Customer Segmentation In The Full Service Restaurant Business Marketing Essay

Customer Segmentation In The Full Service Restaurant Business Marketing Essay ABSTRACT. The purpose of this study was to find factors that affect customers intention to revisit restaurants using data mining. A total of 390 usable questionnaires were used in the data analysis. AnswerTree, a data mining software, was used as a major analytical method. AnswerTree enables a researcher to identify and target desirable customer groups and thus is suitable to use in identifying differences between one group with a revisit intention and one group with no revisit intention. Study results indicated that different component factors affected customers revisit intention between the two groups. Three factors (recommendation, living area, and number of dining occasions per week) most affected a groups revisit intention. On the other hand, three factors (recommendation, occupation, and most frequent dining destination) most affected the group with no revisit intention. Study results provide meaningful information for marketing strategies that can be successfully used by full- service restaurant operators. KEYWORDS. Customer segmentation, full-service restaurant, data mining, AnswerTree INTRODUCTION With increasing competition in the restaurant business, marketing strategies need to be more significant to ensure customer satisfaction (Kara et al., 1997; Kivela, 1997; Murphy et al., 1996). Since customer satisfaction affects businesses ability to sustain their status in a competitive restaurant market, many researchers have studied customer satisfaction (Almanza et al., 1994; Andaleeb Conway, 2006; Barsky Labagh, 1992; Domingo, 2002; James, 1995; Johns Tyas, 1996; Oh, 1999, 2000; Oliver, 1980, 1981). Previous researchers have queried restaurant customers to identify the reasons for their decisions to revisit and recommend restaurants to other potential visitors, thereby suggesting effective marketing strategies. Although there is no assurance that customers will make a return visit (Dube et al., 1994), restaurateurs assume that satisfied customers will return while customers who had a poor experience will not come again. Thus, understanding both satisfied and unsatisfied custo mers is an important process in making appropriate marketing strategies. In order to segment customers according to their attributes, knowledge of those attributes is an effective tool in developing appropriate marketing strategies (Bowen, 1998; Gregoire et al., 1995; Klosgen Zytkow, 2002; Reid, 1983; Richard Sundaram, 1994; Swinyard Struman, 1986; Woo, 1998; Yà ¼ksel Yà ¼ksel, 2002). Customer segmentation can be determined by customer attributes such as demographic and behavioral characteristics: buying patterns, attitude and use or response to a product (Kotler et al., 2005). In this study, the term customer attributes was defined as customers demographic and behavior patterns, attitude and use or response to a product via recommendation intention. To identify customer segmentation, data mining was applied in this study. Data mining is the process of finding or classifying trends and patterns in data to analyze customers past behaviors (Adriaans Zantinge, 1997; Chatfield, 1995; Fadairo Onyekelu-Eze, 2008; Kudyba Hoptroff, 2001; Lovell, 1983; Pyle, 1999; Thuraisingham, 1999). If restaurant managers clearly know their target customer attributes, they can develop a better service strategy or make up for their weak points. The aim of this research was to identify factors affecting customers intention to revisit a restaurant. Study results provide meaningful information to the process of developing marketing strategies that can be successfully used by full-service restaurant operators. Further, study results will inform restaurateurs about important and non-important factors in return intention and enable them to select those factors on which to focus on and improve. REVIEW OF THE LITERATURE Customer Segmentation Customer segmentation is a crucial part of todays overly competitive restaurant business. In the mid-1950s, Wendell R. Smith, an American marketer, first introduced the concept of customer segmentation (Nairn Berthon, 2003). Russell Haley then developed segmentation theory in 1968. Customer segmentation can be defined as a way to separate customers into groups for decision making purposes or to support effective management in acquiring or keeping customers (Bowen, 1998; Bahn Granzin, 1985; Chen et al., 2007; Yà ¼ksel Yà ¼ksel, 2002). Since customer segmentation can help restaurants increase revenue, a marketing strategy based on customer segmentation can be more powerful and effective (Auty, 1992; Bojanic Shea, 1997; Chen et al., 2006). Customer segmentation has encouraged researchers to take a closer look at customer segmentation as part of an effective marketing strategy. Lewis (1981) used discriminant analysis to identify the differences between goers and non-goers with food quality, menu variety, price, atmosphere, and convenience. Bahan and Granzin (1985) investigated four customer segments: health, gourmet, value, and unconcerned. They reported that each group had different preferences for service quality. Auty (1992) divided respondents into three customer groups (student, well-to-do middle-aged people, and older people) and examined restaurant image and atmosphere. Oh and Jeong (1996) revealed the characteristics of four customer segments: neat service seeker, convenience seeker, classic dinner seeker, and indifferent dinner seeker. Bojanic and Shea (1997) sought differences between downtown diners and suburban diners. Yà ¼ksel and Yà ¼ksel (2002) identified the attributes of five customer segments (value seekers, service seekers, adventurous food seekers, atmosphere seekers, and healthy food seekers) based on nine factors (service quality, product quality and hygiene, adventurous menu, price and value, atmosphere, healthy food, location and appearance, availability of nonsmoking area, and visibility of food preparation area). In this study, the researcher divided customers into two groups based on revisit intention in order to identify which factors most affect these two groups. The summary of the literature review is presented in Table 1. Table 1 about here Data Mining To divide respondents into two groups based on revisit intention, unlike previous studies, data mining was used to identify revisit intention in the restaurant business. Data mining was introduced in the late 1980s and developed in the 1990s. Its origin is in the fields of statistics and a specialized area within artificial intelligence (AI) that is part of computer science (Ogut et al., 2008; Roiger Geatz, 2003). According to Fadairo and Onyekelu-Eze (2008), data mining is a way to find information in a huge database. Such information can be used to identify relationships between variables. That is, data mining may be utilized in analyzing a specific data set with the intention of identifying patterns and establishing relationships; using data mining, it is possible to sort through massive volumes of data and discover new information or an analytic method for reasoning useful knowledge and predicting future trends (Bolshakova et al., 2005; Chatfield, 1995; Chen et al., 1998; Groth, 2000; Hand et al., 2001; Koyuncugil, 2004; Lovell, 1983; Westphal Blaxton, 1998). Data mining can perform two basic operations: predicting customer behaviors and identifying segmentation (Lampe Garcia, 2004; Wang et al., 2008). For that reason, many researchers have attempted to apply data mining in the business industry (Keating, 2008; Liu et al., 2008; Rygielski et al., 2002; Zambochovà ¡, 2008). Previous studies have stressed that companies could use data mining to identify customer dispositions or trends regarding the patronizing of certain companies. With this information, companies can focus their efforts on good customers from whom they would make the most profit. Further, all industries can take advantage of data mining in seeking to understand inconsistent segmentation of their target customers. In summary, data mining is a powerful technology that may be used in support of companies engaging in decision-making on issues such as customer attrition, customer retention, c ustomer segmentation and sales forecast (Ogut et al., 2008). CHAID in AnswerTree To apply data mining, the AnswerTree program was used in this study. AnswerTree, a data mining software, is a foreseeable model that shows results in a tree model (SPSS, 2009). Variables may be analyzed in the AnswerTree program in three ways: CHAID (chi-squared automatic interaction detector), CART (classification and regression trees), and QUEST (quick, unbiased, efficient, statistical tree). Basically, the CHAID method is a more comprehensive method and generates more accurate results when using categorical variables, while CART and QUEST are suitable when using continuous variables. Since categorical variables were used in this study, the CHAID method was applied here. The original CHAID method grew from a 1975 doctoral dissertation by Kass, who published a more accessible article four years later (Kass, 1980). Since the CHAID method allows marketers to identify segments in relation to a dependent variable having two or more categories based on the combination of independent variables (Chen, 2003), the CHAID method has popularly been applied in the consumer research field (Haughton Oulabi, 1997; Levin Zahavi, 2001). In the CHAID procedure, a dependent variable and key independent variables are initially chosen. According to chi-squared, the dependent variable can be divided by the levels of a certain independent variable that has the strongest association with the dependent variable. That is, the most important and related independent variable with a dependent variable becomes the first node. This analysis process occurs when one of three criteria are met, according to Berson et al. (2000): 1. The segment contains only one record. (There is no other question that you can ask to further refine a segment of just one.) 2. All the records in the segment have identical characteristics. (There is no reason to continue asking further questions because all the remaining records are the same.) 3. The improvement is not substantial enough to warrant asking the question (p. 162). All variables used in this study were categorical measurements with two or more categorical levels. The stopping rules for AnswerTree analyses were a maximum tree depth of 3, minimum number of cases of 25 for a given node, and significance level for splitting of 0.05. RESEARCH METHODOLOGY Data Collection and Questionnaire The data used for this study were collected in Miami via face-to-face interviews. The response rate for the face-to-face interview has revealed it to be the best method among various survey methods (The Monkey Team, 2008). Surveys were administered from May 1 to May 31, 2007. To increase result reliability, we selected respondents who had visited a full-service restaurant within the last one month. The selected full-service restaurants offered full table service and the average guest expenditure was at least $25 per person. Of the 414 questionnaires collected, 24 were incomplete and were eliminated. As a result, a total of 390 questionnaires were used in the data analysis. Since AnswerTree enables a researcher to identify and target desirable customer groups (SPSS, 2009), it is a suitable analysis method for identifying differences between groups. Further, AnswerTree is a more robust method than existing statistical methods in identifying segment characteristics (Byrd Gustke, 2006). There were two groups in the dependent variable: one group with the intent to revisit and one group with no intent to revisit. Questionnaire items for revisit intention were rated on seven-point Likert scale ranging from strongly disagree to strongly agree. To apply CHAID (Chi-square Automatic Interaction Detection) analysis, researchers converted the seven-point scale into categorical variable (agree, so-so, disagree). Even though respondents answers were a five or six on the seven-point scale, the information sufficiently ensured positive responses. Finally, those with five, six and seven points were converted into a positive group (agree). On the other hand, those with one, two and three points were converted into a negative group (disagree). Finally, four on the seven-point scale was converted into a so-so group. Among respondents (n=390), 83.33% (n=325) indicated that they were willing to revisit the restaurant; on the other hand, 11.54% (n=45) thought that they would not revisit the restaurant; and 5.13% (n=20) replied so-so, which means I dont know. Customer segmentation can be sorted by demographic and behavioral characteristics such as buying patterns, attitude and use or response to a product (Johns Pine, 2002; Kotler et al., 2005). Independent variables were composed of demographic profiles (gender, age, marital status, occupation, income, living area) and customer attributes in relation to buying behavior, a pattern of full-service restaurant use (how many dining occasions per week, how much spent at a restaurant, frequency of restaurant visits, with whom did they dine there). Finally, to ascertain a response to a product, recommendation intention was used in this study as an independent variable. The summary of data description is presented in Table 2. Table 2 about here RESULTS Figure 1 shows the general model for revisit intention. The general model makes it easy for readers to figure out how many nodes are included in the results and shows the entire model design. In this study, 13 nodes were used to explain the factors affecting a group who intended to revisit and a group with no intention to revisit. For the dependent variable, originally there were three groups: yes, so-so, and no. However, because of small sample size, the so-so group was not divided by descriptor. The dependent variable was divided by five descriptors: recommend intention, living area, how many times to dine per week, occupation, and when often dine. Figure 1 about here In terms of yes segmentation on revisit intention, in Figure 2, the first split was recommend: yes (à Ã¢â‚¬ ¡2=376.4356, d.f.=4; p=.000). In Node 3, 98.10% (n=309) of respondents thought that they had a revisit intention. Node 3 was divided into two groups: Node 7 and Node 8. The second split was based on the variable of Living area: South Florida, other Florida, and other U.S states (à Ã¢â‚¬ ¡2=68.5075, d.f.=2; p=.000). Node 7 was divided into two groups: Node 11 and Node 12. 98.69% (n=302) of respondents (Node 7) who lived in South Florida, other Florida, and other U.S states showed that they were willing to return to the restaurant. The last split was how many times dine per week: over 3 times (à Ã¢â‚¬ ¡2=19.7972, d.f.=1; p=.000). In Node 12, 100.00% (n=254) of respondents who visited the restaurant over 3 times per week indicated that they had a revisit intention. In summary, there were three descriptors as follows: recommend: yes, living: South Florida, other Florida, and other U.S states, and how many times to dine per week: over 3 times. Figure 2 about here Considering no segmentation on revisit intention, in Figure 3, the first split was recommend (no) (à Ã¢â‚¬ ¡2=376.4356, d.f.=4; p=.000). In Node 1, 81.13% (n=43) of respondents indicated that they would not visit the restaurant again. Node 1 then was divided into three groups: Node 4, Node 5, and Node 6. The second pruning tree was based on the occupation variable: office worker (professional, salesman, and self-employed) (à Ã¢â‚¬ ¡2=20.1046, d.f.=4; p=.000). In Node 4, 94.87% agreed that they were not willing to return to a restaurant. Node 4 was divided into two groups: Node 9 and Node 10. The last split was when often dine: lunch (à Ã¢â‚¬ ¡2=25.2973, d.f.=1; p=.000). In Node 9, all office workers (100.00%, n=36) who often visited the restaurant at lunchtime did not have a recommendation as well as revisit intention. In brief, three descriptors split the node: recommend: no, occupation: office worker (professional, salesman, and self-employed), and when often dine: lunch. Figure 3 about here Figure 4 presents summary statistics. The bar graph makes it easy for readers to understand which node most represents the dependent variable and provides the variation for each dependent variable. In this study, the dependent variable, revisit intention, was classified by a maximum tree depth of 3, minimum number of cases of 25 for a given node, and significance level for splitting of 0.05. The bar graph for AnswerTree showed that the particular nodes most often represent groups intention or non-intention to revisit, respectively (Node 12: agree group, Node 9: disagree group). Figure 4 about here Table 3 presents a gain chart of yes segments. A gain chart is a table that summarizes the entire model descriptively. In the gain chart, we can see the percentage representation of each node for the dependent variable. In the case of the yes segment, the root node was 83.33% (n=325). Node 12 was computed by taking 100.00% (Gain: % computed from Node: N divided by Resp: N) and then dividing it by 83.33% (root node). The result was 120.00%, the index score for Node 12. That is, Node 12 (recommend: yes, living area: south Florida, other Florida, and other U.S states, how many times dine per week: over 3 times) represents a root node about 1.2 times. Thus, in the case of yes segmentation, three variables (recommend, living, and how many times to dine per week) are important factors in dividing respondents into groups that answered yes regarding revisit intention. Table 3 about here Table 4 presents a gain chart for no segment. The root node was 11.54% (n=45). Node 9 was computed by taking 100.00% (Gain: %) and then dividing it by 11.54% (root node). The result was 866.66%, the index score for Node 9. That is, Node 9 (recommend: no, occupation: professional, salesman, and self-employed, when often dine: lunch) represents a root node about 8.6 times. Thus, in the case of no segment, three variables (recommend, occupation, when often dine) are important factors in dividing a group that answered no with respect to revisit intention. Table 4 about here Table 5 offers a risk chart indicating the preciseness of the classification. It resembles the percentage of classified respondents in the discriminant analysis. The risk estimate predicted the risk incurred due to misclassification of the respondents in the AnswerTree analysis. A lower risk estimate indicates a more precisely classified model. According to the results of the assessment of revisit intention, the risk estimate was 0.0615385. This means that the precision of classifying respondents in the AnswerTree analysis was 99.9384615%. That is, about 99.93% of the respondents were classified accurately on split nodes. Table 5 about here DISCUSSION AND IMPLICATION In the restaurant business, customer segmentation enables restaurant managers or marketers to develop effective marketing strategies. The purpose of this study was to identify factors that affect intention to revisit a full-service restaurant. To ascertain differences between group intent to revisit and group non-intent to revisit, data mining was used. There has been little use of data mining in the hospitality field. Because data mining is one way to create decision-making models that predict future behavior based on analyses of past activity, using collected data from segment targeting is the best way to create suitable marketing strategies (Lampe Garcia, 2004; Wang et al., 2008). Among the respondents (n=390), 83.33% (n=325) indicated that they were willing to visit the restaurant again while 11.54% (n=45) thought that they would not visit the restaurant again and 5.13% (n=20) were so-so. As mentioned earlier, due to the small sample size, the so-so group was not divided. The AnswerTree results revealed different component factors between the two groups. In the case of the yes segment, there were three descriptors: recommend: yes (à Ã¢â‚¬ ¡2=376.4356, d.f.=4; p=.000), living area: south Florida, other Florida, and other U.S states (à Ã¢â‚¬ ¡2=68.5075, d.f.=2; p=.000), and how many times dine per week: over 3 times (à Ã¢â‚¬ ¡2=19.7972, d.f.=1; p=.000). Analysis results revealed that the more customers like to dine out, the more they intend to revisit a restaurant. In other words, people who have a great interest in dining are more likely to be loyal customers. Moreover, as they intend to recommend the restaurant to other people, keeping such a customer means other potential customers could be positively affected. According to Sà ¶derlund (1998), Word-of-mouth is defined here as the extent to which a customer informs friends, relatives and colleagues about an event that has created a certain level of satisfaction (p. 172). As word-of-mouth can be a significant determinant of behavioral intentions, recommendation intention greatly affects a restaurant businesss sales (Babin et al., 2005; Edwards Meisleman, 2005; Mangold et al., 1999; Mattila, 2001; Sà ¶derlund, 1998). This studys results supported the finding that loyal customers are more likely to encourage other people to also have their exceptional experience, which is consistent with previous studies (e.g., Bowen, 1998). It is less costly to keep an existing loyal customer than to attract a new customer. Also, loyal customers return to make more repeat purchases at the restaurant. Therefore, identifying loyal customers is an important part of the restaurant businesses (Bowen Shoemaker, 1998; Fierman, 1994; Jang Mattila 2005; OBrien Jones, 1995; Orr, 1995; Schneider et al., 1998). From a managerial standpoint, the results of this study based on data mining help restaurateurs in identifying the characteristics of a loyal customer segment. In the case of the no segment , three descriptors split the node: recommend: no (à Ã¢â‚¬ ¡ 2=376.4356, d.f.=4; p=.000), occupation: office worker (professional, salesman, and self-employed) (à Ã¢â‚¬ ¡2=20.1046, d.f.=4; p=.000), and when often dine: lunch (à Ã¢â‚¬ ¡2=25.2973, d.f.=1; p=.000). The results of this analysis revealed that office workers who often visited a restaurant at lunch did not intend to revisit. In other words, office workers, salesmen, and the self-employed were less satisfied with lunch at the full-service restaurant. In general, few full-service restaurants focus on lunch for office workers, who usually eat lunch away from the office. Even though some lunch menus do focus on the office worker, offering this meal is usually expensive and time-consuming. However, as office workers tend to eat lunch in a brief span of time, they prefer to eat lunch at fast food restaurants rather than at a full-service restaurant. From a managerial viewpoint, if full-service restaurateurs could provide a lunch menu with low prices as well as quick service, they could obtain part of the office worker market currently going to fast food restaurants. The summary of segments is presented in Table 6. Insert Table 6 The purpose of customer segmentation is to target a certain type of customer when developing a marketing strategy. If a restaurant cannot develop correct and appropriate marketing strategies, they may not sustain their existence in this highly competitive business. As attributes have changed and become more complicated, customer segmentation is becoming more important in providing basic source material for marketing strategies. In order to respond quickly to changing customer attributes, restaurant marketers require rapid access to information on various customer attributes. In this context, restaurant marketers need to be able to identify customers past behaviors in order to predict future tendencies. This ability can be provided and maintained in the restaurant business by using data mining technologies. Data mining enables restaurant marketers to draw information more effectively from databases (Bolshakova et al., 2005; Chatfield, 1995; Chen et al., 1998; Groth, 2000; Hand et al., 2001; Koyuncugil, 2004; Lovell, 1983; Westphal Blaxton, 1998). Through the effective use of data mining, managers can more quickly analyze customer attribute in the restaurant business. This study of customer revisit intention occurred in the Miami area only. Thus, findings might not be generalized to other areas. Another limitation is that study results cannot be applied to all restaurant services, because our focus was on full-service restaurants only. Therefore, findings must be applied to other restaurants with due caution. Lastly, this study did not use a large enough sample size for data mining. Although we used 390 samples to identify and target customer groups, data mining is typically used with a large database. Thus, future research may use data mining with a larger sample size. Data mining was a useful method in predicting restaurant customers intentions to revisit. Unfortunately, very few studies have used this method. Thus, further research in other hospitality fields would benefit from data mining.