You are on page 1of 9

Executive Summary

This project is based on drivers of store image data to identify market segments as well as to determine entry strategy
for Whole Foods. After estimating mixture regression models in GLIMMIX, the segment with higher demand for
atmosphere and service (Quality Segment), is recommended for Whole Foods to target. Seven regions are chosen to
enter initially: Hamburg, Koln, Asturias, Galicia, Alsace, Friuli-Venezia Giulia and Alentejo, considering their
population and likelihood of belonging to Quality Segment. When targeting Quality Segment, recommended
positioning for Whole Foods is to offer the best atmosphere and service. More recommendations are also made for
Whole Foods to enter Europe market successfully.

Introduction and Background


Globalization, an irresistible trend, is happening in all the industries. Hence, identifying market segments not only
inside one country but also across the world has become the first-line for international expansion strategy. Although
customers live in different regions, they often have more in common with each other if they are in the same segment.
When retailers consider global expansion, Rollout strategies are often used, that is to start from a market similar to
their original market, and then enter neighboring regions to expand.
Whole Foods is an American supermarket chain that specializes in selling natural and organic food products. Having
494 stores in North America, Whole Foods is famous for their premium Quality Standard and environmental
involvement. Following the dominant growth strategy and international expansion, Whole Foods is planning to get
into European market. With over $750 billion in sales a year, the food retailing industry is among the largest
industries in the European Union. The data of retail stores in seven European countries, sponsored by the European
Commission is used by the team to give recommendations for Whole Foods’ Europe expansion plan.

Data and Methodology

A survey dataset for the household’s primary food retail outlet is identified and the most frequently visited food
store, has been collected from a sample of 1669 respondents in 105 regions across seven countries. The respondents
rate store image and its drivers including service, atmosphere and price on a likert scale of 1 to 7. Two datasets have
been provided, one is “EURetail.dat” which is used as the data input to estimate mixture regression models. The
other is “EURetail.GL1”, a file having variable definitions.
Mixture Regression Model is a model in which the most optimal number of segments can be identified. And within
each segment, a dependent variable is associated with a group of independent variables according to some form of
response model (linear or nonlinear), and the parameters of the response model differ between the segments. The
analysis has been done using GLIMMIX software, which is used for Mixture Regression Models.,

There are three stages involved :


● Identifying the optimal number of segments
● Developing response model for each segment
● Identifying the probability of belonging to each segment.
The regions are then assigned to each of the segments based on their posterior probability.

The first part is to identify the optimal number of segments. This is done by observing CAIC and by referring to ” the
plot of statistics against the number of segments” in the output result. The lower the CAIC value the better it is.
CAIC has been observed for 1 to 4 number of segments. By observing the CAIC in “statistics” in the output file, we
see that the lowest value is when the number of segments is two.(see Appendix II Figure 1 Figure 2).
The second part is to determine the response model for the segments. The data follows Normal distribution, so it is
necessary to generate mixture regression models for each segment to capture the effects of different
drivers(variables) on the overall store image. Maximum Likelihood Estimation methodology is used over here. The
parameters are different between two segments. (see Appendix II Figure 3).
The third part is to identify the probability of a region belonging to each segment. First, find the segment size, which
is determined by the probability of belonging to each segment and referring to the “coefficients and standard errors”
in the output file. Four iterations are selected to reaffirm our class size for 2 segments. The probability of being
present in the first segment is 0.305 and the probability of being present in the second segment is 0.694. So, the class
size of first segment is 30.5% and the class size of second segment is 69.4%. (see Appendix II Figure 3).
Next step is to assign each region to one of the identified segments. Each region is assigned to one of the segments
by comparing the posterior probability of lying in each of the two segments. The maximum value out of the two
values determines in what segment that particular region will belong to. (see Appendix II Figure 4).

Key Findings

In order to identify market segments based on drivers of Store Image and to choose the entry strategy for a particular
food retailer, mixture regression analysis is carried out. Store Image is analyzed with respect to factors such as
Region, Respondent, Service, Atmosphere and Price. The independent variables Service, Atmosphere and Price are
all categorical and the variables Region and Respondent and the Dependent variable Store image is continuous in
nature.

The GLIMMIX software is run with the following input for a Mixture Regression Analysis:
Store_image = a + b1*Service + b2*Atmosphere +b3*Price
Link Function - Identity
Distribution Level - Normal Distribution
Estimation Method - Maximum Likelihood Estimation (MLE)
Number of Classes - Min -1 and Max - 5

BIC/CAIC is a criterion for model selection among a finite set of models; the model with the lowest BIC/CAIC is
preferred.It has been found that the BIC/CAIC value (3410.67) and AIC value (3351.06) is the lowest for two
Segments/classes. Hence, we choose only two segments for making the analysis.

Coefficient estimates given by GLIMMIX (See Appendix) can be interpret as the importance of each driver to each
segment. Segment 2 has a size of 0.695 and are more sensitive to Service (Estimate Seg2 - 0.3528 > Estimate Seg1 -
0.2628) and Atmosphere (Estimate Seg2 - 0.3028 > Estimate Seg1 - 0.1571) than to Price (Estimate Seg2 - 0.2395 <
Estimate Seg1 - 0.2968) . Whereas, Segment 1 with a size of 0.305 is more sensitive to Price (Estimate Seg1 -
0.2968 > Estimate Seg2 - 0.2395 ) than Segment 2. In following report, Segment 2 will be referred as Quality
segment, and Segment 1 will be referred as Price segment. Whole Foods, a leading Organic food retailer who has
stringent quality standards and premium pricing, should focus more on Quality Segment (Segment 2) than Price
Segment (Segment 1). Also, the market size for Quality Segment is greater (0.695) than the Price Segment (0.305).
Henceforth, Whole Foods can focus on the Quality Segment.

Regions are assigned to one of the two segments based on the value of posterior membership probability, which is
the probability of the customers belonging to each group. A threshold value of 0.6 is chosen, for the posterior
membership probability in order to assign the regions to one of the two segments. A posterior membership
probability of 0.6 gives a clear separation between the two segments whereas a posterior membership probability of
0.5 shows that people from this region have almost similar preference for both the segments and there is no clear
distinction. Over 74 regions belong to Segment 2( Quality segment) based on the value of Posterior membership
probability.

Hamburg , Koln, Asturias, Galicia, Alsace,Friuli-Venezia Giulia and Alentejo are chosen as the regions to target first
for the following reasons. Firstly, they have a clear preference towards Quality segment (Segment 2) with a
Posterior Membership Probability of 0.999/1.000. They are insensitive to Price and are more conscious about Service
and Atmosphere. Secondly, considering the factors such as Population Density, number of Competitors, Supply
Chain efficiency, it is better to focus on regions such as Hamburg, Koln, Asturias and Galicia, as the Population
Density and the Purchasing Power are higher for people in these regions. (Refer Appendix.III) Focusing on these
regions could help to save time and efforts of the Whole Foods Marketing team as well in comparison to the time and
effort they would spend on regions having sparse population with low purchasing power. Also, Hamburg and Koln
are some of the major port cities and have well connected land routes and sea routes. Transportation of goods and
ingredients to these areas would be much easier than other places as well.
Conclusions and Recommendations

Conclusions

In conclusion, based on drivers of store image and Whole Foods’ competencies, regions that value atmosphere and
service should be taken as the target segments for Whole Foods entering Europe market. To take population as well
as the posterior probability into consideration, seven regions are chosen to enter: Hamburg, Koln, Asturias, Galicia,
Alsace, Friuli-Venezia Giulia and Alentejo.

Recommendations

According to analysis, three major recommendations are made for Whole Foods to enter into European market
effectively.
● Based on Mixture Regression Model and taking geographical factors into consideration, seven regions from the
quality segment are suggested for Whole Food to target first: Hamburg, Koln, Asturias, Galicia, Alsace, Friuli-
Venezia Giulia and Alentejo. Then, in order to expand market size, Whole Foods could turn to other regions of
Quality Segment as its next step. In addition, Whole Foods could lower its price to target at more price-sensitive
segments.
● To target at Quality Segment, Whole Food is suggested to focus on best atmosphere and service, which could
help them differentiate from other organic food store. Also, promotion is recommended for Whole Foods to
increase its brand awareness; and loyalty program is recommended to increase customers’ life value.
● More background knowledge in these regions need to be researched when targeting. There might be other
factors in these regions such as competition, customers’ shopping habits, and distribution that will affect
customers’ purchase intention.

Limitations

In terms of data limitation, firstly, based on the contents of questionnaires, the data counts on a 1-7 point likert scale,
which provides little information about the reference of these numbers since these numbers do not have any
numerical meaning behind and vary among individuals. Therefore, it is hard to understand the descriptive statistics
from this database. Secondly, the data was obtained in 2007, which might not reflect the current situation of
European market effectively, since customers’ purchase power and preference might change over time. Thirdly, it is
hard to evaluate whether these samples could represent the real situation of the whole region, since any background
information about them is unknown.
In terms of information insufficiency, internal factors such as cost structure, and external factors such as competition
and economics environment are not clear, which makes it hard to detect segments’ profitability and growth rate
accurately.
Appendices: Tables, Exhibits, Figures
Appendix.I Data Description

Variable Description
REGION ID number for a region (there are 105 regions in the data, ranging from 1-120,
with a few numbers absent).
CONSUMER ID number for a respondent within a region (= 1, 2, 3, …), each region may
have a different number of respondents.
STORE_IMAGE Overall store image rating for a respondent’s primary food retail outlet, on a
1-7 point scale (missing values are imputed with their corresponding sample
averages and thus may contain half points)
SERVICE Service quality perception of a respondent’s primary food retail outlet, on a 1-
7 point scale (a store image driver)
ATMOSPHERE Store atmosphere perception of a respondent’s primary food retail outlet, on a
1-7 point scale (a store image driver).
PRICE Price perception of a respondent’s primary food retail outlet, on a 1-7 point
scale (a store image driver). Note that higher value indicates more favorable
price perception (i.e., low prices, reasonable prices).

Appendix.II Segment

Figure.1 AIC,CAIC,BIC,MAIC Comparing


Figure.2

Figure 3
#classes: 2 current: 1 startnr: 1
Independent Coefficient estimates STD.ERR T-value
Service 0.268627 0.025496 10.536228
Atmosphere 0.157096 0.025302 6.208827
Price 0.296865 0.020900 14.204236
intercept 1.724799 0.127082 13.572384

CLASS SIZE FOR SEGMENT 1 = 0.305145


VARIANCE FOR SEGMENT 1 = 0.612576

#classes: 2 current: 2 startnr: 1


Independent Coefficient estimates STD.ERR T-value
Service 0.354847 0.020562 17.257265
Atmosphere 0.302888 0.020597 14.705321
Price 0.239546 0.014525 16.492079
intercept 0.695696 0.093255 7.460167

CLASS SIZE FOR SEGMENT 2 = 0.694855


VARIANCE FOR SEGMENT 2 = 0.333251
Figure 4
Posterior Posterior
Membership Membership Predicted
Probability: Probability: Store
Region Country Region Name Segment 1 Segment 2 Image
1 BRDeutschland Arnsberg 0.107 0.893 5.345
2 BRDeutschland Berlin 0.306 0.694 5.415
3 BRDeutschland Brandenburg 0.034 0.966 5.318
4 BRDeutschland Braunschweig 0.534 0.466 5.497
5 BRDeutschland Bremen 0.226 0.774 5.387
6 BRDeutschland Chemnitz 0.198 0.802 5.377
7 BRDeutschland Darmstadt 0.794 0.206 5.589
8 BRDeutschland Dessau 0.140 0.860 5.356
9 BRDeutschland Detmold 0.422 0.578 5.457
10 BRDeutschland Dresden 0.109 0.891 5.345
11 BRDeutschland Dusseldorf 0.016 0.984 5.312
12 BRDeutschland Freiburg 0.017 0.983 5.312
13 BRDeutschland Giessen 0.147 0.853 5.359
14 BRDeutschland Halle 0.046 0.954 5.323
15 BRDeutschland Hamburg 0.001 0.999 5.307
16 BRDeutschland Hannover 0.753 0.247 5.575
17 BRDeutschland Karlsruhe 0.114 0.886 5.347
18 BRDeutschland Kassel,Landkreis 0.350 0.650 5.431
19 BRDeutschland Koblenz 0.177 0.823 5.370
20 BRDeutschland Koln 0.001 1.000 5.307
21 BRDeutschland Leipzig 0.767 0.233 5.580
22 BRDeutschland Luneburg 0.011 0.989 5.310
23 BRDeutschland Magdenburg 0.290 0.710 5.410
Mecklenburg-
24 BRDeutschland Vorpommern 0.232 0.768 5.389
25 BRDeutschland Mittelfranken 0.288 0.712 5.409
26 BRDeutschland Münster 0.194 0.806 5.375
27 BRDeutschland Niederbayern 0.165 0.835 5.365
28 BRDeutschland Oberbayern 1.000 0.000 5.663
29 BRDeutschland Oberfranken 0.171 0.829 5.367
30 BRDeutschland Oberpfalz 0.494 0.506 5.482
31 BRDeutschland Rheinhessen-Pfalz 0.980 0.020 5.656
32 BRDeutschland Saarland 0.085 0.915 5.337
33 BRDeutschland Schleswig-Holstein 1.000 0.000 5.663
35 BRDeutschland Stuttgart 0.018 0.982 5.313
36 BRDeutschland Thuringen 0.010 0.990 5.310
37 BRDeutschland Trier 0.442 0.559 5.464
39 BRDeutschland Unterfranken 0.007 0.993 5.309
41 Belgique-Belgie Antwerpen 0.997 0.003 5.662
42 Belgique-Belgie Brabant 0.139 0.861 5.356
43 Belgique-Belgie Hainaut 0.045 0.955 5.322
44 Belgique-Belgie Liege 1.000 0.000 5.663
45 Belgique-Belgie LimburgB 0.297 0.703 5.412
46 Belgique-Belgie LuxembourgB 0.021 0.979 5.314
47 Belgique-Belgie Namur 0.147 0.853 5.359
49 Belgique-Belgie West-Vlaanderen 1.000 0.000 5.663
50 Espana Andalucia 0.975 0.025 5.654
51 Espana Aragon 0.960 0.040 5.649
52 Espana Asturias 0.001 0.999 5.307
53 Espana Cantabria 1.000 0.000 5.663
54 Espana Castilla-LaMancha 0.856 0.144 5.611
55 Espana Castilla-Leon 0.099 0.901 5.342
56 Espana Cataluna 0.055 0.945 5.326
57 Espana ComunidadValenciana 0.076 0.924 5.333
58 Espana Extremadura 0.552 0.448 5.503
59 Espana Galicia 0.001 1.000 5.307
60 Espana Madrid 0.428 0.572 5.459
61 Espana Murcia 0.104 0.896 5.343
62 Espana Navarra 0.774 0.226 5.582
63 Espana PaisVasco 0.340 0.660 5.428
64 Espana Rioja 0.019 0.981 5.313
65 France Alsace 0.001 0.999 5.307
66 France Aquitaine 0.133 0.867 5.354
67 France Auvergne 0.248 0.752 5.395
69 France Bourgogne 0.170 0.830 5.367
70 France Bretagne 0.279 0.721 5.406
71 France Centre 0.674 0.326 5.546
72 France Champagne-Ardenne 0.055 0.945 5.326
75 France IledeFrance 0.226 0.774 5.387
76 France Languedoc-Roussillon 0.057 0.943 5.327
77 France Limousin 0.319 0.681 5.420
78 France Lorraine 0.071 0.929 5.332
79 France Midi-Pyrenees 0.042 0.958 5.321
81 France PaysdelaLoire 0.056 0.944 5.326
82 France Picardie 0.141 0.859 5.357
83 France Poitou-Charentes 0.586 0.414 5.515
85 France Rhone-Alpes 0.353 0.647 5.432
86 Italia Abruzzi 0.961 0.039 5.649
90 Italia Emilia-Romagna 0.202 0.798 5.378
91 Italia Friuli-VeneziaGiulia 0.000 1.000 5.307
93 Italia Liguria 0.225 0.775 5.387
94 Italia Lombardia 0.042 0.958 5.321
95 Italia Marche 0.009 0.991 5.310
96 Italia Molise 0.067 0.933 5.330
97 Italia Piemonte 0.996 0.004 5.661
98 Italia Puglia 0.409 0.591 5.452
99 Italia Sicilia 0.412 0.588 5.453
100 Italia Toscana 0.080 0.920 5.335
102 Italia Umbria 0.058 0.942 5.327
103 Italia Veneto 0.535 0.465 5.497
104 Nederland Drenthe 0.002 0.998 5.307
105 Nederland Flevoland 0.662 0.338 5.542
106 Nederland Friesland 0.999 0.001 5.662
107 Nederland Gelderland 0.008 0.992 5.309
108 Nederland Groningen 0.215 0.785 5.383
109 Nederland LimburgNL 0.036 0.964 5.319
110 Nederland Noord-Brabant 0.148 0.852 5.359
111 Nederland Noord-Holland 0.033 0.967 5.318
112 Nederland Overijssel 0.661 0.339 5.542
113 Nederland Utrecht 0.139 0.861 5.356
114 Nederland Zeeland 0.072 0.928 5.332
115 Nederland Zuid-Holland 0.061 0.939 5.328
116 Portugal Alentejo 0.001 0.999 5.307
117 Portugal Algarve 0.049 0.951 5.324
119 Portugal LisboaeValeDoTejo 0.999 0.001 5.663
120 Portugal Norte 0.007 0.993 5.309

Appendix. III

Hamburg - Population - 1,707,901


Population Density - 2,424 per Square Km

Koln - Population - 976,485


Population Density - 2,641 per Square Km

Asturias - Population - 1,034,960


Population Density - 99.9 per Square Km

Galicia - Population - 2,741,000


Population Density -92 per Square Km

Alsace - Population- 1,859,900


Population Density- 224.6 per Square Km

Friuli-Venezia Giulia- Population- 1,213,000


Population Density- 150 per Square Km

Alentejo - Population- 769,333


Population Density- 24 per Square Km

You might also like