See discussions, stats, and author profiles for this publication at: [Link]

net/publication/363505825

Data Analytics in Customer Segmentation and RFM Method

Thesis · June 2022

DOI: 10.13140/RG.2.2.28067.94244

CITATIONS READS

0 2,193

2 authors:

Aylanur Cuce Eda Tıṙ yakı ̇

Istanbul Technical University Istanbul Technical University
2 PUBLICATIONS 0 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Aylanur Cuce on 13 September 2022.

The user has requested enhancement of the downloaded file.

ISTANBUL TECHNICAL UNIVERSITY  FACULTY OF MANAGEMENT

DATA ANALYTICS IN CUSTOMER SEGMENTATION AND RFM METHOD

[Link]. THESIS

Aylanur CÜCE
Eda TİRYAKİ

Department of Industrial Engineering

JUNE 2022
ISTANBUL TECHNICAL UNIVERSITY  FACULTY OF MANAGEMENT

DATA ANALYTICS IN CUSTOMER SEGMENTATION AND RFM METHOD

[Link]. THESIS

Aylanur CÜCE
(070180101)
Eda TİRYAKİ
(070180085)

Department of Industrial Engineering

Thesis Advisor: Assoc. Prof. Esra BAŞ

JUNE 2022
 ISTANBUL TEKNİK ÜNİVERSİTESİ  İŞLETME FAKÜLTESİ

MÜŞTERİ SEGMENTASYONUNDA VERİ ANALİTİĞİ VE RFM YÖNTEMİ

LİSANS TEZİ

Aylanur CÜCE
(070180101)
Eda TİRYAKİ
(070180085)

Endüstri Mühendisliği Anabilim Dalı

Tez Danışmanı: Doç. Dr. Esra BAŞ

HAZİRAN 2022
 To our loved ones,

v
vi
FOREWORD

This thesis was written for our Bachelor's Degree in the Department of Industrial
Engineering at Istanbul Technical University.
We would like to express our gratitude to our thesis advisor, Esra Baş, for always being
reachable and supportive, and guiding us throughout the semesters in every aspect. We
would also like to underline the emotional support we have recived from our families
and friends throughout the year and to thank them for always being there for us.

July 2022 Eda Tiryaki

Aylanur Cüce

vii
viii
TABLE OF CONTENTS

Page

4.3.1 Segmentation with Agglomerative Hierarchical Clustering ..................... 37

4.3.2 Segmentation with K-means Clustering ................................................... 38

ix
x
ABBREVIATIONS

AHC : Agglomerative Hierarchical Clustering

AHP : Analytic Hierarchy Process
CDA : Customer Data Analytics
CLV : Customer Lifetime Value
CRM : Customer Relationships Management
CS : Customer Segmentation
DBSCAN : Density-Based Spatial Clustering of Applications with Noise
GMM : Gaussian Mixture Model
L-RFM : Length, Recency, Frequency, Monetary
MDS : Multidimensional Scaling
RFM : Recency, Frequency, Monetary
RSQ : R-squared
SOM : Self Organizing Map
STP : Segmentation-Targeting-Positioning

xi
xii
LIST OF TABLES

Page

Table 3.1 : Some researches from literature about classical RFM applications. ....... 27
Table 3.2 : Some researches from literature about improved RFM applications. ..... 30
Table 4.1.1 : Demographic characteristics of respondents. ....................................... 34
Table 4.1.2 : Distribution of mostly purchased product categories. .......................... 35
Table 4.1.3 : Average scores of some possible reasons to choose online market
services over physical markets. .......................................................................... 35
Table 4.3.1: Similarities within-cluster and between-clusters for AHC. .................. 37
Table [Link]: Similarities within-cluster and between-clusters for k-means. .......... 40
Table [Link]: Number of objects by cluster. ............................................................ 40
Table 4.4.1: Encoding of R score in RFM. ............................................................... 40
Table 4.4.2: Encoding of F score in RFM. ................................................................ 41
Table 4.4.3: Encoding of M score in RFM. ............................................................... 41
Table 4.5.1: Average points of each brand given by target segment. ....................... 42
Table 4.5.2: Euclidean distances between the brands. .............................................. 42
Table 5.1.1: Table showing the cluster centroids for gender. ................................... 44
Table 5.1.2: Table showing the cluster centroids for age intervals. .......................... 45
Table 5.1.3: Table showing the cluster centroids for working status. ....................... 45
Table 5.1.4: Table showing the cluster centroids for education level. ...................... 45
Table 5.1.5: Table showing the cluster centroids for marital status. ......................... 46
Table 5.1.6: Table showing the cluster centroids for individual monthly income. ... 46
Table 5.1.7: Table showing the cluster centroids for importance questions. ............ 47
Table 5.2.1: Numbers and percentages of RFM segments in the dataset. ................. 48
Table 5.2.2: Counts of RFM segments in each cluster. ............................................. 49
Table 5.2.3: Counts and percentages of RFM segments in Cluster 1. ...................... 49

xiii
xiv
LIST OF FIGURES

Page

Figure 4.1.1 : Graph showing the gender distribution of respondents. ..................... 33

Figure 4.1.2 and Figure 4.1.3: Graphs showing the education levels and age
percentages of respondents. ............................................................................... 33
Figure 4.1.4 : Graph showing the marital status distribution of respondents............ 33
Figure 4.1.5 : Graph showing the monthly individual income distribution of
respondents. ........................................................................................................ 34
Figure 4.2: An overview of the encoded dataset. ...................................................... 36
Figure [Link]: Dendrogram and number of clusters obtained from AHC................ 37
Figure [Link]: Dendrogram obtained in SPSS. ........................................................ 38
Figure [Link]: Evolution of inertia. .......................................................................... 39
Figure [Link]: Evolution of silhouette score. ........................................................... 40
Figure 5.2: RFM segments and descriptions ............................................................. 48
Figure 5.3.1: RSQ and Stress values of MDS analysis. ............................................ 50
Figure 5.3.2: Positioning Map. .................................................................................. 50

xv
xvi
 DATA ANALYTICS IN CUSTOMER SEGMENTATION AND RFM
METHOD

SUMMARY

Today, with the increasing consumer population and diversifying product portfolios,
companies have to exert more effort to have a good market share and maintain their
position in the industries in which they operate. Under these tough competitive
conditions, the value of each customer served increases exponentially for the
companies. It is easier for companies to retain their existing customers than to find and
acquire new potential customers. For this reason, it is necessary to better know, define
and classify existing customers in order to provide accurate and effective service.
Every company that needs its customers in order to continue its activities has to give
the necessary value to its customers. Every business that is aware of this should reach
its customers in the most accurate and effective way and provide the necessary service
to their customers. In order to realize all these, companies today care about their
customers' data and collect, process and make sense of all data correctly. With the
effective use of customer data, customer satisfaction and loyalty can be measured and
necessary actions can be taken to increase it.
By making use of data analytics methods, companies that can identify and group their
customers with similar approaches and characteristics can predict their future behavior
and adopt the right marketing approaches. In this context, the segmentation of
customers and the effective interpretation of their data are very important for
competitive advantage. In this thesis, data analytics methods that are frequently used
in customer data processing and segmentation and cluster analysis algorithms are
discussed in detail.
Another approach examined in this study is the RFM method, which is frequently used
in segmentation and targeting. The name of this method comes from the initials of the
words recency, frequency and monetary which are the three parameters used. In this
method, customers are classified according to the date they last made a purchase, the
frequency of their purchases and the amount of expenditure they made.
In this thesis, STP (segmentation-targeting-positioning) analysis, which is a three-
stage marketing strategy, was carried out with the aim of understanding the
preferences, ideas and priorities of users/customers using online market applications
better. In order to make these evaluations in the study, a questionnaire was created to
collect data from online market service users. As the first step of the analysis process,
a consumer segmentation study was conducted using the responses collected from the
survey participants. In the segmentation phase, hierarchical and non-hierarchical
clustering approaches were used and the results were compared. The clusters formed
as a result of the segmentation analysis were then subjected to a conventional RFM
scoring analysis. Finally, using a positioning analysis, inferences were generated
regarding the participant consumer group's perspectives and positions towards online
market service providers.

xvii
xviii
MÜŞTERİ SEGMENTASYONUNDA VERİ ANALİTİĞİ VE RFM YÖNTEMİ

ÖZET

Günümüzde artan tüketici nüfusu ve çeşitlenen ürün portföyleri ile birlikte, şirketler
faaliyet gösterdikleri endüstrilerde iyi bir pazar payına sahip olabilmek ve konumlarını
korumak için daha fazla efor sarf etmek zorunda kalmaktadır. Bu zorlu rekabet
koşulları altında, hizmet sunulan her müşterinin firmalar için değeri katlanarak
artmaktadır. Firmalar için halihazırda var olan müşterilerini elde tutmak, yeni
potansiyel müşteri bulmak ve kazanmaktan daha kolaydır. Bu sebeple, var olan
müşterilerine doğru ve etkili hizmeti sunmak için onları daha iyi tanımaları,
tanımlamaları ve sınıflandırmaları gerekmektedir.
Faaliyetlerini sürdürebilmek için müşterilerine ihtiyaç duyan her firma, müşterilerine
gereken kıymeti vermek durumundadır. Bunun bilincindeki her işletme, müşterilerine
en doğru ve etkili şekilde ulaşmalı ve müşterilerine gerekli hizmeti sunmalıdır. Tüm
bunları gerçekleştirebilmek amacıyla, günümüzde şirketler müşterilerinin verilerini
önemsemekte ve tüm verileri doğru şekilde toplamakta, işlemekte,
anlamlandırmaktadır. Müşteri verilerin etkili kullanımı ile müşterilerin memnuniyeti
ve bağlılığı ölçülebilmekte ve artırmak için gerekli aksiyonlar alınabilmektedir.
Veri analitiği yöntemlerinden de faydalanarak, benzer yaklaşım ve özelliklere sahip
müşterilerini belirleyerek gruplayabilen firmalar, onların gelecek davranışlarını
tahminleyebilir ve doğru pazarlama yaklaşımlarını benimseyebilir. Bu bağlamda,
müşterilerin segmentasyonu ve verilerinin etkili yorumlanması rekabet avantajı
açısından oldukça önemlidir.
Bu çalışmada incelenen diğer yaklaşım ise yine segmentasyon ve hedeflemede sıkça
kullanılmakta olan RFM methodudur. Bu yöntemin ismi, kullanılan 3 parametre olan
Yenilik (recency), Sıklık (frequency) ve Parasallık (monetary) kelimelerinin baş
harflerinden gelmektedir. Müşteriler bu yöntemde en son satın alma yaptıkları tarihe,
satın alma sıklıklarına ve yaptıkları harcama miktarına göre sıralanarak
sınıflandırılmaktadır.
Bu tez çalışmasında, online market uygulamalarını kullanan
kullanıcılarının/müşterilerinin tercihleri, fikirleri ve önceliklerinin daha iyi anlaşılması
amacıyla, üç aşamalı bir pazarlama stratejisi olan STP (segmentasyon-hedefleme-
konumlandırma) analizi gerçekleştirilmiştir. Çalışmada bu değerlendirmeleri
yapabilmek için online market servisi kullanıcılarından veri toplamak için bir anket
oluşturulmuştur. Analiz sürecinin ilk aşaması olarak, anket katılımcılarından toplanan
yanıtlar kullanılarak bir tüketici segmentasyonu çalışması gerçekleştirilmiştir.
Segmentasyon aşamasında hiyerarşik ve hiyerarşik olmayan kümeleme yaklaşımları
kullanılmış ve sonuçlar karşılaştırılmıştır. Segmentasyon analizinin bir sonucu olarak
oluşturulan kümeler daha sonra geleneksel bir RFM skorlama analizine tabi
tutulmuştur. Son olarak, bir konumlandırma analizi kullanılarak, katılımcı tüketici
grubunun çevrimiçi pazar hizmet sağlayıcılarına yönelik bakış açılarına ve
konumlarına ilişkin çıkarımlar üretilmiştir.

xix
xx
 INTRODUCTION

Data analytics allows businesses to investigate all of their data to generate insights to
identify and inform patterns and, in some cases, automate decisions. Organizations
may use data analytics to digitally adjust their culture, making their decision-making
processes more inventive and forward-thinking. Aktan states that making strategic
decisions based on data is significant for companies to continue their existence and
maintain the competitive advantage they have (2018).

Companies value the customer-oriented strategy more than ever before in the face of
rising competition and shifting market dynamics. Marketing units try to collect as
comprehensive and detailed information as possible about their customer portfolio and
use this information in more personalized marketing activities. Customer
segmentation: it is a strategy that aims to get to know the customer well by classifying
the customers in the market with a wide target audience into different needs and
expectations. With segmentation, marketing activities can be better shaped by looking
at the needs and characteristics of the determined target classes, so that products and
services become more attractive to target groups.

Customer segmentation is an easy way to organize and manage your company's

relationships with your customers. The customer segmentation process helps you tailor
your marketing, service and sales plan to the needs of your various customer groups
and increase brand loyalty and customer satisfaction. With the help of customer
segmentation, you can create more targeted campaigns and ads to get better results
from your campaigns and increase your return on investment. You will get to know
your valuable customers better and you can offer different services for them.
According to the research, customer relationship management has many benefits for
companies, they can be listed as follows: increasing customer profit, customizing
services, providing better quality service to customers in a shorter time, increasing
customer retention and loyalty (Kim et al., 2006).

RFM scoring is a widely used method of customer data analytics. RFM expression
consists of the concepts of Recency, Frequency, Monetary. Customer-specific sales

21
and marketing tactics are established by segmenting customers in this sort of a study.
It is a technique that helps determine marketing and sales strategies based on
customers' purchasing habits. RFM helps you get rid of the high costs of mass
marketing by reaching the right customers at the right time. Examining your customers'
past purchasing behavior and segmenting them helps to get to know your customers
closely and increase your sales. Thanks to targeted and personalized marketing
campaigns with RFM, you can increase your return on investment, increase your
customer loyalty and retention rate, so you can have a higher CLV. Baskol says, the
main purpose with RFM is to make predictions about the future purchasing behavior
of customers. While making predictions, it assumes that the past purchasing behavior
of customers may continue over time. While making this assumption, recency,
frequency and monetary value are used as parameters (2020).

In this thesis, the RFM method, which is a method used in customer segmentation, is
emphasized. It was aimed to learn about the characteristics ideas, and preferences of
users/customers of online market services, as well as to conduct a STP (segmentation-
targeting-positioning) analysis which is a three-step strategy that evaluates how
organizations convey their products or services and their advantages to distinct client
categories. In the study, a questionnaire is created to collect data from consumers of
online market services in order to conduct these evaluations.

A consumer segmentation study is conducted as the first phase in the analysis process,
using the responses gathered from the survey participants. Hierarchical and non-
hierarchical clustering approaches are used in the segmentation phase, and the results
were compared. The clusters created as a consequence of the segmentation analysis
are then subjected to a traditional RFM analysis. Finally, inferences regarding the
viewpoints and positioning of the participant consumer group toward online market
service providers are produced using a positioning analysis.

22
 DATA ANALYTICS IN CUSTOMER SEGMENTATION

Today, where digitalization and new communication technologies affect the whole
world, computers and systems connected to computers gradually carry out human
activities. The rapid increase in the world's population and the more diverse human
activities have led to the emergence of high volume of data, more precisely “big data”.
Data analysis is process of examining and compiling the data collected and creating
meaningful analyzes for companies. It can generally be used to consider the current
situation and make predictions for the future. As pointed out by Verdenhofs and
Tambovceva, it is critical for businesses to grasp the notion of predictive analysis
methods and how to perform segmentation with the help of big data (2019). In the
following section, data analysis methods and their use in segmentation are discussed.

Customer Data Analytics

When we analyze CRM operations in two parts, these can be marketing

communication activities and data analytics applications. As stated in a review article,
the usage of data analytics in CRM operations gives businesses the opportunity to
recognize similar habits and behaviours of customers, to segment them, and to even
make prediction about them or potential new buyers (Perera et al., 2018). In this regard,
Customer Data Analytics (CDA) can be defined as processes that help businesses
while making crucial decisions by segmentation and prediction based on the data
obtained from customers’ behaviors.

In a study that draws attention to the importance of effective customer clustering, the
most often used data analytics methodology is found as supervised learning algorithms
with a common usage rate of 32%. Other techniques utilized in numerous business
sectors are hypothesis tests and other statistical analyses (Saha et al., 2021).

Considering the increasing value and power of data, CDA applications start with the
collection of processable customer data through various interaction channels, and
should be followed by the proper analytical applications of customer segmentation,
targeting, and positioning.

23
 Clustering algorithms for customer segmentation

Nowadays, understanding customers and classifying them based on demographics and

purchasing tendencies are significant. This is an important part of CS since it helps
businesses to better customize their marketing jobs to different buyer subgroups in
terms of promotions, campaigns, and product creation. As previously said, the goal of
CS is to model the buyers by distinguishing them based on their wants and attitudes.
With today's world's tremendous data flow, it's more crucial than ever to get to know
your customers so you can reach the proper target audiences with the right messages.

The clustering algorithms help us better understand customers because the customers
having similar characteristics generally interacts with businesses in a similar way, so
the businesses can take advantage of these clustering techniques by creating a proper
marketing strategy for each segment they obtained.

As concluded by Tripathi et al., clustering methods can be divided into two depending
on the approaches they follow: Hierarchical and Non-Hierarchical. The most popular
ones of these two branches are Agglomerative Hierarchical Clustering (AHC) and K-
Means respectively (Tripathi et al., 2018).

Hierarchical methods are the ones used for determining clusters sequentially by
bringing the units together at different stages and determining at what distance level
the units that will enter these clusters are cluster elements. These methods are seen to
be classified into to subcategories as Agglomerative and Divisive.

On the other hand, the non-hierarchical methods are used in order to divide the data
set into k initially determined clusters and can be applied to larger data sets compared
to hierarchical methods. While aiming to create k clusters, each cluster contains at least
one unit and each unit is in only one group in non-hierarchical clustering. K-Means is
one of the most often utilized techniques. The main purpose here is to assign the best
“centroid” points that form the centers of k clusters. The algorithm aims to find the
center values that minimize the loss function (inertia) in each iteration by choosing
these points randomly. An important condition for using this method is that the desired
number of clusters to be created must be known beforehand. Another method called
MiniBatch K-Means works faster than K-Means and can sometimes cluster with small
differences.

24
In addition to these methods mentioned before, there are numerous other data analytics
tecniques used for clustering such as Density-Based Spatial Clustering of Applications
with Noise (DBSCAN), Gaussian Mixture Model (GMM), MeanShift, etc. As an
example, GMM operates by assuming that the data is formed by mixing more than one
normal distribution. This method looks for normal distributions for each of the clusters
in the data. It determines the number of clusters by looking at the amount of normal
distribution in the data set. On the other hand, MeanShift method works by moving
data points towards the centers repeatedly to allocate data points to clusters, where
data has the highest density. There is no need to know the required cluster amount at
the beginning.

25
 RFM METHOD AND APPLICATIONS

There are various methods and algorithms for targeted purposes in order to make sense
of the data collected from the customers. One of these methods is the RFM (Recency,
Frequency, Monetary) Analysis. It is a frequently used, basic and powerful approach
applied for customer segmentation in terms of customer analytics.

Traditional/Classical RFM Analysis

The purpose of using the RFM method is to segment customers by calculating a score
based on their past purchasing behaviors i.e. the dates and prices of the purchases they
made (Christy et al., 2021). Each letter in the word RFM delivers another word: R
stands for Recency, F stands for Frequency, and M stands for Monetary. Based on
empirical analyzes it can be assumed that there is a direct connection between the
ordering behavior of customers and the three variables (Mengen, 2009). Consisting of
the initials of these three variables and combining these metrics after calculation, this
score helps businesses analyze customers’ current status and segment them according
to these scores.

In traditional RFM analyzes R, F, and M are each usually divided into groups in a scale
between 1 and 5 and then ranked. While customers with R, F and M values of 5-5-5,
that is, customers who have made the last shopping very recently, shop frequently, and
spend large amounts on their shopping, are very valuable, on the contrary, customers
with 1-1-1 scores are less valuable for the businesses.

In traditional RFM calculations, all customers of an institution are evaluated with the
same scales, and it is believed that information about which customers are valuable
can be found as a result of the analysis. In this regard, one of the most important points
that has become inadequate in the traditional RFM approach today is segmenting the
customers according to the RFM results. However, segmentation should be done first,
and then RFM method should be applied for each segment. As an example for that, in
the study conducted by Gustriansyah et al., an RFM analysis was combined by the k-
Means application with three clusters (Gustriansyah et al., 2020).

26
Table 3.1 contains some researches and studies from the literature about the classical
RFM application.

Table 3.1 : Some researches from literature about classical RFM applications.

Paper title Basic Basic tools Basic outcomes

objectives

RFM-analysis as a To understand RFM-analysis, Applying RFM as

tool for the consumer ABC-analysis segmentation tool is a
segmentation of behavior to gain for partition of powerful approach to
high-tech profitability in data. improve customer
products’ high-tech tools relationships and, as a
consumers (Tsoy retail sector. result, enhance sales
& Shchekoldin, and profitability.
2016)

RFM-based To detect RFM, The compelling nature

repurchase behavior of the computed RFM
multi-layer
behavior for patterns for features is
perceptron,
customer customers in demonstrated, and it is
support vector
classification and retail sector. concluded that the
machine and
segmentation suggested technique can
decision tree
(Rahim et al., be used with a high
classification
2021) degree of classification
methods for
accuracy.
validation.

Clustering and To provide RFM, data Recommendations and

Profiling of CRM advices to mining methods advices for CRM
Customers Using a middle-sized for implementation are
RFM For company by CS. transaction/sales concluded in order to
Customer data. assist the firm while
Relationship making decisions.
Management
Recommendations
(Maryani &
Riana, 2017)

Customer To suggest a RFM, self RFM scores, loyalty

Segmentation and new model by organizing map and potential values are
Evaluation Based CS while using (SOM) considered, customers
On RFM, Cross- customer value are classified into
selling and and loyalty. sixteen categories in a
Customer Loyalty new SOM.
(Liu, 2011)

27
 Table 3.1 (continued) : Some researches from literature about classical RFM
applications.

Paper title Basic Basic tools Basic outcomes

objectives

Using data mining To profile most RFM RFM analysis found the
techniques for profitable most profit generating
profiling vacationers. customers efficiently
profitable hotel and resulted that there
customers: An are eight clusters.
application of
RFM analysis
(Dursun & Caber,
2016)

An Empirical To conduct RFM and k- Four clusters are

Study on value analysis means obtained based on the
Customer and CS by using clustering purchasing behavior.
Segmentation by data of online To achieve a high
Purchase sales of a firm in degree of customer
Behaviors Using a China. satisfaction, many
RFM Model and CRM tactics are
K-Means recommended.
Algorithm (Wu et
al., 2020)

Improved RFM Methods

Advanced methods have been developed so that the results can be interpreted more
accurately and better. Calculations were tried to be made by preventing information
loss in the data set. 2-tuple fuzzy linguistic RFM model has been developed to obtain
more precise results and to interpret more accurately. The most important feature of
the improved RFM model is that it focuses on giving more precise, cleaner results. In
this way, more effective customer segmentation can be made, customer loyalty can be
calculated, and necessary marketing strategies can be created in this context.
According to the researches when the traditional RFM method and 2-tuple fuzzy
linguistic RFM in the customer database in the home furnishing industry are compared,
2-tuple fuzzy linguistic RFM has defined much cleaner segmentation. In this way,
loyal customers and inactive customers are determined more precisely, marketing
activities are defined according to the results obtained, and customer loyalty to the

28
brand can increase (Martinez et al., 2019). The 2-tuple fuzzy linguistic RFM gives
more precise results than the traditional RFM method. Carrasco et al. show that the
traditional RFM method does not perform as well as the 2-tuple fuzzy linguistic RFM
in terms of precision. In the traditional RFM method, it may not be certain which
customer is actually the best customer. A 2-tuple fuzzy linguistic RFM with novelty,
frequency, money, and RFM scores may be preferred for its height to be precise and
easy to interpret (Carrasco et al., 2015).

In another advanced method, it is seen that parameter weights can be differentiated.

As pointed out in Aggarwal and Yadav’s research, the AHP method was used in a gift
shop. While customers are classified into clusters, different weights are given to the
parameters in the RFM method. The parameter that is important for the study was
found to be monetary with a weight of 0.603. According to this result, the customer
who provides more monetary benefits within the cluster will provide more benefits for
the companies than the customer who shopped more frequently and shopped most
recently (Aggarwal & Yadav, 2020). As it was discussed in Shih and Liu’s paper back
in 2003, AHP has been applied to weight the RFM parameters to calculate customer
lifetime value at a hardware retailer. Since the importance of these parameters may
differ in products and sectors, using this method will be useful and valuable in
providing marketing services to satisfy the customer, predicting the behavior of the
customer, and identifying profitable and loyal customers (Shih & Liu, 2003).
Determining the weights of the variables is an important issue and leaving it to the
subjective decisions of the experts may not yield an objective result. According to
Jiang et al., the weights of the parameters may differ from industry to industry. Finding
the most valuable customer in companies is an important issue to increase the profit of
the company. Therefore, the determination of parameter weights should not be
subjective, but should be calculated by considering the purchasing data of the
customers. Calculating the value of customers to the business can be more precise this
way (Jiang et al., 2011).

Another development area has been in the direction of expanding the parameters.
According to the researches, in addition to the parameters that exist in the traditional
RFM method in the training e-commerce sector, the C parameter, which includes
social relations, should be added. In the educational e-commerce sector, where there
are different community features, the use of the social relationship parameter C will

29
be more beneficial in customer segmentation ( Huang, Zhang & He, 2020). Another
new parameter added improved RFM model is L-RFM (Length, Recency, Frequency,
Monetary). Ozkan and Kocakac state that adding the L parameter to the RFM method
in the health sector, especially in areas such as aesthetics, dietetics, and dentistry, will
be beneficial in terms of getting to know customers better and providing appropriate
service in a competitive environment. This parameter is the length of time customers
make their first and last purchases. With the addition of this parameter, the customers
that bring the most profit to the hospitals can be segmented, promotions and marketing
methods can be developed that can increase the customer's shopping volume (Ozkan
& Kocakoc, 2019).

Table 3.2 contains some researches and studies from the literature about the
applications of improved RFM methods:

Table 3.2 : Some researches from literature about improved RFM applications.

Paper title Basic objectives Basic tools Basic outcomes

A comparison To provide a clearer RFM and 2- It can be said that

between Fuzzy idea of how customers tuple Model the two models are
Linguistic RFM will behave in the actually useful, but
Mdel and future in a dataset when the 2-tuple
traditional RFM related to household model is used, much
model applied to goods, to reach the clearer clusters are
Campaign best target customer. obtained.
Management. Case try this using the 2-
study of retail tuple model instead of
business. (Martinez using the traditional
et al., 2019) method.

Customer By calculating the Fuzzy AHP It has been shown

Segmentation customer life value, that giving different
Using Fuzzy-AHP giving more attention weights to the RFM
and RFM Model to the customer group parameters gives
with the highest life much better results,
(Aggarwal &
value and giving and in fact, the
Yadav, 2020).
different weights to customer who
the RFM parameters. brings in more
money has a greater
benefit.

30
 Table 3.2 (continued) : Some researches from literature about improved RFM
applications.

Paper title Basic objectives Basic tools Basic outcomes

A Method for Assigning different AHP and The weights of the

Customer Lifetime weights to parameters RFM parameters can
Value Ranking- and determining differ in different
Combining the weights to better sectors and
analytic hierarchy measure customer life products.
process and value and testing this
clustering analysis out on the hardware
(Shih & Liu, 2003) company.

Determination of Applying RFM with AHP and Although it is an

customer value customer sales data of RFM important issue to
measurement an entrepreneurial determine the
model RFM index company and better parameter weights
weights (Jiang et grouping them by objectively, it can
al., 2011) giving different vary from sector to
weights to the sector.
parameters.

Research on Expanding the K-Means , The structure of

improved RFM parameters to achieve RFM and segmentation
customer better results when RFMC influences the
segmentation segmenting the trueness of customer
model based on K- customer in the segmentation.
Means algorithm training e-commerce
(Huang, Zhang & venture company.
He, 2020) Adding the social
relations parameter C
to these parameters.

31
 APPLICATION OF SEGMENTATION AND RFM ANALYSES FOR
ONLINE SUPERMARKET SERVICES: AN STP APPROACH

In this study, it was aimed to understand the preferences, thoughts and prorities of
users/customers of online market services and to perform an STP (segmentation-
targetin-positioning) analysis. STP stands for Segmentation, Targeting, and
Positioning, a three-step strategy that looks at how businesses promote their products
or services and their benefits to certain client categories. In a nutshell, the STP model
recommends segmenting the market's consumers, targeting particular customer
segments with marketing campaigns customized to their preferences, and adjusting the
firm's positioning based on their needs and expectations.

For the purpose of performing these analyses, a questionnaire was designed to collect
the necessary data from the users of online market services. The questions and possible
options of the questionnaire are given as appendix to this thesis.

As the first step of analyses, a customer segmentation study was carried out with the
answers obtained from the participants by the designed survey. For the segmentation
phase, hierarchical and non-hierarchical clustering methods were applied and
examined. As mentioned before in the previous sections of this paper, the most
common two algorithms of hierarchical and non-hierarchical approaches were
Agglomerative Hierarchical Clustering (AHC) and K-means clustering methods
respectively.

Then, a classical RFM application was performed on the clusters obtained as a result
of the segmentation analysis. Finally, by performing a positioning analysis, inferences
were made about the perspectives and positioning of the participant customer group
towards the online market service providers.

Data Collection and Profile of Questionnaire Respondents

A total of 202 responses were received but during the segmentation analysis, data
obtained from 1 respondent were seen to occur as an “outlier” and removed from the

32
dataset. While conducting the survey with these people, the survey link was sent to our
immediate surroundings, friends, acquaintances, colleagues, and relatives via social
media, direct message, e-mail, etc. It was surely aimed to obtain a random sample to
answer the survey. Therefore, it was aimed to reach not only our own friends who are
mainly students, but also participants from various income, education and age ranges.

When the demographic characteristics of respondents are examined, it is seen that

44,3% of the respondents are male and 55,2% is female.

0,5%
44,3% Female

Male

I prefer not to say/ Other

55,2%

Figure 4.1.1 : Graph showing the gender distribution of respondents.

The educational status and age ranges of the survey participants are shown on the graph
with the percentage distribution as follows:

1% Primary/Secondary 11% 6% 3% 18-23

22% 35%
Education 10% 24-29
Bachelor's Degree
5% 30-35
36-41
6% High School
42-47
Associate Degree/ 12% 48-53
Vocational School 22% 54+
65%

Figure 4.1.2 and Figure 4.1.3: Graphs showing the education levels and age
percentages of respondents.

The options in the question asked about the marital status of the participants and the
percentage distribution of their answers are shown below in Figure 4.1.4.

21% Single, living apart from

33%
my family
Single, living with my
family
Divorced

1% Married
44%

Figure 4.1.4 : Graph showing the marital status distribution of respondents.

Figure 4.1.5 represents the monthly individual incomes of 201 people who participated
in the survey according the given intervals:

33
 29%
29%
0-2500₺
10001-20000₺
20001₺ and more
11% 2501-5000₺
5001-10000₺
21%
10%

Figure 4.1.5 : Graph showing the monthly individual income distribution of

respondents.

Table 4.1 shows the summary of demographic characteristics of the survey

respondents who forms the dataset of the study.

Table 4.1.1 : Demographic characteristics of respondents.

Percentage Percentage
Gender Monthly individual income
Female 55.2% 0-2500₺ 29,4%
Male 44.3% 2501-5000₺ 10,9%
Prefer not to say 0.5% 5001-10000₺ 28,9%
10001-20000₺ 20,9%
Age 20001₺ and more 10,0%
18-23 35,3%
24-29 22,4% Working status
30-35 12,4% Student/Part-timer/Intern 35,3%
36-41 10,0% Full time worker 57,2%
42-47 10,9% Retired 1,5%
48-53 6,0% Housewife 1,5%
54+ 3,0% Jobless/Looking for a job 4,5%

Marital status Education level

Single, living apart from my 21,4% Primary/Secondary 1,0%
family Education
Single, living with my family 43,8% High School 6,5%
Divorced 1,5% Associate Degree/ 5,5%
Vocational School
Married 33,3% Bachelor's Degree 64,7%
Master’s Degree/PhD 22,4%

An important thing to be pointed out in this study is the fact that online market
applications offer almost all of the products and services offered by physical
supermarket stores to customers virtually. To underline it, the survey participants were
asked to choose the 3 categories in which they order products most frequently in the

34
online market applications. Table 4.1.2 shows the product categories provided to
customers in these applications and the distribution of the answers given by 201 people
at the questionnarie.

Table 4.1.2 : Distribution of mostly purchased product categories.

Product categories Number of times Percentage

they are chosen
Water/Beverage 91 15,1%
Fruits and Vegetables 46 7,6%
Basic Food (legumes, pasta, flour, oil, etc.) 120 19,9%
Dairy products/Breakfast 74 12,3%
Meat-Chicken-Fish Products 25 4,1%
Snack (cookie, wafer, chocolate, chips, etc.) 128 21,2%
Personal Care/Cosmetics 43 7,1%
Home Care/Cleaning 34 5,6%
Baby Care 6 1,0%
Pet Care 7 1,2%
Technology/Electronics 29 4,8%

Possible reasons for participants to prefer virtual market applications over physical
market shopping are also examined in the study. The respondents were asked to give
a score between 1 and 5 to some reasons that may led them to use online market
services instead of physical ones, considering 1 as “Not effective/important at all” and
5 as “Extremely effective/important”.

Table 4.1.3 : Average scores of some possible reasons to choose online market
services over physical markets.

Possible factors Average score given by

respondents (between 1-5)
Provides time advantage 4,16
Does not require physical effort 4,19
Offers ease of price comparison 3,56
Product/brand diversity 3,22
Special discount coupons/campaigns for virtual
3,69
shopping
Provides the opportunity to avoid physical contact
3,67
during the pandemic

Data Preprocessing

During the data collection phase, we tried to create a random sample of participants by
having 201 people respond to the questionnaire. The answers collected will be

35
analyzed in software programs like XLSTAT, SPSS, R or Python in order to be used
in the segmentation part of the study.

As a result of the questionnaire that were conducted, the dataset consisting of the
responses of participants was filled with non-numerical data. Since many machine
learning algorithms cannot work directly with categorical data, it was necessary to
convert these categorical data to numerical data. Therefore, for the purpose of
preparing the data to be analyzed in these programs, encoding was performed to
convert these categorical data to processable numerical data. At this part of the study,
One-hot encoding using the necessary Python libraries was performed for the columns
containing categorical data like marital status, education level, or working status. One-
hot Encoder in the “sklearn” library in Python allowed us to perform the binarization
of these categorical data. For each column, it gives "1" to the current value in that row
and "0" to the non-existent ones.

The columns that contain categorical data were the columns of age, education level,
working status, marital status, and income level. After One-hot encoding was
performed, the overview of the dataset obtained can be seen below in the Figure 4.2:

Figure 4.2: An overview of the encoded dataset.

Segmentation Analyses

At this part of the study, the dataset obtained as the result of encoding process will be
utilized for segmentation analyses. Firstly, the most commonly used hierarchical
clustering algorithm (i.e. Agglomerative Hierarchical Clustering) was applied. By

36
using the outputs of AHC, segmentation with k-means algorithm, which is a well-
known non-hierarchical clustering approach, was performed.

4.3.1 Segmentation with Agglomerative Hierarchical Clustering

Hierarchical clustering methods, as previously stated in this thesis, are used to

determine clusters progressively by bringing units together at various stages and
establishing at what distance level the units that will enter these clusters are cluster
elements. Agglomerative and Divisive approaches are recognized as two subcategories
of these strategies.

In this part of the study, Agglomerative Hierarchical Clustering, AHC in short, will be
used. XLSTAT, which is a useful tool of Microsoft Excel to be used in data analysis
applications, was used in order to perform AHC. While performing AHC in XLSTAT,
dissimilarity type to be used in the algorithm was chosen as “Euclidean distance”, and
the agglomeration method was selected as “Ward's method”.

The results produced from the AHC analysis is explained below. The following Figure
[Link] shows the dendrogram obtained from the analysis.

Figure [Link]: Dendrogram and number of clusters obtained from AHC.

According to the dendrogram obtained, it can be concluded that AHC algorithm

proposes that the number of clusters should be 3.

Another result of the AHC analysis can be seen by checking the variance
decomposition for the optimal classification (case of n=3 clusters). The table 4.3.1
below contains the similarity percentages within-cluster and between-clusters.

Table 4.3.1: Similarities within-cluster and between-clusters for AHC.

37
 Absolute Percent
Within-cluster 1353,039 81,66%
Between-clusters 303,807 18,34%
Total 1656,846 100,00%
The similarity rate within-cluster was found to be 81,66 percent, and the similarity
between-clusters was obtained as 18,34 percent. This high percentage of within-cluster
similarity shows that the cases assigned to the same segments are significantly similar
to each other. Accordingly, the low percentage of 18,34 for between-clusters similarity
indicates that the cases from different clusters are not found to be similar to each other.

For the purpose of double-check, AHC analysis was also performed in the statistical
software program SPSS. The dendrogram obtained as the output of SPSS analysis is
given below in Figure [Link].

Figure [Link]: Dendrogram obtained in SPSS.

4.3.2 Segmentation with K-means Clustering

As previously stated, non-hierarchical approaches are used to partition the data set into
k originally selected clusters and, in comparison to hierarchical methods, may be used
to bigger data sets. Non-hierarchical clustering aims to construct k clusters, with each
cluster containing at least one unit and each unit belonging to only one group.

Since the AHC partition is not always ideal, it might be improved by using non-
hierarchical methods to reinforce it and to make the classes more homogenous. At this
part of the study, k-means, the most often used non-hierarchical clustering strategy,
will be performed.

The outcome of AHC analysis indicated and proposed that the optimal number of
clusters to obtain (i.e. “k” in k-means clustering) for our dataset should be 3. This result

38
was surely taken into consideration. Nevertheless, the ideal number of clusters was
also checked from XLSTAT while performing k-means analysis by trying different k
values changing from 2 to 5. In k-means analysis, number of clusters k=2,3,4, and 5
were tried and the results were examined.

For this purpose, Elbow Method was also utilized, because one of the most prominent
ways for determining the best value for k among multiple k values is the Elbow
Method. The value of k at the "elbow," i.e. the point at which the distortion/inertia
starts reducing in a linear way, should be used to find the best number of clusters,
according to this technique. Figure [Link] below shows the change in inertia with
respect to the different k values tried, the elbow occurs in k=3.

Figure [Link]: Evolution of inertia.

Accordingly, Figure [Link] below shows the evolution of the sillhouette score
depending on the number of clusters. Both graphs indicate that k=3 is the optimal point
and the ideal number of clusters for segmentation should be 3.

39
 Figure [Link]: Evolution of silhouette score.

Similar to the variance decomposition obtained in AHC analysis, the Table [Link]
below contains the similarity percentages within-cluster and between-clusters for k-
means analysis with k=3 clusters. The similarity rate within-cluster was found to be
81,90 percent, and the similarity between-clusters was obtained as 18,10 percent. It
can be said that these percantage values are similar to the ones obtained in AHC
analysis.

Table [Link]: Similarities within-cluster and between-clusters for k-means.

Absolute Percent
Within-cluster 1356,934 81,90%
Between-clusters 299,912 18,10%
Total 1656,846 100,00%
K-means analysis with k=3 in XLSTAT resulted in the following distribution of cases
as first cluster containing 88, cluster 2 containing 50, and cluster 3 having 63
participants as seen in Table [Link].

Table [Link]: Number of objects by cluster.

Cluster 1 2 3
Number of objects
88 50 63
by cluster

RFM Analysis for Targeting

In the questionnarie applied, the questions regarding recency, frequency, and monetary
behaviors of participants were designed with suitable options. Therefore, it was proper
to use simple fixed changes method for the determination of RFM scores according to
the following tables. Table 4.4.1 shows the possible expressions of recency and the
corresponding recency scores. Similarly, Table 4.4.2 and 4.4.3 represents the
expression of frequency and monetary, and their corresponding F and M scores
repectively.

Table 4.4.1: Encoding of R score in RFM.

Recency R SCORE

40
 More than 1 month 1
3-4 weeks ago 2
1-2 weeks ago 3
3-7 days ago 4
Today/ 1-2 days ago 5

Table 4.4.2: Encoding of F score in RFM.

Frequency F SCORE
Less frequently 1
Once a month 2
Once in 10-15 days 3
1-3 times a week 4
4-5 times a week /almost every day 5

Table 4.4.3: Encoding of M score in RFM.

Monetary M SCORE
0-100 ₺ 1
101-200 ₺ 2
201-300 ₺ 3
301-450 ₺ 4
451 ₺ and more 5

The answers collected from the participants regarding these 3 RFM questions were
converted to RFM scores according to these encoding tables. Later, the RFM scores of
each individual respondent were determined and examined. According to the RFM
scores of customers at each segment, Cluster 1 was selected as the target segment of
the study. The outputs of RFM analysis will be discussed and investigated in Part 5 in
detail.

Positioning Analysis

According to the results of RFM analysis, Cluster 1 was chosen as the target segment,
as it is the segment containing the customers with the highest RFM scores. The scores
(between 1-5) given by 88 people in Cluster 1, evaluating 8 brands and ideal regarding
8 different attributes, were used here for positioning analysis.

The attributes investigated are: Price, Product/brand diversity, Delivery speed,

Delivery within the specified time frame, Discounts/campaigns, Ease of use of the

41
mobile app/website, Courier's attitude/behavior, Payment diversity and security. The
brands evaluated by the participants are: Getir, Yemeksepeti Banabi, İstegelsin,
Migros Sanal Market, Trendyol Hızlı Market, A101 Kapıda, Macroonline,
CarrefourSA Online, and the ideal brand.

Table 4.5.1 below contains the averages of these points out of 5 that 88 respondents in
Cluster 1 gave to the brands regarding those attributes:

Table 4.5.1: Average points of each brand given by target segment.

delivery
payment
brand/ within the discounts ease of use
delivery courier’s options
price product specified and of the
speed attitude &
variety time campaigns app/website
security
interval
Getir 2,4933 3,0833 3,8406 3,7536 2,8788 3,9429 3,8507 3,9571
Yemeksepeti
2,5161 3,0952 3,6349 3,5968 2,6140 3,7813 3,6833 3,6825
Banabi
İstegelsin 2,3250 3,1220 3,4444 3,4444 2,4545 3,6286 3,5000 3,5714
Migros Sanal
2,6230 3,6000 3,6226 3,6182 2,5306 3,8704 3,7647 3,9615
Market
Trendyol
2,8923 3,7500 3,7869 3,6393 3,2807 4,0328 3,8276 3,8644
Hızlı Market
A101 Kapıda 3,0286 2,9394 2,9259 2,9310 2,8519 3,6400 3,4348 3,4348
Macroonline 2,0667 3,4348 3,0476 3,0476 2,2727 3,7000 3,4762 3,7500
CarrefourSA
2,4375 3,3793 3,0476 3,0909 2,5417 3,6087 3,3478 3,7000
Online
Ideal 4,8068 4,7955 4,7614 4,8182 4,6705 4,8182 4,4659 4,7500

Dissimilarity matrix containing the euclidean distances was formed and given as input
to the SPSS program in order to perform Multidimensional Scaling (MDS). Table 4.5.2
below includes the Euclidean distances between the brands.

Table 4.5.2: Euclidean distances between the brands.

Yemek- Migros Trendyol Carrefour

İste- A101 Macro-
Getir sepeti Sanal Hızlı SA Ideal
gelsin Kapıda online
Banabi Market Market Online
Getir 0,0000
Yemek-
sepeti 0,5169 0,0000
Banabi

42
İstegelsin 0,9126 0,4375 0,0000
Migros
Sanal 0,6953 0,6051 0,8159 0,0000
Market
Trendyol
Hızlı 0,8935 1,0755 1,3822 0,8435 0,0000
Market
A101
1,5358 1,1968 1,1143 1,3095 1,5023 0,0000
Kapıda
Macro-
1,4291 1,0651 0,7417 1,0433 1,6747 1,2402 0,0000
online
Carrefour
SA 1,3050 0,9141 0,6395 0,8720 1,3848 0,8249 0,4724 0,0000
Online
Ideal 3,9049 4,2373 4,6075 3,7964 3,1058 4,2700 4,7361 4,6783 0,0000

43
 OUTPUT ANALYSIS

As mentioned at Part 4 of the study, three segments were acquired, and the outputs of
the analyses performed will be investigated at this part. The data obtained from 201
survey participants were analyzed and separated into three segments by taking the
attributes regarding the gender, age, working and marital status, education and income
level, and the points they gave for various characteristics of online market services into
account.

Segment Profiles

In this part of the study, the outputs obtained from the k-means clustering analysis will
be interpreted and discussed. The segments obtained differ in terms of the listed
attributes.

In terms of gender, Cluster 1 and 3 seem to have a more balanced distribution of male
and female. On the other hand, most of the participants in Cluster 2 are female. The
Table 5.1.1 shows the gender ratios of these three clusters.

Table 5.1.1: Table showing the cluster centroids for gender.

Cluster Gender (0=female, 1=male)

1 0,455
2 0,280
3 0,571

When the age distribution among clusters are considered, it can be seen in the Table
5.1.2 that Cluster 1 has elderly people when compared to the other clusters. More than
50% of participants in Cluster 1 are above the age of 35. In contrast to that, Cluster 2
can be defined as the youngest one among the clusters. More than 80% of Cluster 2
are younger than the age of 30, and only 8% is above the age of 35. Similar to Cluster

44
2, Cluster 3 does also contain young people, but it has a 15% of population older than
the age of 35.

Table 5.1.2: Table showing the cluster centroids for age intervals.

Cluster 18-23 24-29 30-35 36-41 42-47 48-53 54+

1 0,057 0,216 0,193 0,170 0,205 0,102 0,057
2 0,540 0,320 0,060 0,020 0,020 0,020 0,020
3 0,619 0,159 0,079 0,063 0,048 0,032 0,000

In the sense of working status, Cluster 1 has the highest ratio of full time employees
with a percentage of 91. On the other hand, Cluster 2 and Cluster 3 consist of nearly
60% students and interns/part-timers. Cluster 2 has 12% jobless participants, while
Cluster 3 has only 3%. Moreover, there are 35% of full time workers, which is a higher
ratio than Cluster 2. Table 5.1.3 shows the related ratios for all three clusters.

Table 5.1.3: Table showing the cluster centroids for working status.

Full time Student/ Intern Jobless/ Looking

Cluster Retired Housewife
worker Part-timer for a job
1 0,909 0,045 0,011 0,023 0,011
2 0,260 0,600 0,020 0,000 0,120
3 0,349 0,587 0,016 0,016 0,032

As shown below at Table 5.1.4, the education level differs among clusters. Although
Cluster 1 includes 55% of Bahchelor’s degree owners/students, Cluster 2 has the
highest ratio of Bahcelor’s owners with a percentage of 80. The majority of Masters’s
and PhD degree owners are classified into Cluster 1. Cluster 3 occurs as a mixture of
those, containing 67% Bahcelor’s and 20% Master’s degree owners.

Table 5.1.4: Table showing the cluster centroids for education level.

Primary/ Associate Degree/

Bachelor's Master’s
Cluster Secondary High School Vocational
Degree Degree/PhD
Education School
1 0,000 0,091 0,068 0,545 0,295
2 0,020 0,000 0,060 0,800 0,120
3 0,016 0,079 0,032 0,667 0,206

45
According to Table 5.1.5, Cluster 1 stands out as it includes 60 percent of married
individuals. Contrarily, Cluster 2 and 3 both contains single people with a percentage
more than 85.

Table 5.1.5: Table showing the cluster centroids for marital status.

Single, living apart Single, living with

Cluster Married Divorced
from the family the family
1 0,148 0,216 0,602 0,034
2 0,260 0,620 0,120 0,000
3 0,270 0,603 0,127 0,000

When the individual monthly income distributions of respondents assigned in different

clusters are examined by using Table 5.1.6, it can easily be seen that Cluster 1
distinguishes with a higher level of income. Because 90% of individuals in Cluster 1
has a monthly income more than 5000₺, and 45% of them earns more than 10000₺ while
13,6 percent earns even more than 20001₺ per month. In contrast to that, 72% of individuals
in Cluster 2 have a monthly income level of less than 5000₺ and 50% of them actually earns
between 0-2500₺ each month. Lastly, Cluster 3 has a more balanced distribution of incomes
as 58% of people in this segment earns monthly between 0-5000₺, and 42% of them earns
more than 5001₺.

Table 5.1.6: Table showing the cluster centroids for individual monthly income.

20001₺ and
Cluster 0-2500₺ 2501-5000₺ 5001-10000₺ 10001-20000₺
more
1 0,011 0,080 0,466 0,307 0,136
2 0,500 0,220 0,140 0,100 0,040
3 0,524 0,063 0,159 0,159 0,095

In the questionnarie, the participants asked to give points between 1-5 (1 for Extremely
important, 5 for Not important at all) to evaluate the importance of some characteristics
of online market services for them. They faced expressions as “How important are
prices in an ideal virtual market application for you?”, and gave points to these
expressions between 1-5. These characteristics evaluated were price, brand and
product variety, delivery of speed, importance of delivery within the specified time

46
 interval, ease of use of the app or website, courier’s attitude, payment security and
options.

According to Table 5.1.7, people in Cluster 1 give the most importance to (4,8 out of
5 points) delivery within the specified time interval and the ease of use of the mobile
apps. Cluster 2 seems to give not much importance to the mobile app/website design
and the courier’s attitudes, because they gave 3,8 and 2,9 out of 5 points to these
aspects respectively. Lastly, individuals in Cluster 3 do not give as much importance
as others to discounts and campaigns since they gave 3,9/5 to that characteristic.

Table 5.1.7: Table showing the cluster centroids for importance questions.

Importance
Importance
Importance of delivery Importance Importance
Importance of Importance
Importance of brand/ within the of ease of of payment
Cluster of delivery discounts of courier’s
of price product specified use of the options &
speed and attitude
variety time app/website security
campaigns
interval
1 4,807 4,795 4,761 4,818 4,670 4,818 4,466 4,750
2 4,320 4,080 4,400 4,700 4,660 3,820 2,960 4,280
3 4,444 4,365 4,048 4,317 3,968 4,476 4,492 4,762

By taking all these information into consideration, some characteristics of each cluster
are attained. The segment profiles can be summarized as follows:

Cluster 1 appears to be the segment of middle-aged/elderly, married, full time

employees with a higher level of income and education who think that delivery within
the specified time interval and the ease of use of the mobile apps are highly important
while using online supermarket services.

On the other hand, Cluster 2 can be defined as the segment of single, young part-time
workers with relatively lower income level who have/attend Bachelor’s degree. These
young people give not much importance to the mobile app/website design and the
courier’s attitudes.

Last but not least, Cluster 3 seems to be more like a mixture of the other segments. It
contains people with different income levels from different age intervals who give
relatively less importance to the discounts and campaigns while using online markets.

47
 RFM Scores of Customers and Segments for Targeting

In this part of the study, the outputs obtained from the RFM analysis will be interpreted
and discussed. As seen in Figure 5.2, there exists 11 customer segments in RFM
scoring. These segments are widely accepted and used by researchers and
organizations according to the assigned RFM scores of each individual in the dataset.

Figure 5.2: RFM segments and descriptions (Rsquared Academy Blog, 2019).

At the previous stages of this study, RFM scores were firstly assigned to the 201
respondents in the dataset. Then, by considering the 11 RFM segments mentioned, a
corresponding RFM segment were appointed to them. Table 5.2.1 shows the numbers
and percentages of each of these RFM segments occuring in the dataset.

Table 5.2.1: Numbers and percentages of RFM segments in the dataset.

RFM Segment Total number in the dataset Percentage

About to Sleep 19 9,45%
At Risk 3 1,49%
Cannot Lose Them 7 3,48%
Champions 7 3,48%
Need Attention 5 2,49%
Hibernating Customers 22 10,95%
Lost 33 16,42%
Loyal Customers 6 2,99%
New Customers 6 2,99%
Potential Loyalists 91 45,27%
Promising 2 1,00%
After that, the RFM scores were analyzed specifically for each segment (i.e. Cluster 1,
2 and 3).Table 5.2.2 shows the number of RFM segments, which were assigned to each
individual depending on their RFM scores, at each cluster.

48
 Table 5.2.2: Counts of RFM segments in each cluster.

RFM Customer Segment Cluster 1 Cluster 2 Cluster 3

About to Sleep 7 7 5
At Risk 2 0 1
Cannot Lose Them 6 0 1
Champions 3 1 3
Need Attention 4 1 0
Hibernating Customers 10 7 5
Lost 10 9 14
Loyal Customers 4 1 1
New Customers 3 2 1
Potential Loyalists 38 22 31
Promising 1 0 1
TOTAL 88 50 63

As a result of these analyses, the attractiveness of each segment can be seen by looking
at the customers they contain. Respondents in Cluster 1 have higher RFM scores, and
they seem to be classified more as “loyal”, “champion” and “potential loyalist”
according to their RFM segments. Therefore, it can be concluded that Cluster 1 is more
attractive with its customers than the others. The counts and percentages of RFM
customer segments of people within Cluster 1 are shown in Table 5.2.3:

Table 5.2.3: Counts and percentages of RFM segments in Cluster 1.

RFM Customer Segment Count Percentage

About to Sleep 7 7,95%
At Risk 2 2,27%
Cannot Lose Them 6 6,82%
Champions 3 3,41%
Need Attention 4 4,55%
Hibernating Customers 10 11,36%
Lost 10 11,36%
Loyal Customers 4 4,55%
New Customers 3 3,41%
Potential Loyalists 38 43,18%
Promising 1 1,14%

49
 Positions of Popular Brands and the Ideal Brand

In this part of the study, the outputs obtained from the Positioning analysis will be
interpreted and discussed. The matrix containing the average points given to 8 brands
of interest about 8 characteristics, and the matrix including the Euclidean distances
between the brands were utilized.

Figure 5.3.1 shows the RSQ value obtained in the positioning analysis of SPSS. RSQ
(R-squared) value is the fraction of variation in the partition's scaled data that may be
explained by their associated distances. The value is seen as 0.99692 which is really
close to 100% (RSQ = 99.692%).

Figure 5.3.1: RSQ and Stress values of MDS analysis.

As a result of the MDS analysis performed in SPSS, the Positioning map obtained
shown below in Figure 5.3 was obtained:

Figure 5.3.2: Positioning Map.

A configuration map was created as a result of the positioning analysis applied to the
target segment which was Cluster 1. By looking at the positioning map, the vertical
dimension above seems like combination of more than one attribute, therefore it is not
possible for us to predict which attributes it refers to. Similarly, even though dimension
1 seems to be matching up with “ease of use of mobile app/website” attribute, it is
again not certain and it may be a combination of more than one attributes.

50
For a company that will enter the market with a new online supermarket
application/service, it can be easily seen from the map that a service close to the ideal
brand should be established. Even though the brands investigated here look far from
the ideal, “Trendyol Hızlı Market” still seems to be the one closer to the ideal brand
image in the eyes of the target customers of this study.

51
 CONCLUSIONS AND RECOMMENDATIONS

The purchasing power of customers is expanding as technology advances and market

conditions shift. Due to the facts that they provide time advantage and do not require
pyhsical effort, online shopping platforms have started to be used widely and
frequently by many customers. Since online supermarket purchase forms a significant
part of online shopping, online supermarket service providers must provide customers
a cause to like them in order to be chosen. Only by precisely understanding client needs
and providing value propositions that fit those needs can this be accomplished.
Customers, on the other hand, are not all same in terms of their requirements and
expectations while using online market services, as well as the value they bring to the
organization. As a result, rather of treating each client the same, it's vital to segment
them based on the distinctions in their demands and contributions to the companies.

In order to be successful, online supermarket brands should adjust and adopt their
marketing channels, product portfolios, methods of improving product variety, speed
of deliveries, design of mobile applications and websites, pricing policies, and
campaings to their target consumer groups. In this context, customer segmentation is
an easy way for these companies to organize and manage their relationships with their
customers.

In this thesis, firstly a literature review is performed for the data analytics techniques
that are used in CS, and the RFM scoring method. Later, an STP approach is followed
with the purpose of understanding the preferences, ideas and priorities of customers
using online market applications. By using the data collected data from online market
service users via the designed questionnaire, a segmentation analysis is performed.
The customer segmentation process helps adjust the marketing, service, and sales plan
to the needs of various customer groups, increasing brand loyalty and customer
satisfaction. In the segmentation phase, hierarchical and non-hierarchical clustering
approaches were used and the results are compared. The results of AHC analysis imply
that the individuals in the dataset are assigned to 3 clusters. The outcomes of k-means

52
clustering analysis show that Cluster 1 appears to be the segment of middle-
aged/elderly, married, full time employees with a higher level of income and education
who think that delivery within the specified time interval and the ease of use of the
mobile apps are highly important while using online supermarket services. On the
other hand, Cluster 2 can be defined as the segment of single, young part-time workers
with relatively lower income level who have/attend Bachelor’s degree. These young
people give not much importance to the mobile app/website design and the courier’s
attitudes. Last but not least, Cluster 3 seems to be more like a mixture of the other
segments. It contains people with different income levels from different age intervals
who give relatively less importance to the discounts and campaigns.

Later, clusters formed as a result of the segmentation analysis were then subjected to
a traditional RFM analysis. RFM has a great importance because decision makers may
identify critical consumers more effectively and design right strategies by utilizing the
RFM model (Wei et al., 2010). In this study, the outcomes of RFM scores propose that
Cluster 1 containing individuals with higher RFM scores should be the target segment
among all customer groups. Finally, using a positioning analysis, inferences are
generated regarding the participant consumer group's perspectives and positions
towards online market service providers.

The results of this study shows that Cluster 1, which is the segment of middle-
aged/elderly, married, full time employees with a higher level of income and
education, should be the target segment for online supermarket service providers.
Since these customers think that delivery within the specified time interval and the
ease of use of the mobile apps are highly important for them while using online
supermarket services, the brands should attach importance to these issues more in
order to have a greater market share.

Delivery within the promised time interval has a great impact on the impression of the
online supermarket service provider in these customers’ mouth. Not maintaining an
“on-time” delivery service is a problem for both for the companies’ logistics operations
and the customers by causing frustration and loss of faith in the brand. Therefore, it
would be sensible to recommend giving importance to their logistics systems and their
delivery promises to these companies.

53
Lastly, the online supermarket service providers, which aims to serve more customers
similar to the one in Cluster 1, should take the fact that these individuals give
importance to ease of use of the mobile applications and websites into consideration.
They should pay attention to the fact that consumers from different generations have
different interest and knowledge level with the technology, and design their apps and
websites by taking this into consideration.

From the global aspect, it is a fact that the concept of online shopping has gained
importance during the case of Covid-19 pandemic. Due to some reasons like health
concerns and physical interaction, people all around the world tend towards online
platforms for shopping. Moreover, the economic conditions of emerging countries like
Turkey led consumers to be careful with their expenses, and to follow the campaigns
and discounts more while making purchases.

Another aspect to be pointed out is that there exists also an ethical and safety dimension
of this subject. All participants of this study demonstrated that security of their data
while making payments in online supermarket services is significantly important for
them. It is reasonable because due to the digitalization of our everyday lives, the safety
of our data is gaining more and more importance everyday as we share more data
online. In this regard, it is easy to say that online supermarket service providers should
also take these aspects into account and give the necessary attention to these subjects
in order to maintain a decent relationship with their customers.

54
REFERENCES

Aggarwal, A. G., & Yadav, S. (2020). Customer segmentation using Fuzzy-AHP and
RFM model. 2020 8th International Conference on Reliability, Infocom
Technologies and Optimization (Trends and Future Directions)
(ICRITO). [Link]
Aktan, E. (2018). Büyük Veri: Uygulama Alanları, Analitiği ve Güvenlik Boyutu.
Bilgi Yönetimi, 1(1), 1–22. [Link]
Carrasco, R. A., Blasco, M. F., & Herrera-Viedma, E. (2015). A 2-tuple fuzzy
linguistic RFM model and its implementation. Procedia Computer
Science, 55, 1340–1347. [Link]
Christy, A., Umamakeswari, A., Priyatharsini, L., & Neyaa, A. (2021). RFM
ranking – An effective approach to customer segmentation. Journal Of
King Saud University - Computer And Information Sciences, 33(10),
1251-1257. [Link]
Dursun, A., & Caber, M. (2016). Using data mining techniques for profiling
profitable hotel customers: An application of RFM analysis. Tourism
Management Perspectives, 18, 153-160.
[Link]
Gustriansyah, R., Suhandi, N., & Antony, F. (2020). Clustering optimization in
RFM analysis Based on k-Means. Indonesian Journal Of Electrical
Engineering And Computer Science, 18(1), 470-476.
[Link]
Huang, Y., Zhang, M., & He, Y. (2020). Research on improved RFM customer
segmentation model based on K-means algorithm. 2020 5th
International Conference on Computational Intelligence and
Applications (ICCIA). [Link]
Jiang, L. W., Duan, S. Y., Xue, Y., & Feng, W. X. (2011). Determination of customer
value measurement model RFM index weihts. Africal Journal of
Business Management, 5(14), 5572-5572.
Kim, S.-Y., Jung, T.-S., Suh, E.-H., & Hwang, H.-S. (2006). Customer segmentation
and strategy development based on Customer Lifetime Value: A case
study. Expert Systems with Applications, 31(1), 101–107.
[Link]
Liu, C. (2011). Customer Segmentation and Evaluation Based on RFM, Cross-Selling
and Customer Loyalty. 2011 International Conference On
Management And Service Science, 1-4.
[Link]
Martinez, R. G., Carrasco, R. A., García-Madariaga, J., Gallego, C. P., &
Herrera-Viedma, E. (2019). A comparison between fuzzy linguistic
RFM model and traditional RFM model applied to campaign
management. case study of retail business. Procedia Computer Science,
162, 281–289. [Link]
Maryani, I., & Riana, D. (2017). Clustering and profiling of customers using RFM
for customer relationship management recommendations. 2017 5Th

55
 International Conference On Cyber And IT Service Management
(CITSM), 1-5. [Link]
Mengen, A. (2009). Verfahren der Kundenwertermittlung: Darstellung und
Bewertung der Kundenwertmessung als Bestandteil des Marketing-
Controlling. Fachhochschule Koblenz - University of Applied Sciences.
[Link]
Ozkan, P., & Kocakoc, İ. D. (2019). Sağlık Sektöründe LRFM Analizi ile Pazar
Bölümlendirme. PPAD Pazarlama Kongresi.
Perera, W., Dilini, K., & Kulawansa, T. (2018). A Review of Big Data Analytics for
Customer Relationship Management. 2018 3Rd International
Conference On Information Technology Research (ICITR), 1-5.
[Link]
Rahim, M., Mushafiq, M., Khan, S., & Arain, Z. (2021). RFM-based repurchase
behavior for customer classification and segmentation. Journal Of
Retailing And Consumer Services, 61, 1-9.
[Link]
Rsquared Academy Blog. (2019). RFM Segments and Descriptions [Image].
Retrieved 2 June 2022, from [Link]
[Link]/2019/07/customer-segmentation-using-rfm-analysis.
Tripathi, S., Bhardwaj, A., & E, P. (2018). Approaches to Clustering in Customer
Segmentation. International Journal Of Engineering &
Technology, 7(3.12), 802-806.
[Link]
Tsoy, M., & Shchekoldin, V. (2016). RFM-analysis as a tool for segmentation of
high-tech products' consumers. 2016 13Th International Scientific-
Technical Conference On Actual Problems Of Electronics Instrument
Engineering (APEIE), 1-4. [Link]
Verdenhofs, A., & Tambovceva, T. (2019). Evolution of customer segmentation in
the era of Big Data. Marketing and Management of Innovations, 238–
240. [Link]
Wei, J. T., Lin, S. Y., Wu, H. H. (2010). A review of the application of RFM model.
African Journal of Business Management. 4(19), 4199–4206
Wu, J., Shi, L., Lin, W.P., Tsai, S.B., Li, Y., Yang, L., & Xu, G. (2020). An
empirical study on customer segmentation by purchase behaviors using
a RFM model and K-means algorithm. Mathematical Problems in
Engineering. 1–7. [Link]

56
APPENDICES

APPENDIX A: Questionnaire in Turkish

57
APPENDIX A

Questionnaire in Turkish

Sayın katılımcı,

Kullanıcıların online market servisleri konusundaki tercihlerini anlamak amacıyla

tasarlanan bu anketten toplanan yanıtlar, İTÜ Endüstri Mühendisliği bölümü
bünyesinde gerçekleştirilen lisans bitirme tezi çalışmasında kullanılacaktır.

Yanıtlarınız yalnızca bitirme projesi kapsamında kullanılacak olup üçüncü

kişi/kurumlarla paylaşılmayacaktır.
Sorularınız ve önerileriniz için cuce18@[Link] veya tiryakie15@[Link] mail
adresleri üzerinden bizimle iletişime geçebilirsiniz.

Yanıtlarınız için teşekkür ederiz.

Aylanur Cüce & Eda Tiryaki

Cinsiyetiniz nedir?
( ) Kadın
( ) Erkek
( ) Belirtmek istemiyorum/Diğer

Yaşınızı belirtir misiniz?

( ) 18-23
( ) 24-29
( ) 30-35
( ) 36-41
( ) 42-47
( ) 48-53
( ) 54+

Eğitim seviyeniz nedir? (Şu anda öğrenciyseniz okuduğunuz eğitim seviyesini,

değilseniz son mezun olduğunuz okulu seçebilirsiniz)
( ) İlkokul/Ortaokul
( ) Lise
( ) Önlisans/Meslek Yüksek Okulu
( ) Lisans
( ) Yüksek Lisans/Doktora

Medeni durumunuzu belirtir misiniz?

( ) Bekarım, ailemle yaşıyorum
( ) Bekarım, ailemden ayrı yaşıyorum
( ) Evliyim
( ) Boşandım

Çalışma durumunuzu belirtir misiniz?

( ) Öğrenciyim/Stajyerim/Part-time çalışanım
( ) Çalışıyorum (tam zamanlı çalışan)

58
( ) Ev hanımıyım
( ) Emekliyim
( ) İşsizim/İş arıyorum

Aylık bireysel gelir durumunuzu belirtir misiniz?

( ) 0-2500₺
( ) 2501-5000₺
( ) 5001-10000₺
( ) 10001-20000₺
( ) 20001₺ ve üzeri

Hangi sanal market uygulamalarını kullanıyorsunuz?

[ ] Getir
[ ] Yemeksepeti Banabi
[ ] İstegelsin
[ ] Migros Sanal Market
[ ] Trendyol Hızlı Market
[ ] A101 Kapıda
[ ] Macroonline
[ ] CarrefourSA online

Bu uygulamalarda en sık ürün sipariş ettiğiniz 3 (üç) kategoriyi işaretleyiniz.

[ ] Su/İçecek
[ ] Meyve/Sebze
[ ] Temel Gıda (baklagiller, makarna, un, sıvıyağ vb.)
[ ] Süt ürünleri/Kahvaltılık
[ ] Et-Tavuk-Balık ürünleri
[ ] Atıştırmalık (çerez, gofret, çikolata, cips vb.)
[ ] Kişisel Bakım/Kozmetik
[ ] Ev Bakım/Temizlik
[ ] Bebek bakım
[ ] Evcil Hayvan
[ ] Teknoloji/Elektronik

En son ne zaman sanal market uygulamalarından sipariş verdiniz?

( ) Gün içinde/1-2 gün önce
( ) 3-7 gün önce
( ) 1-2 hafta önce
( ) 3-4 hafta önce
( ) 1 aydan daha önce

Ne sıklıkla sanal market uygulamalarından sipariş veriyorsunuz?

( ) Haftada 4-5 kere/ Neredeyse her gün
( ) Haftada 1-3 kere
( ) 10-15 günde bir kere
( ) Ayda bir kere
( ) Daha seyrek

59
Sanal market uygulamalarını kullanırken bir siparişteki ortalama harcama tutarınız ne
kadardır?
( ) 0-100₺
( ) 101-200₺
( ) 201-300₺
( ) 301-450₺
( ) 451₺ ve üzeri

Sanal market uygulamalarını fiziksel market alışverişi yerine tercih etme

sebeplerinizi düşündüğünüzde, aşağıdaki etmenleri 1-5 arası skalada puanlayınız.
1 2 3 4 5
Zaman avantajı sağlaması ( ) ( ) ( ) ( ) ( )
Fiziksel çaba gerektirmemesi ( ) ( ) ( ) ( ) ( )
Fiyat karşılaştırma kolaylığı sunması ( ) ( ) ( ) ( ) ( )
Ürün/marka çeşitliliği ( ) ( ) ( ) ( ) ( )
Sanal alışverişe özel indirim kuponları olması ( ) ( ) ( ) ( ) ( )
Pandemide fiziksel temastan kaçınma imkanı sunması ( ) ( ) ( ) ( ) ( )

"1- Hiç Önemli Değil" "5- Çok Önemli" olmak üzere, aşağıdaki ifadelerde size
uygun gelen puanı işaretleyiniz.

Sizin için ideal bir sanal market uygulamasında fiyatlar ne kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market uygulamasında ürün/marka çeşitliliği ne kadar
önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market alışverişinde teslimat hızı ne kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market alışverişinde belirtilen zaman aralığında teslimat
yapılması ne kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market uygulamasında indirimler/kampanyalar (sepette

indirim, kuponlar, ücretsiz kargo vb.) ne kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

60
Sizin için ideal bir sanal market uygulamasında mobil uygulamanın/websitesinin
kullanım kolaylığı/tasarımı ne kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market alışverişinde kuryenin tutumu/davranışı ne kadar
önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Sizin için ideal bir sanal market uygulamasında ödeme çeşitliliği ve güvenliği ne
kadar önemlidir?
1 2 3 4 5
Hiç Önemli Değil ( ) ( ) ( ) ( ) ( ) Çok Önemli

Aşağıdaki tablolarda verilen her bir kriter için söz konusu sanal market
uygulamalarını değerlendiriniz.

Fiyatlar açısından:
Fiyatlar Fiyatlar Fiyatlar
Bilgim Ortal Fiyatlar
çok ortalamanın ortalamadan
Yok ama çok uygun
pahalı üstünde uygun
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

Ürün/marka çeşitliliği açısından:

Ürün/marka
Ürün/marka Ürün Ürün/marka
Bilgim Orta çeşitliliği
çeşitliliği çeşitliliği çeşitliliği
Yok lama ortalamadan
çok yetersiz yetersiz çok fazla
fazla
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

61
Teslimat hızı açısından:
Hiç
Bilgim Memnun Çok
memnun Ortalama Memnunum
Yok değilim memnunum
değilim
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

Belirtilen zaman aralığında teslimat yapılması hususunda:

Hiç
Bilgim Memnun Çok
memnun Ortalama Memnunum
Yok değilim memnunum
değilim
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

İndirimler/kampanyalar açısından:
İndirimler Ortalama İndirimler İndirimler
Bilgim
çok az/ dan az/ Ortalama ortalama- fazla/
Yok
yetersiz yetersiz dan fazla yeterli
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

Mobil uygulamanın/websitesinin kullanım kolaylığı açısından:

Uygulama Kullanı- Kullanı-
Bilgim Uygulama
çok Ortalama mı mı çok
Yok kullanışsız
kullanışsız kolay kolay
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

62
Kuryenin tutumu/davranışı açısından:
Hiç
Bilgim Memnun Çok
memnun Ortalama Memnunum
Yok değilim memnunum
değilim
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

Ödeme çeşitliliği ve güvenliği açısından:

Hiç
Bilgim Memnun Çok
memnun Ortalama Memnunum
Yok değilim memnunum
değilim
Getir ( ) ( ) ( ) ( ) ( ) ( )
Yemeksepeti Banabi ( ) ( ) ( ) ( ) ( ) ( )
İstegelsin ( ) ( ) ( ) ( ) ( ) ( )
Migros Sanal Market ( ) ( ) ( ) ( ) ( ) ( )
Trendyol Hızlı Market ( ) ( ) ( ) ( ) ( ) ( )
A101 Kapıda ( ) ( ) ( ) ( ) ( ) ( )
Macroonline ( ) ( ) ( ) ( ) ( ) ( )

63
CURRICULUM VITAE

Name Surname : Aylanur Cüce

Place and Date of Birth : Istanbul, 19.01.2001

E-Mail : cuce18@[Link]

EDUCATION :

• [Link]. : 2022, Istanbul Technical University, Faculty of

Management, Industrial Engineering
• [Link]. : 2023, Istanbul Technical University, Faculty of
Electrical and Electronics Engineering, Electronics and Communication
Engineering (Double Major)

64
65
 CURRICULUM VITAE

Name Surname : Eda Tiryaki

Place and Date of Birth : 21.10.1997

E-Mail : tiryakie15@[Link]

EDUCATION :

• [Link]. : 2022, Istanbul Technical University, Faculty of

Management, Industrial Engineering

66

View publication stats