A Category Role Aded Market Segmentaton Approach to Convenence Store Category Management Yongje Ye 1, Shuhua Han Abstract Ths paper presents an nnovatve market segmentaton model whch s drven by category-role, for the frst tme, to support category management n chan CVSs. The usefulness and applcablty of ths study s llustrated by means of an emprcal study. The derved results are dscussed and compared wth the exstng works. Keywords: Category Management, Market Segmentaton, Chan Convenence Stores Management Introducton In recent years, category management (CM) (Basuroy, Mantrala, & Walters, 2001; Morgan, Kaleka, & Gooner, 2007)gradually ganed popularty n retalers to ncrease ther core compettveness, maxmze the profts and ensure long-term healthy customer relatonshp. CM (Dussart, 1998; Heller, 2012; Zufryden, 1987) s a busness fundamental whch uncovers a lot of untapped potental to explore. It ams to analyze consumer purchasng behavor and stock the products consumers are most lkely to buy. More specfcally, the defned categores can be used to target the consumer groups to gan a better understandng of ther needs. Although t attracts ncreasng attenton and some ntal results have been obtaned (Dussart, 1998; Kamakura & Kang, 2007; Trved, 2011), applyng conventonal CM methods to chan Convenence Stores (CVS) management t s stll a challengng task due to ts dstngushed features. Frst, CVSs are often dstrbuted n a varety of areas and each store has mpulsve consumers. Therefore, dfferent CVSs may have to provde dfferent products to meet the dverse requrements of ther consumers. Second, due to the moblty of consumers, t s dffcult to employ common CM attrbutes (e.g. locaton, populaton, age) to determne the categores. Thrd, CVSs are often small and have very lmted storage space, such that t cannot provde large number of product categores. Ths leads supplers to nvest most of ther tme and effort on ther large retalers, but not small chans CVSs. To overcome these dffcultes, ths paper proposes an nnovatve market segmentaton model, whch for the frst tme employs a category role (CR) to support CM n chan CVSs. Ths new method takes three CR dmensons (mportance to retalers, 1 Correspondng author. Emal: fjyeyongje@126.com. 1
consumers, and marketplace) nto account to segment the market. Intally, the hstorcal transacton data provdes a full vew of the market. Such data are then used to derve a global category ndex (CIX) for each store. CVSs whch share smlar CIXs are clustered nto the same group by usng a new smlarty measure, named HCsm() and an mproved weghted fuzzy K-mean clusterng algorthm (named WFKM). Based on the obtaned clusterng results, the retaler can desgn vared CM and marketng strateges for dfferent CVSs clusters. The remander of ths paper s organzed as follows: In Secton 2, a novel method, the CR-based market segmentaton model, s proposed to cluster the CVSs. In addton, an mproved clusterng algorthm whch employs HCsm() s also ntroduced n ths secton. The applcablty and utlty of the proposed method s demonstrated n Secton 3 va an emprcal study. The fnal secton concludes ths paper and ponts out further work drectons. CR-based market segmentaton approach Category role A key tenet of CM s that the retaler should decde on the role each category plays n the overall portfolo. Accordng to cross-category quanttatve analyss (Dussart, 1998; Kamakura & Kang, 2007), the typcal four CRs are: Destnaton category: Make the store the prmary category provder. When consumers would lke to purchase products n a certan category, the store s ther man choce. It takes 5% - 10% of categores n the store. Routne category: Make the store one of the preferred category provders. Ths category generates sold profts for retalers, whle meetng the customers' daly varous needs. It takes 50% - 70% of categores n the store. Occasonal/seasonal category: Make the store a major category provder when consumers would lke to buy a gven occasonal/seasonal product. It takes 10% - 15% of categores n the store. Convenence category: Ths category helps the store to provde convenence to consumers. It takes 10% - 15% of categores n the store. CRs are essental to mantan a consstent strategc and tactc plan, whch provdes cohesve market ntents across prce, promoton and assortment. It s ponted out that the same category may act as a dfferent CR n dfferent stores(trved, 2011), so that customzed category strategy should be mplemented n dfferent market segments. Category ndex (CIX) Cross-category quanttatve analyss (Basuroy, et al., 2001; Dussart, 1998; Kamakura & Kang, 2007) has been wdely used n the assgnment of CRs. In ths method, three dmensons (as shown n Table 1) should be taken nto consderaton and they are descrbed as below: Three dmensons The mportance to consumers Table 1: Three dmensons of CR Measures Annual expendture; purchase frequency; purchase volume 2
The mportance to retalers The mportance to marketplace Sales volume; sales revenue; sales gross prot; effcency plateau Average growth rate; purchase trends; market share changes trends Due to the avalablty of dmenson measures, fve segmentaton attrbutes are selected to aggregate a new category ndex (CIX): CIX F N S R G (1) F N S R G Where CIX s the ndex value of category, and F, N, S, R, and G represent the average sales frequency, average sales volume, average sales revenue, average gross proft and average growth rate of category, respectvely. F, N, S, R and G represent the correspondng attrbute weghts and F + N + S + R + G = 1. A new smlarty measure (HCsm()) Based on the features of CVSs category data, coupled wth both the smlarty between category structures and the smlarty n data space, a revsed category smlarty measure, HCsm(), s defned as: HCsm( s, s ) j k L p jk ( c ) 1 cx cx 1 j k L (2) Where L C j C k s the total number of categores whch store j and k sell, the pjk ( c ) represents the rato of common sub-categores or products n category c that shared by store j and k, c x j s the CIX values of category n store j. Ths smlarty measure has the followng propertes: The greater smlarty value ndcates the two CVSs are more smlar to each other. 0 HCsm( s, s ) 1, f and only f two CVSs are dentcal, j k then HCsm( s j, sk) 1. In contrast, when HCsm(s j,sk ) 0, store j and k are totally dfferent. Ths measure s symmetrc,.e. HCsm( s, s ) HCsm( s, s ). j k k j Ths dstance functon dffers from the tradtonal dstance functons. Its value reflects the smlar dmensons (categores, n ths case) between any two stores and s less affected by dssmlar categores, thus, the more categores shared by two stores, the more smlar the two stores are, and vce versa. It matches the natural human-beng percepton well. Clusterng Clusterng s a commonly used technque n market segmentaton (Lu, Kang, & Brusco, 2012; Shn & Sohn, 2004). In ths work, an mproved weghted fuzzy K-means clusterng algorthm (WFKM) whch employs prevously proposed HCsm() smlarty measure s proposed heren to cluster CVSs. The new CIX, whch consders consumers, retalers, and marketplace, s also employed to help cluster CVSs (as shown n Fgure1). 3
Ultmately, by analyzng the clusterng results help to gan a better understandng of market segments and consumers. It also provdes sold evdence for the mplementaton of CM and market strategy n CVSs. Fgure 1: CR-based market segmentaton model Data and emprcal results An emprcal study s conducted to verfy the proposed model n ths secton. A dataset from a gas staton convenence store n Chna s used. It contans transacton records that are collected from 82 petrol chan CVSs between January 2009 and June 2009 n Guangdong provnce, Chna. The dataset contans 2 category levels: 21 major categores and 95 sub-categores, and 3,456 products. Data preprocessng Intally, nosy, error and mssng data was removed from the dataset. In ths case, 7 out of 82 CVSs (whch only contaned transacton data n January and February) were fltered out. Meanwhle, the Fast food category that does not contan transacton data was also removed. Thus, a total of 20 major categores and 75 CVSs reman. The fve segmentaton attrbutes used n the proposed model (.e. F, N, S, R and G) are derved on a monthly bass. To start wth, normalzaton s performed on these attrbutes by usng the Mn-Max method, n whch the value of each attrbute s mapped onto a range of [0, 1]. The clusterng analyss of CVSs market segmentaton In order to avod bases, the entropy method s selected to determne the weghts of the fve attrbutes, and the obtaned results are shown n Table 2. Table 2: Attrbute weghts Attrbutes F N S R G 4
Weghts 0.2022 0.1834 0.1870 0.2137 0.2130 Based on Table 2 and collected transacton data, the CIXs (whch ndcates the global contrbuton of a gven CVS) of dfferent CVSs can be calculated, by the use of Equaton (1). The results are lsted n Table 3. Table 3: CIXs of dfferent CVSs Stores # category1 category2 category3 Category20 1 0.3895 0.5226 0.4591 0.6574 2 0.3129 0.5377 0.4729 0.7216 3 0.3447 0.5383 0.4844 0.7256 75 0.3059 0.3658 0.3356 0.4094 For clusterng analyss, t s crucal to select the number of clusters, K. In the lterature, three evaluatons, namely partton coeffcent(bezdek, 1973), XB coeffcent (Xe & Ben, 1991) and fuzzy entropy(kosko, 1986), are often employed to assess the clusterng performance. For the partton coeffcent, the greater ndex value ndcates better performance, whlst for the XB coeffcent and fuzzy entropy, the smaller ndex value ndcates better performance. In ths work, based on the collected dataset, these three ndexes are calculated by usng dfferent K values, respectvely. The obtaned results are depcted n Fgure 2. Wth the ncrease of K, the ndex value of partton coeffcent gradually decreases; the ndex value of fuzzy entropy ncreases at frst, reaches the peak ndex value when K = 3, and then slowly decreases; the ndex value of the XB coeffcent has a downward trend when K s n [2, 4], and then tends to become stable when K s n [4, 10]. Guded by Fgure 2, ths work chooses K as 4 n an effort to acheve optmal clusterng results. Fgure 2: The changes of the ndcators of effectveness 5
Results and dscusson Based on Table 3 and the selected K value, we get the clusterng results. The CVSs are grouped nto four market segments (Cluster1, Cluster2, Cluster3 and Cluster4). The detaled results are lsted n Table 4. In terms of CIX, Cluster2 (0.4579) > Cluster1 (0.3780) > Cluster3 (0.3514) > Cluster4 (0.3364). Ths ndcates that CVSs n Cluster2 outperform other CVSs n the market, whle CVSs n Cluster4 performs the worst n the market. In addton, Cluster2 only conssts of 8 CVSs, whch s much less than other clusters. Therefore, further nvestgatons on these CVSs n Cluster2 are necessary to help mprove other CVSs performance. For category roles, the destnaton category acheves a hgh level of consstency. Except Cluster2, the destnaton category all conssts of Drnks and Sweets categores. Snce these categores contrbute most towards the store performance. It s natural to pay more attenton on ther sub-categores and provde more varety of products for consumers to choose. Also, frequent and dverse promoton offers on these categores are necessary to gan more products for the CVSs. For Cluster2, Tobacco appears n the destnaton category, ths s qute dfferent from other market segments. One possble reason may be that ther consumers nclude more smokers than other clusters. Further nvestgatons on ther smokers-consumers may help ths segment to develop valuable market strateges. The routne category manly contans ready-to-eat categores (e.g. Sweets, Crsps/snacks, Bscuts, Mlk, Hot meals, etc), ths ndcates the clusterng results s meanngful, snce the derved CR well reflects the realty. Such categores are conssts of routne products n daly lfe. Due to the very lmted storage space n chan CVSs, the ncluded products n routne categores should be carefully selected. Performance analyss on sngle product should be conducted frequently, those products perform worst should be removed from the shelf. In addton, the products n routne category are often provded n other compettors as well. Therefore, when dentfyng the prce of such products, the prces from other compettors need to taken nto consderaton. In the seasonal category, the ncluded categores are qute dverse; none of the market segments consst of the dentcal categores. Moreover, for these categores, the consumer demands are not stable and normally are short-term. Especally for seasonal categores, varous promoton offers can be provded accordng to festvals or seasons. For example, n Cluster3, Ice-cream appears n the Seasonal category; therefore, for those CVSs n Cluster3, t s a good dea to provde promotons on Ice cream n summer. The convenence category manly contans Newspaper/magazne and Ice-cream; ths once agan reveals the relablty of the proposed clusterng algorthm. Ths s because, t s well known that categores such as Newspaper/magazne and Ice-cream are low proft, but essental to consumers. The goal of these categores s to provde convenence to consumers, so that the consumers can buy all they need n the store n one go. Consumers tend to be less senstve to prce n ths category, and care more about the product really meets ther requrements. Therefore, promotons n convenence category may not be very attractve to consumers. 6
Type of market # of CVSs Average CIX Convenent category Seansoal category Rountne category Targeted category Cluster1 23 0.3780 Famly plannng, Ice cream Newspaper/magazne ce cream Pharmacy, Wne, Toy/gft, Crsps/snack, Bscut, Hot meal, Mlk, Tobacco, Engne ol, Household good, Statonery, Health & beauty, Bread, Frut, Automoble supples Drnks, Sweets Cluster2 8 0.4579 Table 4 The clusterng results Ice cream, Newspaper/magazne,Toy/gft, Wne, Famly plannng, Pharmacy, Sweets, Crsps/snacks, Bscut, Hot meal, Mlk, Household good, Health & beauty, Bread, Engne ol, Frut, Statonery, Automoble supples Drnks, Tobacco Cluster3 23 0.3514 Famly plannng,toy/ gft, Newspaper/magazne, Household good, Pharmacy, Ice cream Crsps/snacks, Hot meal, Bscut, Mlk, Engne ol, Statonery, Bread, Tobacco, Automoble supples, Health& beauty, Wne, Frut, Drnks, Sweets Cluster4 21 0.3364 Ice cream, Household good, Newspaper/magazne, Wne, Famly plannng, Toy/ gft, Crsps/snacks, Hot meal, Bscut, Statonery, Mlk, Engne ol,health & beauty, Bread, Automoble supples, Tobacco, Frut, Pharmacy Drnks, Sweets 7
Concluson In ths paper, we have proposed an nnovatve market segmentaton model for CVSs. Startng from the three dmensons of CR, each category n dfferent CVSs get a derved CIX (whch ndcates the global contrbuton of a gven category to CVS). In addton, by the jont use of an nnovatve smlarty measure (HCsm()) and an mproved weghted fuzzy K-means clusterng algorthm (WFKM), CVSs wth smlar CIXs are grouped nto the same market segment. The applcablty and utlty of the proposed clusterng model s demonstrated va an emprcal study on a CVSs dataset provded by a gas staton convenence store n Chna. By usng the CR model, the current retal market s dvded nto four clusters. And they show a hgh level of consstency n destnaton category, whle performance qute dfferences n other CRs. Moreover,the cluster results helps to defne approprate market strateges for dfferent categores n dfferent market segments. Although the proposed approach s promsng, much may be done through further research. Frstly, the applcablty of the proposed CR model can be further mproved, so that the model can be used n varous applcaton domans. Secondly, n order to better smulate the changes of consumers demand and purchasng behavor, tme seres analyss wll be employed to analyze hstorcal sales data. Acknowledgements Ths work s supported by the Natonal Nature Scence Foundaton of Chna (Grant No. 70971112). References Basuroy, S., Mantrala, M. K., & Walters, R. G. (2001). The mpact of category management on retaler prces and performance: Theory and evdence. The Journal of Marketng, 16-32. Bezdek, J. C. (1973). Cluster Valdty wth Fuzzy Sets. Journal of Cybernetcs, 3(3), 58-73. Dussart, C. (1998). Category management:: Strengths, lmts and developments. European Management Journal, 16(1), 50-62. Heller, A. (2012). Consumer-Centrc Category Management: How to Increase Profts by Managng Categores Based on Consumer Needs: Wley. Kamakura, W. A., & Kang, W. (2007). Chan-wde and store-level analyss for cross-category management. Journal of Retalng, 83(2), 159-170. Kosko, B. (1986). Fuzzy entropy and condtonng. Informaton Scences, 40(2), 165-174. Lu, Y., Kang, M., & Brusco, M. (2012). A unfed framework for market segmentaton and ts applcatons. Expert Systems wth Applcatons, 39(11), 10292-10302. Morgan, N. A., Kaleka, A., & Gooner, R. A. (2007). Focal suppler opportunsm n supermarket retaler category management. Journal of Operatons Management, 25(2), 512-527. Shn, H. W., & Sohn, S. Y. (2004). Segmentaton of stock tradng customers accordng to potental value. Expert Systems wth Applcatons, 27(1), 27-33. Trved, M. (2011). Regonal and Categorcal Patterns n Consumer Behavor: Revealng Trends. Journal of Retalng, 87(1), 18-30. Xe, X. l., & Ben, G. (1991). A valdty measure for fuzzy clusterng. Pattern Analyss and Machne Intellgence, IEEE Transactons on, 13(8), 841-847. Zufryden, F. S. (1987). New computatonal results wth a dynamc programmng approach for product selecton and supermarket shelf-space allocaton. Journal of the Operatonal Research Socety, 201-203. 8