJUDWLRQDQGHI¿FLHQF\ of the product development, engineering design, e-sourcing, and e-procurement processes. This research is a qualitative action research study on how different organizational, social, political, and WHFKQLFDOIRUFHVLQÀXHQFHGWKHVRFLDOFRQVWUXF- tion of an enterprise-wide information system. Understanding the dynamics and power of these socio-technical forces in shaping the development environment and change process of enterprise systems is the focal point of this chapter’s dis- cussion. INTRODUCTION In mid-2001, Invensys, a multi-billion dollar multi-national corporation initiated a project to implement an enterprise-wide electronic database system accessible via the Web. The envisioned database was to form part of the corporation’s growing e-business system. This database was geared toward helping engineers sift through millions of electrical and mechanical components offered by various suppliers and component vendors. The database system was envisioned to be integrated with their e-procurement sys- tem, product data management systems (PDM), 402 Planning and Designing an Enterprise Wide Database Systems for E-Business enterprise resource planning (ERP) systems, and computer-aided design and manufacturing (CAD/CAM) systems. The objectives of initiating this enterprise- ZLGH V\VWHP ZHUH WR VLJQL¿FDQWO\LPSURYH the product development process by providing Invensys engineers a better and faster means of identifying/choosing product components and cutting product development cost (by lowering product development errors caused by sub-stan- dard components); (2) to improve e-sourcing (or the online search process for the right suppliers) by having access to a much wider range of sup- plier catalogs internationally and locally and be able to compare/analyze them; and (3) to improve the e-procurement process and lower procure- ment cost. 7KHFRQWULEXWLRQDQGVLJQL¿FDQFHRIWKLV research is to provide meaningful insights into different socio-technical milieus and their pivotal LQÀXHQFHLQWKHVKDSLQJRIDQHZHQWHUSULVHZLGH system. Although systems requirements and functionality are delineated and envisioned at the onset of system planning, the resulting system is often implemented and developed differently from what was initially planned due to the underlying V R F L R  W H F K QLFD O U H D O LW L H V W K D W V X U I D F H D Q G L Q ÀX H Q F H  the systems planning as more stakeholders and systems users become involve in it. Mitigating socio-technical factors ultimately determine the path of systems development and adoption. Therefore, it is important that more studies and research are conducted to shed light on how these underlying forces come into play when shaping enterprise systems. ,QYHQV\VLVDGLYHUVL¿HGFRQJORPHUDWHWKDW manufactures and provides various products and services. In the United States, Invensys is a leading global provider of heating systems, air conditioning, building systems, and commercial refrigeration. In Europe, Invensys acquired the AVP group of companies which specializes in engineering processes (such as brewery and dairy V\VWHPV'XHWRWKHFRQJORPHUDWH¶VGLYHUVL¿HG global business, it was of great interest to pursue this study to see how various Invensys subsidiar- LHVZRUOGZLGHFRXOGEHQH¿WIURPRUFKDQJHZLWK a new enterprise-wide system. What was also interesting from a socio-technical research point of view was that at the time of this project’s imple- mentation (2001-2002), Invensys Corporation just acquired Baan, a leading ERP solutions provider (Baan was re-acquired by another company in 2003). This created an underlying situation that Baan, being an enterprise solutions provider, ZRXOGKDYHFRQVLGHUDEOHLQÀXHQFHLQWKHGLUHFWLRQ of this enterprise-wide project. RESEARCH INTEREST AND APPROACH The objective of this research is to determine the key factors that affect the planning and develop- ment of enterprise-wide systems. We were hired as information systems consultants tasked to plan and design the implementation of this project from 2001-2002. When hired, we realized it was a great research opportunity, because it would allow us to experience how an enterprise-wide systems development goes through the intricacies of a con- glomerate environment. The retrospective value of this research was discussed and agreed upon with Invensys Technology vice president, Tim Matt. This research is a result of our documented analyses and insights as to how we strategically planned and tactically developed the system on a day-to-day basis considering the organizational, social, political, and environmental forces that ultimately shaped the systems design. We want to continue the discourse of Kim, Lee, and Go- sain (2005) and Gosain (2004) who claimed that enterprise information systems are subjected to institutional forces and processes, and Soh and .LHQZKRGLVFXVVHGWKHQHHGWR¿WFXOWXUH with enterprise systems solutions or there will be gaps that could lead to mismatch between the solution and the enterprise’s needs. 403 Planning and Designing an Enterprise Wide Database Systems for E-Business Since the researchers were involved in the proj- HFWWKLVFKDSWHULVFOHDUO\FDWHJRUL]HGDV³action research”. While quantitative methods are good for some type of research, we argue that qualita- tive research is the more appropriate approach to determining the casual effects of socio-technical and organizational factors in shaping the develop- ment of enterprise-wide systems. 1XPHURXVVWXGLHVLQWKH¿HOGRILQIRUPDWLRQ V\VWHPVKDYHDFNQRZOHGJHGWKDW³DFWLRQUHVHDUFK´ is a well-suited method for dissecting the complex social dimensions of IS planning and development. Previous action research methodology studies by Baskerville (1999), Wood-Harper (1985), and Hult and Lennung (1980) discussed action research as appropriate for understanding the social set- ting of the information systems environment. To quote Baskerville on his adoption of Hull and /HQQXQJ¶VGH¿QLWLRQRIDFWLRQUHVHDUFKKHFLWHG four major characteristics of the action research methodology: 1. Action research aims at an increased under - standing of an immediate social situation, with emphasis on the complex and multi- variate nature of this social setting in the IS domain. 2. Action research simultaneously assists in practical problem solving and expands sci- HQWL¿FNQRZOHGJH7KLV JRDOH[WHQGVLQWR two important process characteristics: First, there are highly interpretive assumptions being made about observation; second, the researcher intervenes in the problem set- ting. 3. Action research is performed collabora - tively and enhances the competencies of the respective actors. A process of partici- patory observation is implied by this goal. Enhanced competencies (an inevitable result of collaboration) is relative to the previous competencies of the researchers and subjects, and the degree to which this is a goal, and its balance between the actors, will depend upon the setting. 4. Action research is primarily applicable for the understanding of change processes in social systems. Although action research can be viewed as subjective, the intention of this research is to learn from experience, so we will not attempt to cloud it with our biases as we, too, want to fully learn from facts and events that transpired. We accepted the consulting role based on the op- portunity to learn more about enterprise-wide systems planning and development. Baskerville VWDWHGWKDW³FRQVXOWDQWVDUHXVXDOO\SDLG to dictate experienced, reliable solutions based on their independent review. Action researchers act RXWVFLHQWL¿FLQWHUHVWWRKHOSWKHRUJDQL]DWLRQLWVHOI to learn by formulating a series of experimental solutions based on an evolving, untested theory.” As academic researchers taking on the consultant’s role, we fully concur with this statement. THEORETICAL FRAMEWORK: SOCIO-TECHNICAL FACTORS SHAPING ENTERPRISE-WIDE SYSTEMS There are internal and external factors affecting the way enterprise-wide information systems (EIS) are planned and developed. Figure 1 maps WKHVHLQÀXHQFLQJIDFWRUV$OWKRXJKVRFLRFXOWXUDO forces (Bijker, Hughes, & Pinch, 1987), political forces (Robey, 1995), and business process (Ham- mer & Champy, 1993) are familiar factors that we occasionally come across as shaping information systems, it is also abstract as to how these factors help shape enterprise-wide systems. Our objective LV WR LQVWDQWLDWH LWV LQÀXHQFH LQ PRUH FRQFUHWH terms with qualitative data. . S t UHSUHVHQWVWKHSUR¿WPDGHEWKHVHOOHU during interval t, and G t represents the amplitude of the price change and is drawn randomly from the uniform distribution U[l, u], where l > 0 and u > 0. In the DF algorithm,. wider range of sup- plier catalogs internationally and locally and be able to compare/analyze them; and (3) to improve the e-procurement process and lower procure- ment cost. 7KHFRQWULEXWLRQDQGVLJQL¿FDQFHRIWKLV research. forces and processes, and Soh and .LHQZKRGLVFXVVHGWKHQHHGWR¿WFXOWXUH with enterprise systems solutions or there will be gaps that could lead to mismatch between the solution and