Characteristic Analysis of Each Store in Japanese Hair Salon

Abstract

There are 223,645 Hair salons in Japan in 2011. Those same about quadruple amount of Japanese convenience store. From these things, we can know that there are many Hair salon in Japan. Hair salon moved and there are close stores about 9,000 a year. However new open chain stores about 12,000 therefore Hair salon increasing about 3,000 stores a year of Hair salon in Japan. Hair salon were group at Kanto region and Kinki region because they are high population density in Japan. The struggle for existence Hair salon are store excess also a lot of hair salon are small scale and they are micro enterprises. In that sense, Hair salon are faces severe competition. Customer’s hair salon usages frequency is woman 4.5 times a year, men 5.38 times a year. The amount of usage per one times is women 6429 yen, men 4067 yen. Men are more frequently used, and the usage amount is increasing. However, the usage women’s rate frequency of Hair salon more than men’s rate frequency of hair salon. We use the data is all over Japan of a certain hair salon chain stores of this study. This was provided by Joint Association Study group of Managements Science (JASMAC) 2017 Data Analysis Competition. According to basic statistics, there are many customers with one visit to the store in this hair salon thus high customer rates. A certain hair salon have many people who are 30 to 60 years old. One of 12 stores are a male salon. We used Quantification category 3 that infer the characteristics of customers also, we predict store characteristics from the characteristics of customers. However, it was mixed Mathematical data and Qualitative data thus we had to unify the scale. Therefore, we converted Mathematical data and Qualitative data. As a result, we used Quantification category 3. We Interpreted compound variable answer1 is “Neighboring a working woman and office worker who want to quietness”. Answer2 is a customer who emphasize of high temperature and high-quality care. Cluster analysis has Hierarchical approach and Nonhierarchical approach. Nonhierarchical approach be able used for small data on the other hands Hierarchical approach can be used to big data. Therefor we adopted Hierarchical approach. We needed to decide on a group when we used Hierarchical approach. However, it was objectively lacking, and it had disadvantages of reduced reliability if you have decided the number of groups before this analysis. Therefore, we did cluster analysis in several times in addition we decided the number of groups by squared residual sum of squares in cluster. We created an elbow diagram. The difference in the residual sum of squares within the group is It was shrinking sharply smaller than Groups 4 to 5 thus we decided the number of groups to 5 in addition we did k-means method. When initial value of each result, big differences occur in size of group and convergence value. K-means method needs the best solution in multiple analyzes. In this research, we had the purpose to stores characteristic. We developed one-way analysis of variance and multiple comparison by compound variable, answer1 and answer2. We had the purpose to develop that analysis. The purpose was if we did not have significant difference, we would not classify every characteristic. One-way analysis of variance used compound variable, answer1 and answer2. Therefore, we used nonparametric method because those were not normal distribution. Nonparametric method has multiple test method. We used Games-Howell method in this research. Games-Howell method was matching in this research because it did not assumption homoscedastic. We used result hierarchical approach cluster analysis from First time to fourth time. We developed one-way analysis of variance and multiple comparison. The result, we adopted 2’nd time because it is classified definite characteristic. In addition, first time, 3’rd time and fourth time had significant difference one-way analysis of variance. However, there were combination in multiple comparison that did not have significant difference. We developed one-way analysis of variance with result of 2’nd time and compound variable. If answer1 shows a large value to minus, we infer that the customer want glamorous and has enough time to coming from afar. If answer2 shows a large value to minus, we infer that customer important care for low temperature and low humidity. We founded by one-way analysis of variance that there was a difference each group. We divided for each store the group. Then we were find difference in store characteristics. In addition, we suggested optimal marketing strategy based on result of analysis against each store. We will increase the synthesis and make finer interpretations.

After that, we will make suggestions for winning potential customers.

1 Introduction

1.1 Background

Hair salon serves us such as perm, cut and make up.

Figure 1 shows the trends in the number of stores in the beauty industry by year. The number of hair salon is flat from 2000 to 2002 but since 2003, this number has continued to increase. Tokyo is the number with the largest number of hair salons. Next is Osaka prefecture, then Saitama prefecture. In addition, hair salon moved and there are close stores about 9,000 a year. However new open chain stores about 12,000 therefore Hair salon increasing about 3,000 stores a year of Hair salon in Japan. There are 223,645 Hair salons in Japan in 2011. Those same about quadruple amount of Japanese convenience store. From these things, we can know that there are many Hair salons in Japan.

Fig. 1.

The trends in the number of stores in the beauty industry [1]

Figure 2 shows the number of Japanese hair salons by a heat map. Hair salon were group at Kanto region and Kinki region because they are high population density in Japan.

Fig. 2.

Heat map of hair salon [1]

Fig. 3.

Service fee consumer price index [1]

The average number of employees per facility in the hair salon nationwide is 2.03. Figure 4 shows the percentage of employees. One person is the highest at 29.9%, accounting for about 30% of the total. Next, 2 people followed 21.0%, 5–9 people followed 14.6%. About 70% of the facilities account for less than 5 people. From these facts, it can be inferred that many hair salons are small-scale and microscopic facilities. Figure 3 shows the trend of service fee provided by hair salon indexed as 100 in 2005. Prices offered services are increasing little by little every year. Haircut is the highest price increase, but it is kept at 1 point. Therefore, competition of hair salon is felt.

Fig. 4.

Trends in services by category of beauty markets as a whole [2]

Figure 4 is a graph showing the changes in market size of beauty services as a whole from surgery for 2008 to 2012. The service scale is decreasing in any service. From this, it is expected that this trend will continue in the future.

Customer’s hair salon usages frequency is woman 4.5 times a year, men 5.38 times a year. Men are more frequently used than females. In addition, the amount of usage per one times is women 6429 yen, men 4067 yen.

1.2 Purpose

In recent years, in the retail industry, customer purchasing behavior is gathered by POS data combined with customer ID and POS system and membership number and they use it to manage the store. By utilizing POS data with customer ID, we were able to approach each customer, and we were able to establish a strong relationship with customers. The trend of maintaining or increasing existing customers by strengthening relationships is used in many retail industries. Nakahara and Morita [3], Okuno and Nakamura [4] and Hisamatsu et al. [5] are mentioned as prior studies conducting customer purchase analysis from POS data with customer ID and purchase history. However, in the previous research, the approach to customers is mainly, and there seems to be few approaches to shops selling. In addition, there are many papers on beauty care, but among them there are few papers using statistical methods. Therefore, this research aims to analyze the features of stores using statistical methods and to help them in their management.

2 Data

The data we use is data from the hair salon chain store in Japan. We show below the overview. This data is targeted at 12 hair salons in Kanto area.

2.1 Data Overview

• Provide: Joint Association Study group of Managements Science (JASMAC)

  • 2017 Data Analysis Competition

• Period: July 1, 2015–Jun 30, 2016 (For 2 years).

2.2 Basic Aggregate

2.2.1 Customers Master Data

There are more female customers than men in certain hair salon (Fig. 5).

Fig. 5.

Percentage of gender

Fig. 6.

Heat map of customer’s address

Customers live in various parts of Japan (Fig. 7).

Fig. 7.

Percentage by birth year

They do not change the address after moving (Fig. 6).

The 2030’s and the 2080’s were clearly abnormal.

There were 7915 customers who were no answer.

Most often born in 1980.

“Store I” has a high ratio of men because this store is for men only (Fig. 8).

Fig. 8.

Gender ratio by store

2.2.2 Accounting Master Data

Before we do basic aggregate of accounting master data, we do an error handling.

• Total of accounting master data: 166,922

  • Free customer: 2,425 (1.5%)

  • Return: 5,757 (3.4%)

  • Sale: 5,148 (3.1%).

These three were abnormal values. So, the total number is 153592 after error handling (Figs. 9, 10 and 11).

Fig. 9.

Nominate ratio (Most customers choose the Hair dresser.)

Fig. 10.

Percentage of payment amount (People with one payment 5,400 JPY–8,640 JPY are the most frequent. The next highest percentage is 10,800 JPY–17,280 JPY.)

Fig. 11.

Number of visitors and sales by store

Most of the customers are not using the point. We thought it would be difficult to save points. A simple calculation means that 1% to 2% of the amount is targeted. Therefore, it is estimated that customers using 1,000 points are excellent customers.

B is the highest in both the number of guests and sales. A, B, G, L the ratio of the number of visitors is below the ratio of sales. The ratio of visitors to C, D, E, F, H, I, J exceeds the ratio of sales. Based on the above, it can be estimated that A, B, G, j is a high percentage of sales per visit.

There are a lot of people who visit only once. This can be said that the rate of churn is high (Fig. 12). The most used is 4:00 pm. Also, the opening was estimated at about 10:00 and closing at around 10 pm (Fig. 13).

Fig. 12.

The number of visits

Fig. 13.

Number of accounting by time

2.2.3 Accounting Details Master Data Sample

Before we do basic aggregate of accounting details master data, we do an error handling.

• Total of accounting master data: 409,347

  • Return: 16,043(3.9%)

  • Sale: 14,846 (3.6%).

These two were abnormal values.

So, the total number is 378,458 after error handling.

2.2.4 Product Master

There are many products of about 3,000 JPY to 5,000 JPY (Figs. 14 and 15).

Fig. 14.

Price setting

Fig. 15.

Treatment time

I can see that there are a lot of treatments that take 120 min.

3 Analyze

3.1 Analysis Method

We used Quantification category 3 that infer the characteristics of customers also, we predict store characteristics from the characteristics of customers. However, it was mixed Mathematical data and Qualitative data thus we had to unify the scale. Therefore, we converted Mathematical data and Qualitative data. As a result, we used Quantification category 3.

For the Quantification category 3, we aimed to conduct statistical analysis on the Internet. We used statistical analysis software provided by Shigenobu Aoki.

3.1.1 Analysis Target

The number of customers to be analyzed is 31,776 out of 31,862 people who are registering customers. 86 people who do not have purchasing behavior between January 2015 and February 2016 (2 years), which is the analysis period, were excluded from the analysis as the reason for the decrease in the number of people. The reason for narrowing down the analysis target to the customers who had the purchase history is that there is data indicating the characteristics of the customer with the purchase history customer. On the other hand, for customers without purchase history, there are few data showing characteristic. In addition, because it can be considered that the ID of the free customer has multiple customer information, it was necessary to think separately from other customer IDs. Therefore, it is excluded in this research.

3.1.2 Result of Analysis

Table 1 shows the Quantification category 3 using 29 variables.

Table 1. Result of quantification category 3 ①

From Table 1, the contribution ratio of the components is low, and the eigenvalues are also low. Therefore, from this fact, it can be seen that useful results are not obtained with 29 variables.

We analyzed again except for those with low commonality. We used 9 variables used for analysis, Kanto, Spring use, Summer use, Autumn use, Winter use, permanent use, nomination, morning use and average expenditure 6,429 yen or more. The results of the analysis are shown in Table 1.

A Scree plot of eigenvalues of synthetic variables obtained by analysis is shown in Fig. 16. As shown in Fig. 16, it can be seen that the inclination of the line of the Scree plot suddenly changes from the second line. Therefore, we adopt a second number of synthetic variables according to Scree plot criteria. When adopting up to two synthetic variables, the cumulative contribution ratio of the synthetic variable is 51.1% (Table 2).

Fig. 16.

Scree plot of Quantification category 3 analysis result

Table 2. Quantification category 3 analysis result item category

3.1.3 Consideration

3.1.3.1 Interpretation of Composition Variable ∙ Solution 1

Table 3 shows coloring of variables belonging to each composition variable.

Table 3. Coloring compound variable scores

Elements living in Kanto (4 prefectures) have the greatest influence on composition variable ∙ solution 1. Next, factors with a high degree of influence are morning use and autumn use. Composite variable ∙ solution 1 is composed of the above three. What can be thought of as a factor that reduces the use in the morning and the use of autumn when living in Kanto (4 prefectures) is the ease of accessing the store from home. The positional relation of the hair salon chain store used in this research is relatively close. Therefore, if you live in Kanto (4 prefectures) it can be said that it is easy to access. On the contrary, it is speculated that customers coming from outside of Kanto (4 prefectures) with poor access are not due to hair salons, but another main schedule. Those who do not live in Kanto (4 prefectures) have long hair trip round trip time thus it is difficult to think of going to a hair salon at the end of work. In other words, I guess that we will use a hair salon when visiting the city center on a holiday etc. or before the schedule from the afternoon. We think that autumn use is not a big factor because of its low value. However, the characteristics of autumn are strong wind, strong temperature and temperature difference, heavy rain due to typhoon. Therefore, the influence of access is considered to be related. From the above, the composition variable ∙ solution 1 was interpreted as “Ease of access”.

3.1.3.2 Interpretation of Composition Variable ∙ Solution 2

Synthetic variable ∙ solution 2 is the one most influenced by permanent use. Next, nomination, Average expenditure 6,429 yen or more, summer use, winter use, spring use. The composition variable ∙ solution 2 consists of the above six elements. The reason why summer use, nomination, Average expenditure 6,429 yen or more decreases when perming is used is the hair length. Because people sweat in summer, I think that perm is not suitable. In addition, permanent and color are correcting hair and discoloration with dyes thus It does not directly affect the change in hair length. It is possible to redo if the dye is washed as soon as possible. On the other hand, cuts cannot be redone to cut hair thus we speculate that customers tend to appoint stylists that will fulfill their wishes. Moreover, cutting tends to be cheaper compared to perm. As the reason that winter use and use of spring have a negative value, the temperature in spring and winter tends to be low, thus I speculated that it is suitable for the season when using perm. In addition, we think that those who use permanence will give preference to gorgeousness rather than quietness. From the above, the composition variable ∙ solution 2 was interpreted as “styling to produce glamorous”.

3.2 Cluster Analysis

We got two solutions of the synthetic variables. Perform cluster analysis from the sample score of it.

3.2.1 Analysis Method

Used version: SPSS Statistics 22 from IBM Company.

In this research, we used hierarchical cluster analysis to used large data. When performing hierarchical cluster analysis, it is necessary to determine the number of groups first. Then, we do cluster analysis multiple times. Using the residual sum of squares in the cluster obtained at that time, perform the elbow method. Also, the hierarchical cluster analysis used in this research used the k-means method.

3.2.2 Result of Analysis

3.2.2.1 Determining the Number of Groups

The residual sum of squares within the group used in shown in Table 4.

Table 4. The residual sum of squares corresponding to the number of groups

Also, an elbow diagram is shown in Fig. 17.

Fig. 17.

Elbow diagram

As can be seen from this figure, the numerical value sum of squares residual within group decreases toward the group number 5 (Table 5).

Table 5. Hierarchical cluster analysis

The result was changed by riding in the order of data. Therefore, rearrange the data several times and proceed with the analysis.

• First time, when the customer number is in ascending order.

• Second time, when the store number is in ascending order.

• Third time, when the compound variable solution 1 is in ascending order.

• Forth time, when the compound variable solution 2 is in ascending order.

3.2.3 Consideration

There was a big difference in each initial value, group size, group convergence value from 1^st time to 4^th time.

Indicate the positional relationship for each group of solutions 1.2 of the first to 4^th synthesis variables.

From this figure, it is found that the average value of the composition variable solution 1.2 of each group from the 1^st time to the 4^th time is different.

In the k-means method, it is necessary to adopt the best solution from multiple analyzes. In this research, the purpose is to calculate the characteristics of the store. Therefore, one-way analysis of variance and multiple comparison are performed based on the composition variable solution 1.2 obtained by Quantify 3 types.

3.3 One-Way Analysis of Variance and Multiple Comparison

3.3.1 Analysis Method

Used version: SPSS Statistics 22 from IBM Company.

In this research, we used the Welch method when did One-way analysis. Also, we used nonparametric method because the data we used did not follow the normal distribution. In the multiple comparison, the Games-Howell method was used.

3.3.2 Result of Analysis

We were able to judge that only the second time has a significant difference in One-way analysis of variance and multiple comparison. On the other hand, In the first, third and fourth rounds, there was a combination which was judged as significant difference in one-way analysis of variance but there was no significant difference in multiple comparison. As a result, we judged that only the second times output useful results in hierarchical method cluster analysis.

Perform the characteristic analysis of each group from Fig. 18.

Fig. 18.

Composite variable solution of each group average

First, when Synthetic variables answer 1 shows a large positive value, it is considered to be a customer living in the area where access is easy. Conversely, when it shows a large numerical value to a negative, it is speculated that it is a customer other than Kanto 4 prefectures.

When Synthetic variables answer 2 shows a large value to plus, we thought that it is a customer who likes directing and styling. Conversely, when I showed a large value to minus, I thought that it was a customer who likes gorgeousness and styling. When interpreting with only the value of the synthetic variable solution, Customers may live in places where access to stores is reasonably good. Also, we speculated that it is a customer group that likes ornate styling.

3.3.2.1 “Group1” Consideration

Group1’s Synthetic variables answer 1 was 0.765 and answer 2 was 1.059. The Synthetic variables answer 1 shows the third highest positive value among the five groups, and the value of the Synthetic variables answer 2 shows the highest positive value among the five groups.

Looking at the results of basic statistics

• : The customer group with the highest average age group.

• : The highest proportion of female customers.

• : Color usage rate is as high as 89%.

• : The average usage price of 6,429 yen or more shows the highest value at 100%

• : As the proportion of only the first visitor is 1%, and the ratio of more than 4 times per year is 93%, we estimate that it is a customer group with a high maintenance rate

• : Spring, summer, autumn and winter are equally used.

: Based on the above, I interpreted the group 1 group of customers as “Changing hair color according to the season A good customer group of young people and elderly who live in the neighborhood”.

3.3.2.2 “Group2” Consideration

Group1’s Synthetic variables answer 1 was 0.971 and answer 2 was −1.109. Synthetic variables answer 1 has the second highest positive value among the five groups and Synthetic variables answer 2 shows the lowest negative value among the five groups

Looking at the results of basic statistics

• : The average age is the third among the five groups, and the largest age group is 26–37 years old

• : The only customer group whose male and female ratios are reversed in 64% males and 36% females

• : The group with the highest use in winter

• : A group with the highest percentage of customers visiting on holiday morning or in the morning

• : Store with low color utilization ratio

Based on the above, I interpreted the group 2 group of customers as “A group of customers who prefer a quiet style, centering on men living nearby”.

3.3.2.3 “Group3” Consideration

Group3’s Synthetic variables answer 1 was −0.677 and answer 2 was −0.816. Group 3 is the only group showing the negative values for both synthetic variable answer 1 and 2. It is a customer group that has a high possibility of living in areas where access to stores is bad and tends to prefer quiet styling.

Looking at the results of basic statistics

• : The group with the lowest average age

• : The proportion of men is relatively high at 45%

• : Residents in Kanto, use in the morning, designated use, Average usage amount 6429 yen or more, use of perm, use of summer/autumn is low percentage

• : Many people use inexpensive menus

• : There are few people using coupons

Based on the above, I interpreted the group 3 group of customers as “This group is a group of people trying cut menu, young people who are non-regular”.

3.3.2.4 “Group4” Consideration

Group4’s Synthetic variables answer 1 was 1.305 and answer 2 was 0.084. People living in areas that are easy to access store, people who prefer quiet styling, and people who like luxury styling are mixed in Group 4.

Looking at the results of basic statistics

• : The average age is the second highest among the 5 groups

• : Kanto resident, nominated use, summer/fall utilization rate is the highest among the 5 groups

• : Visit uniformly in spring, summer autumn and winter

Based on the above, I interpreted the group 4 group of customers as “Group of young women in the neighborhood who change hairstyles by season”.

3.3.2.5 “Group5” Consideration

Group5’s Synthetic variables answer 1 was −0.904 and answer 2 was 1.001. There are many people who live in areas with poor access to stores or prefer quiet styling.

Looking at the results of basic statistics

• : Coupon usage rate is the highest among 5 groups

• : People who have an average usage price of 6429 yen or more are very high, 99%

• : Color menu usage rate 60%

Based on the above, I interpreted the group 5 group of customers as “This group is a group of people trying out the color menu, Young people who are save money”.

3.3.2.6 Percentage of Customers by Group

We found that there are many customers in groups 3 and group 5. Therefore, we guess that there are many new customers overall (Fig. 19).

Fig. 19.

Number of customers by group for each store

On the other hand, group 1 and group 4, which were presumed to be good customers, are not so many. What the group’s customers are seeking is “cut menu”, but because customers in Groups 1 and 4 are “color permanent menus”, move Group 3 customers to groups 1 and 4 Is difficult. Therefore, Group 3 can also be judged as a customer group that is highly likely to become a departed customer. Because the usage rate of “permanent menu” is high in the group 5 customer, there is a high possibility of moving to group 1 and group 4. Therefore, it is judged that it is a prospective customer highly likely to become a good customer. Therefore, it can be judged that “Styling by color menu is a strength” as the whole 12 stores.

3.3.3 Consideration by Store

From Table 6, it can be seen that Group 3 with the largest number of samples has the largest belonging number in most shops. As Group 3 customers always exist in high proportion at all stores, we can guess that they are basic customer groups. Therefore, it is not considered for interpretation unless it is extremely high or low.

Table 6. Number of group customers by store

As a result of considering each store based on these, it was possible to divide it into four store patterns.

3.3.3.1 Royal Customer Store

Royal customer stores are “A”, “B”, “G” and “K” (Fig. 20).

Fig. 20.

Number of customers by group by Royal customer store

The common thing to the four is that the value of Group 1 is above the average. Group 1 is considered a good customer in a certain hair salon chain. So, it can be said that these four stores are able to acquire good customers and the management is relatively stable.

3.3.3.2 Lead Nurturing Store

Lead Nurturing Stores are “H” and “I” (Fig. 21).

Fig. 21.

Number of customers by group by Lead Nurturing Store

H store

The strength of the hair salon I have handled this time is “Color Menu”, and the customer of Group 5 can be said to be a promising customer. Because the value of Group 5 is higher than the average, we thought that we could acquire good customers by approaching existing customers.

I store

As the value of group 2.3 with many men is high, what is required in this store is quiet styling (cut). From that we can say that a promising customer of “I” is Group 3. We thought it would be nice to have successful guidance to excellent customers because the value of group 3 is high.

3.3.3.3 Lead Generation Store

Lead Generation Stores are “C”, “E”, “F” and “L” (Fig. 22).

Fig. 22.

Number of customers by group by Lead Generation Store

These four stores are stores where the value of the prospect’s fifth group is lower than the average or there are few good customers. It is necessary to acquire new customers and promising customers.

3.3.3.4 Miss Match Store

Miss match store is “D” (Fig. 23).

Fig. 23.

Number of customers by group by Miss match store

There are few promising customers, and the proportion of groups that may not be repeat customers is very high. In fact, the proportion of repeat customers is very low, and this hair salon may not correspond to the area.

3.3.3.5 Lead Nurturing and Lead Generation Store

Lead Nurturing and Lead Generation Store is “J” (Fig. 24).

Fig. 24.

Number of customers by group by Lead Nurturing and Lead Generation Store

The city of the location is under development. Therefore, it is necessary to simultaneously perform the Lead Nurturing of the current customer and the lead generation of the new customer.

3.3.4 Suggestion by Store

3.3.4.1 Royal Customer Store

Continue to lead nurturing and continue to nurture superior customers. Management will be stable if we maintain current status.

3.3.4.2 Lead Nurturing Store

Because “H store” has many future customers, it is important to promote them to promising customers.

I store is a new store, firstly it is necessary to improve store brand loyalty. In order to secure continuous high-quality customers, it is necessary to develop strengths centering on cuts that tend to favor men.

3.3.4.3 Lead Generation Store

It is urgent to attract prospective customers. Due to the fact that many lead generation shops have many competing stores in the vicinity, differentiation from other stores is necessary. Should appeal about ornate hair style menu which is the strength of certain hair salon.

3.3.4.4 Miss Match Store

J store is not a glamorous hairstyle, which is the strength of a certain hair salon, should enhance the cut menu that this store customer is seeking.