102

Dissertação de Mestrado Mestrado em Estatística - UMinho

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Dissertação de Mestrado Mestrado em Estatística - UMinho
Page 2: Dissertação de Mestrado Mestrado em Estatística - UMinho

Dissertação de MestradoMestrado em Estatística

Professora Doutora Raquel MenezesDoutora Ana Moreira

Trabalho realizado sob a orientação de

Outubro de 2018

Page 3: Dissertação de Mestrado Mestrado em Estatística - UMinho
Page 4: Dissertação de Mestrado Mestrado em Estatística - UMinho
Page 5: Dissertação de Mestrado Mestrado em Estatística - UMinho

Acknowledgement

I thank my advisors, Professor Dr Raquel Menezes and Dr Ana Moreira for the trans-

mission of their knowledge and for the support provided in the execution of this thesis.

To all colleagues of Inovretail company, for all the help and advice, thank you very

much.

To my girlfriend, Sara Ferreira, who accompanied me tirelessly along this route. Thank

you for the encouragement, patience, love and understanding shown.

And my family, my big thank you.

i

Page 6: Dissertação de Mestrado Mestrado em Estatística - UMinho

ii

Page 7: Dissertação de Mestrado Mestrado em Estatística - UMinho

Abstract

In the current economic context, retail has been confronted with a significant increase

in competitiveness indicators that lead to the equation of optimization tools in the most

diverse business processes, namely in stock management.

The definition of stock distribution is a key element for the management of a business

in any retail segment. In the Fashion segment, this challenge is particularly sensitive

to be managed, given the rapid product lifecycle, the high heterogeneity of the type of

products between the various stores and the strong variability in demand for each of

the products by consumers.

In this context, there is a strong potential of using predictive tools to support the

decision of the business managers so that the management of the stock is conducted

in a more quantitative way and less based on empirical knowledge, as it often occurs

nowadays.

This project was focused on the problem of the first shipment of products to the stores,

typically referred to as the allocation. Thus, to solve this problem we have proposed

the development of algorithms in R language.

The solution developed was tested by an international fashion retailer, which, with the

its application, obtained measurable gains by reaching days of in-store coverage of up

to 3%, compared to the initial scenarios of traditional human decision making. The

proposed approach has a Business Rules component, which will allow users to create

a diversity of business restrictions, offering a more robust and flexible solution. This

was confirmed by the first tests performed with actual retail data, which allowed us to

conclude that some business constraints could, and should, be automatically included

as algorithm defaults.

Keywords: Allocation; Fashion Retail; Stock Intelligence; Supply Chain

iii

Page 8: Dissertação de Mestrado Mestrado em Estatística - UMinho

iv

Page 9: Dissertação de Mestrado Mestrado em Estatística - UMinho

Resumo

No contexto económico atual, o retalho tem sido confrontado com um aumento sig-

nificativo de indicadores de competitividade que levam a equacionar ferramentas de

otimização nos mais diversos processos de negócio, nomeadamente na gestão de stock.

A definição da distribuição de stock é um elemento chave para a gestão de um negócio

em qualquer segmento de retalho. No segmento Fashion esse desafio é particularmente

sensível de ser gerido, dado o rápido ciclo de vida dos produtos, a elevada heterogenei-

dade da gama de produtos entre as várias lojas e a forte variabilidade na procura de

cada um dos produtos pelos consumidores.

Neste contexto há um forte potencial de utilização de ferramentas preditivas para su-

portar a decisão dos gestores de negócio para que a gestão do stock seja conduzida de

forma mais quantitativa e menos baseada em conhecimentos empíricos, tal como hoje

em dia maioritariamente se verifica.

Este projeto foi debruçado no problema do primeiro envio de produtos para as lojas, a

alocação. Deste modo, para solucionar este problema recorreu-se ao desenvolvimento

de algoritmos em linguagem R.

A solução desenvolvida foi testada por um retalhista internacional de moda, no qual,

com a aplicação da solução se obteve ganhos mensuráveis ao atingir dias de cobertura

em loja de até menos 3% face aos cenários iniciais de decisões por decisão humana

tradicional. A solução desenvolvida possui uma componente de Business Rules, a qual

permitirá aos utilizadores criar uma diversidade de regras de negócio, oferecendo maior

robustez à solução. Tal foi confirmado com os primeiros testes efetuados com dados

reais de retalho, os quais nos permitiram concluir que algumas restrições do negócio

poderiam e deveriam ser incluídos automaticamente como defaults do algoritmo.

Palavras-chave: Alocação; Retalho de Moda; Stock Inteligente; Cadeia de Abasteci-

v

Page 10: Dissertação de Mestrado Mestrado em Estatística - UMinho

vi

mento

Page 11: Dissertação de Mestrado Mestrado em Estatística - UMinho

Contents

1 Introduction 1

1.1 Project Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Motivation and Objective . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Contextualization 7

2.1 Fashion Retail . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1.1 Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1.2 Supply Chain Management . . . . . . . . . . . . . . . . . . . . . 17

3 Related Work 21

3.1 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Business Process Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Allocation Plan - An Optimization Approach 35

4.1 Solution Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

4.1.1 Use Case 1: Baseline . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.1.2 Use case 2: Shipment at different phases than planned . . . . . . 41

4.1.3 Use Case 3: Shipping Gap . . . . . . . . . . . . . . . . . . . . . . 42

4.2 Additional Complexities of the Algorithm . . . . . . . . . . . . . . . . . 43

4.2.1 Coverage Stock . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.2.2 Product Sequencing . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.2.3 Stock Reserves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4.2.4 Assortment Overwrite . . . . . . . . . . . . . . . . . . . . . . . . 45

vii

Page 12: Dissertação de Mestrado Mestrado em Estatística - UMinho

viii

4.2.5 Quantities Cap . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.2.6 Gap-End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.2.7 Gap-Begin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.2.8 Allocation on Low Refreshness . . . . . . . . . . . . . . . . . . . 48

4.3 Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.3.1 Stock Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.3.2 Sales Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.3.3 Business Performance . . . . . . . . . . . . . . . . . . . . . . . . 53

5 Results 65

5.1 Exploratory Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

5.2 Algorithm Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.2.1 Parametrizations . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5.2.2 Algorithm Performance . . . . . . . . . . . . . . . . . . . . . . . 71

5.3 Business Perfomance Metrics . . . . . . . . . . . . . . . . . . . . . . . . 75

6 Conclusion and Future Work 79

Page 13: Dissertação de Mestrado Mestrado em Estatística - UMinho

List of Figures

1.1 Inovretail’s logo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Allocation Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.1 Portuguese Private Consumption Source: Oxford Economics . . . . . . . . . . . 7

2.2 Consumption in Retail Goods by Category Source: Oxford Economics . . . . . . 8

2.3 Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.4 Pre-Season of Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.5 In-Season of Supply Chain . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3.1 Continuous Review System . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2 Periodic Review System . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.3 Allocation Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.4 Product Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.5 Season in Fashion Retail . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Process Diagram Allocation . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.2 Use Case 1 - Forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.3 Use Case 2 - Forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.4 Use Case 3 - Forecast [Store 1] . . . . . . . . . . . . . . . . . . . . . . . 42

4.5 Use Case 3 - Forecast [Store 2] . . . . . . . . . . . . . . . . . . . . . . . 43

4.6 Stock Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.7 Complexity Product Sequencing (Plan FIFO - First In First Out) . . . . 45

4.8 Complexity Cap in store 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.9 Complexity Force Allocation in End-Phase Proximity . . . . . . . . . . 47

4.10 Complexity Allocation in Begin-Phase Proximity . . . . . . . . . . . . . 47

ix

Page 14: Dissertação de Mestrado Mestrado em Estatística - UMinho

x

4.11 Force Allocation on Low Refreshness . . . . . . . . . . . . . . . . . . . . 48

4.12 Cases - Difference between Initial Plan and Executed Plan . . . . . . . . 56

5.1 Monthly Sales History (Dresses & Shorts) . . . . . . . . . . . . . . . . . 66

5.2 Monthly Sales History (Polos & Pajamas) . . . . . . . . . . . . . . . . . 66

5.3 Weekly Stock History (SS16 (Spring-Summer 2016) - SS17 (Spring-

Summer 2017)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.4 Coverage of Stock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5.5 Gap - End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

5.6 Allocation in SS 18 - Products Category Shorts . . . . . . . . . . . . . . 72

5.7 Allocation in SS 18 - Store . . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.8 Assortment Refreshness - Season SS 18 . . . . . . . . . . . . . . . . . . . 75

5.9 Allocation Execution By Store - Season SS 18 . . . . . . . . . . . . . . . 76

5.10 Allocation Execution By Product Category - Season SS 18 . . . . . . . . 77

Page 15: Dissertação de Mestrado Mestrado em Estatística - UMinho

List of Tables

3.1 Summary Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.1 Development Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.2 Summary table of the Complexities/Parametrizations . . . . . . . . . . 49

4.3 Allocation - Assortment Refreshness . . . . . . . . . . . . . . . . . . . . 53

4.4 Input’s - Percentage of Executation Plan Metric (Store) . . . . . . . . . 54

4.5 Input’s - Percentage of Executation Plan Metric (Product) . . . . . . . 54

4.6 Output Executation Plan Metrics . . . . . . . . . . . . . . . . . . . . . . 55

4.7 Input’s - Difference between Initial Plan and Allocation Executated -

Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

4.8 Difference between Initial Plan and Allocation Executed - Case 1 . . . . 56

4.9 Input’s - Difference between Initial Plan and Allocation Executed - Case 2 57

4.10 Difference between Initial Plan and Executed Plan - Case 2 . . . . . . . 57

4.11 Input’s - Difference between Initial Plan and Allocation Executed - Case 3 57

4.12 Difference between Initial Plan and Executed Plan - Case 3 . . . . . . . 57

4.13 Output Difference between Initial Plan and Allocation Executed Metrics 58

4.14 Input’s - Percentage of Difference between Initial Plan and Allocation

Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

4.15 Dataset - Difference Plan . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4.16 Dataset - Difference Plan Store . . . . . . . . . . . . . . . . . . . . . . . 60

4.17 Dataset - Executed Plan . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.18 Dataset - Suggested Plan . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.19 Output - Difference between Executed Plan and Suggested Plan . . . . 62

4.20 Input’s - Range Max . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

xi

Page 16: Dissertação de Mestrado Mestrado em Estatística - UMinho

xii

4.21 Output Range Max . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

5.1 Summary of Monthly Sales History . . . . . . . . . . . . . . . . . . . . . 67

5.2 Summary Stocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

5.3 Summary about Product Category . . . . . . . . . . . . . . . . . . . . . 69

5.4 Execution realized . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.5 Comparison of allocation in SS 18 - Products Category Shorts . . . . . . 73

5.6 Comparison of allocation in SS 18 - Store . . . . . . . . . . . . . . . . . 74

5.7 Difference between initial plan and allocation executed . . . . . . . . . . 77

5.8 Range Max - Season SS 18 . . . . . . . . . . . . . . . . . . . . . . . . . . 78

Page 17: Dissertação de Mestrado Mestrado em Estatística - UMinho

List of Abbreviations

ASC - Agile Supply Chain

CRBE - United States Commercial Real Estate Services

CW - Central Warehouse

ETL - Extract Transform and Load

FIFO - First In First Out

GSC - Global Supply Chain

Informa D&B - Informa Data Base

JIT - Just In Time

QR - Quick Response

R - Release

RRP - Recommended Retail Price

SIT - Stock In Transit

SKU - Stock Keeping Unit

SOH - Stock On Hand

SONAE IM - SONAE Investiment Management

SOO - Stock On Order

SS - Spring-Summer

UC - Use Case

xiii

Page 18: Dissertação de Mestrado Mestrado em Estatística - UMinho

xiv

Page 19: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 1

Introduction

1.1 Project Framework

The present thesis was done in a business environment, more specifically at Inovretail,

with the purpose of obtaining the degree on Master in statistics.

Inovretail was founded in 2011 in the city of Porto. Since 2016 it has belonged to

the SONAE IM group (SONAE Investment Management). The main objective of the

company is to enable and maximizing their clients’ sales, thus turning into a high

performance company specialized in retail. In order to do so, it resorts to technol-

ogy, mathematical and statistical processes, developing prediction tools that allow the

retailers to optimize the commercial actions in their stores.

Figure 1.1: Inovretail’s logo

The theme of the thesis in question “Allocation” is inserted in the roadmap of the

“Inovretail’s Data Science” team. This team is focused in countless challenges and

projects, some of which are described in Figure 1.2. The allocation is inserted in the

(group of projects) projects group of “stock optimization” in the “predictive retail”

area.

1

Page 20: Dissertação de Mestrado Mestrado em Estatística - UMinho

2

Figure 1.2: Allocation Project

1.2 Motivation and Objective

When we make a purchase, we are far from imagining the gigantic machine that “feeds”

the thousands of stores spread out across the country or even around the world. The

operations of the Portuguese retailers are made of impressive numbers. The fleet of

trucks that transports the products from the central storage to the stores travels more

than two and a half million kilometres to make around one hundred thousand deliv-

eries. As so, it is a complicated system with multiple variables and hundreds of steps

and procedures that demand a complex articulation INESC-TEC (2016). Presently the

Page 21: Dissertação de Mestrado Mestrado em Estatística - UMinho

3

retail sector has come to know a new moment, with more well-informed and demand-

ing customers that seek quality at the most competitive price. Taking into account

the decisive weight that clients and suppliers have in this area, companies seek to an-

ticipate the market needs. So the need for creating and using an integrated solution

with the purpose of releasing resources and diminishing the spending allocated to low

value tasks is fundamental, thus allowing more availability for the creation of proximity

relationship between customers and suppliers.

In a scenario marked by a strong competitiveness, in which the consumer has several

options available to him, it is fundamental to satisfy its needs and expectations, or even

to go beyond them for a continued success of a retail company ARTSOFT (n.d.).

Technology has changed the way in which customers browse and shop. The technolog-

ical advances have brought new channels and ways of contact, which forces companies

to re-think their business strategies. In spite of all the variety of options, the customer

wants to have the same experience, regardless of the means of selected purchase/con-

tact, concept known as omni-channel. These forces, more than a simple change in the

operational process of the company, a change in mind-set. The client is found under

the focus of all the strategy Frankenthal (2017).

This thesis has, as a main motivation, the exploration of new ways of thinking in order

to respond to the problem of allocation where the retail companies fight constantly.

The allocation consists of the decision making of the first shipment of products of the

new season, this problem is a problem of the supply chain. The concern with the ques-

tion of allocation came as an answer to the change in the attitude of customers verified

throughout the years. The theme has not been fully covered yet by the researchers, so

there are still very few papers and research concerning this subject. Thusly the main

objective of this thesis lies in the creation of a process/methodology that optimizes the

allocation of products for the stores. After the development, the solution will be tested

in a real environment (fashion retailer).

After its execution the potential gains will be monitored through the process applica-

tion from the retail company doing the follow-up of the spring/summer 2018 season.

Page 22: Dissertação de Mestrado Mestrado em Estatística - UMinho

4

1.3 Problem Definition

Logistics have evolved at a very fast rate during the last few years. Some decades

ago it wasn’t even mentioned in the long term plans of many companies. Nowadays,

it is considered one of the most strategic and fundamental elements in all of these

organizations, being one of the strongest sources in the creation of competitive

advantages Waters (2003). One of the reasons is the fact that its concept has spread to

an integrated function, called supply chain management, which is responsible for all of

the optimization of the flow of goods and materials since its purchase from a supplier

until its delivery at the final destination Tan (2001). All the major retail chains have

great costs associated to the logistical operations. Due to the growing complexity in

logistic operations, there is a growing demand for the optimization of each decision

and each movement along the chain of supply in a way as to optimize the creation of

value as a whole.

In this context, the first shipments of new articles/goods (allocation) towards the

stores is one of the most crucial moments, given that it is at this time that stores

receive the merchandise that they will be selling in the first instance, and where the

first indicator of market sensitivity is set.

In fashion retail stores a considerable part of the articles has a short life span,

depending heavily on the season in which it is inserted. For this reason, the first

shipment becomes yet more preponderant, for there isn’t a lot of margin of adjustment

in the distribution of the first shipments. Furthermore, to the final customer, the

difference towards the competition is often not in the quality or price of the product,

but in the capacity the store has of having the right product available, with all the

desired colours and size variants.

It is necessary to optimize decisions in the chain of supply, since every week millions

of decisions are associated with the shipment of merchandise, a process referred to as

season planning. In this way, there is a greater susceptibility relative to the occurrence

of errors since there is no scientific rigor. Supported on the allocations plan designed

in the season planning phase. These decisions are made by operators, who take into

account innumerable business and logistics rules which revolve and take part of the

Page 23: Dissertação de Mestrado Mestrado em Estatística - UMinho

5

chain of retail and the extensive quantity of data volume. Thusly these decisions make

the allocation process slow and inefficient. These plans are set some time in advance

(9 months in fashion) and when applied it may no longer fully match the needs /

characteristics of the market, customers and products. The fact that the decision

making involves a great number of operators makes the process even more complex.

Another of the great challenges of this project consists in trying to keep products

for a greater time in central storage so that they are only sent to stores at the right

moment, meaning, when these have actual need for them. This challenge comes with

the objective of optimizing transfers to the stores and by doing so reducing the costs of

transfers. Furthermore, to optimize transfers the prevalence of products in the central

storage may cause the product not to be sent, by impulse, to stores where there is

less need than in others. This can help to avoid the problem of stalled stock in store

warehouses (one of the retailer’s most worrying problems).

In conclusion, we will be creating an intelligent process that sends right product to

the rigth store at they right quantity and right time, always taking in to account all

previously explained considerations.

1.4 Thesis Structure

In addition to the introduction, this thesis contains 5 more chapters according to the

evolution of the work over time.

In Chapter 2, a contextualisation of the problem is made revealing much of the fashion

flap.

In Chapter 3, the related work is described, presenting the state of art, that is, a bibli-

ography review on some of the articles that played an important role in the acquisition

of knowledge for the realization of this thesis. It also presents more business concepts

that served for a better understanding.

Chapter 4 presents the development of the solution. It starts with a description of the

solution through the presentation of three use cases. Next, some features that were

added to the solution are presented as well as the presentation of the metrics developed

until now.

Page 24: Dissertação de Mestrado Mestrado em Estatística - UMinho

6

Chapter 5 presents and discusses the results obtained after the implementation of the

solution.

Finally, Chapter 6 ends with presentation of main conclusions of the work developed

and changes to be considered in future work.

Page 25: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 2

Contextualization

According to Wakefield (2018), in the last 15 years more than 6.7 thousand million

euros of investment were channelled towards the retail sector, which corresponds to

more than 50% of the total investments spent in commercial real-estate in Portugal.

The same authors state that private consumption is one of the economic variables that

best reflects the current state of the retail market. They further state that until 2008,

this was one of the main growth enablers in the economy, having then been strongly

affected by the following crisis, as can be seen in Figure 2.1.

Figure 2.1: Portuguese Private Consumption

Source: Oxford Economics

From the year 2014 onwards, economic recovery has been made felt through the increase

in private consumption. In 2018 it is estimated that it will reach the millennium’s

7

Page 26: Dissertação de Mestrado Mestrado em Estatística - UMinho

8

maximum. During this year growth Portugal’s private consumption should match the

European Union’s average at 2.2%, reflecting more moderate growths during the rest

of the analysed period (Wakefield, 2018).

Figure 2.2: Consumption in Retail Goods by Category

Source: Oxford Economics

Through the analysis of Figure 2.2, we can see that there has been a change in purchase

patterns which is demonstrated by comparing the distribution of retail goods consump-

tion by category between 2007 (market peak), 2011 (year of the beginning of the crisis)

and 2017.

The distribution of expenses in 2017 reflects a contraction in consumption in the "food

service", "home living" and others, in opposition to the food and fashion sectors, that

spent the same year getting a greater proportion of family income. The replacement

of restaurant business expenses for a greater consumption of food goods is a natural

effect of the crisis (Wakefield, 2018). We can see on Figure 2.2 that this indicator has

stagnated, which reflects a structural change in habits.

The increase in fashion expenses between 2007 and 2017 is surprising, taking into ac-

count that this category usually suffers during times of crisis. The strength of the retail

operators in this sector as well as a great amount of offer from low-cost clothing chains

Page 27: Dissertação de Mestrado Mestrado em Estatística - UMinho

9

may explain this tendency. For a long time, we have been witnessing a change in sales

policies from retailers, which catch the attention of consumers challenging them with

very diversified promotions and discount (Wakefield, 2018).

At the end of December 2017, there were in Portugal about 62845 business entities

dedicated to the retail sector. This sector represents around 13,6% of the total number

of business entities, data sent by Informa D&B (DB, 2018). According to the statistics

bulletin of 2018 published by the Banco de Portugal, the retail sector may be split

between the trade of food, beverages and tobacco (44.84%) or the same of non-food

products (55.16%). The fashion industry is inserted into the latter, representing about

7.30% of the total retail (Banco de Portugal, 2018).

The retail sector is in a constant state of change having its rhythm become more and

more accelerated during the last few decades. The industry stopped having a dedi-

cated focus on purchasing, meaning, reaching an ideal variety of products, to a broader

approach, where the consumer is at the centre of all the spotlights. In other words,

retailers dedicate a great percentage of their time trying to reach the ideal variety of

products, always taking into account the consumer as its primary focus (Zentes et al.,

2007). Nowadays, retailers face two major challenges regarding about management of

the supply chain, an increase in competition and the greater expectations from con-

sumers. These challenges are fuelled by five great tendencies: globalization; technology;

sustainability and safety; recession and prosperity. They create an increasing pressure

over retailers in order to maintain their success and to meet necessities from the con-

sumers without increasing costs.

Given that the role of the consumer is increasingly important in the supply chain, the

need to optimize retailers main actions becomes fundamental. For instance, selecting

the width and depth of the product range, planograms for the shelves, the allocation

are important aspects of the supply chain that retailers should be aware of (Hübner,

2011).

At the core of the issue is a planning paradigm, that synchronizes financial plans, store

plans, mandatory product plans and product range plans in order to respond to the

consumers need and to promote their loyalty. Therefore, the range of products can

be personalised based on each store’s exclusive attributes and using those reaching an

Page 28: Dissertação de Mestrado Mestrado em Estatística - UMinho

10

ideal price for each product in such a way as to maximize sales (Hübner, 2011).

2.1 Fashion Retail

In the past two decades, deep changes have occurred in the fashion industry, on a global

level, which originated an exponential increase in the complexity of the supply chain

management. In developed countries, as well as in developing ones, companies are

currently selling to the theoretical economic markets of perfect competition “business-

to-consumer” and selling increasingly more to the global supply chains (Global Supply

Chains – GSC) “business-to-business”, which are regulated by predominantly external

and global companies (Kaplinsky, 2005).

According to Gereffi and Memedovic (2003), the production of fashion products has

always been and shall continue to be an activity associated with intensive hand-labour

and low requirements in investment capital. This sector is often linked to poorer coun-

tries or developing ones. During the 70’s, 80’s and 90’s, there was a great number of

companies that moved their factories to countries where labour work is cheaper. They

moved, essentially, the productive tasks of less added value, in an attempt to main-

tain or regain competitive advantages. In the 90’s, southeast Asia was the geographic

area with the greater impact in this dynamic, including countries like China, India,

Malaysia, the Philippines and Indonesia, among others. In order to maintain its com-

petitiveness, the remaining retailers started doing their sourcing increasingly more in

Asian and Eastern European countries (Jackson and Shaw, 2000; Bruce et al., 2004).

2.1.1 Supply Chain

In order to know all of the supply chain management it is important to firstly know a

supply chain. Figure 2.3 illustrates the different stages of a typical supply chain of a

retailer dedicated to the fashion sector.

Page 29: Dissertação de Mestrado Mestrado em Estatística - UMinho

11

Figure 2.3: Supply Chain

In the business retail sector, the supply chain is divided into two stages, pre-season and

in-season, as illustrated by the previous figure. In the pre-season phase, we have stages

like design and product production and then their shipment towards the warehouse. In

the in-season phase, products are sent to the different stores arriving in this way to the

consumers.

Pre-Season

Analysing in more detail the pre-season stage illustrated in Figure 2.4, we can see that

designers are found at begin stage of the pre-season stage.

Figure 2.4: Pre-Season of Supply Chain

Along the times tasks dedicated to the pre-season phase have suffered several changes.

In the past, suppliers outlined the collections with the creation of products and its mak-

Page 30: Dissertação de Mestrado Mestrado em Estatística - UMinho

12

ing hoping that in the end retailers would buy those products. With the information

gathered from the points of sale, retailers started holding privileged information and

decided to incorporate stages that were up until then performed by the manufacturers.

In this way, retailers started to develop their own product creating competition against

the manufacturers that naturally held product development teams. The supply chain

in fashion retail has stopped being guided towards business and started being guided

towards the final consumer. Thus, nowadays, the design task started being made by

retailers which hire a series of designers, suppliers and manufacturers with the purpose

of drawing, producing collections based on the information collected at the sale points.

A current phase of pre-season in the supply chain starts by making studies about the

previous collection and the target consumer. Following these studies, a new collection is

designed. Thusly, the process starts with the elaboration of studies about the presented

collections in fashion shows (trade shows) in the world’s main fashion capitals (Paris,

Milan and New York), or with studies about the street tendencies, movies, sports and

media events on a global scale. Next, the sketches are drawn by the designers according

to previously placed studies.

Due to great geographic distances, intermediary companies have appeared between re-

tailers and the manufacturers of clothing. Since production has shifted and is now

spread among several different countries, the responsibility to explore to its maximum

imported quotas was left to companies specializing in complex supply chain manage-

ment.

Even faced with these difficulties, the economies resulting from cheap labour justified

the effort and risk that retailers were subjecting themselves to. The search and compe-

tition in developed countries demanded that the proposal of value had to be improved.

The sourcing of articles in developing economies was the solution found by retailers

in order to improve the offer through the reduction of sale prices to the public and

simultaneously increase the levels of profitability.

Starting from a generalist division of articles between fashion and basic, we have been

witnessing a separation in production with the first one being produced in a geographic

location near the target markets, and the basic ones in places that are further away with

longer lead-times but where you can get greater scale economies and consequently a

Page 31: Dissertação de Mestrado Mestrado em Estatística - UMinho

13

smaller price cost (Jackson and Shaw, 2000). In the same study, the authors also point

out to an example of a fashion manufacturer that has hired 3 companies for the same

product: one with a high lead-time (90 days) but with low costs in the Asian continent,

one with a medium lead-time (21 days) and with average costs in Latin America, and

a final one with a reduced lead-time (3 to 5 days) but with high costs in the target

production market. Being that what normally happens is that in fashion, in the case

of the product being a success, retailers will resort to the nearest place where they can

manufacture more of the same.

In the last decade, we have been seeing a conflux of interests between retailers, who

want to postpone the order as much as they can, and manufacturers who want to plan

its production and anticipate the placement of orders as early as possible. In spite of

retailers not wanting to compromise with purchase orders 25 to 40 weeks before the be-

ginning of the collection, we have been noticing that they commit to a certain volume of

production, meaning, they assure one manufacturer of the usage of a certain percentage

of its production capabilities. Instead of buying the article beforehand, they buy the

factory’s productive capability during a certain time of the year. This commitment has

come to solve part of the problems manufacturers used to have, since in the first two

months before the beginning of the collection they received numerous requests for or-

ders and its production capacity was limiting the acceptance of these orders. With the

promise of the purchasing of the product, companies can plan the production before-

hand and adjust its productive capacity, recurring frequently to outsourcing to other

manufacturing companies in order to be able to respond to and quantities that exceed

its installed capacity.

The last pre-season phase, that is, the last stage in sending the finalized products to

central warehouse may compromise the consumer’s level of service, as logistics guaran-

tees the moving of the product between several entities, throughout the chain of value,

effectively, maintaining its integrity. Normally, these transportation costs represent in

average about 1/3 of the cost of the supply chain (Pedro, 2015), with the possibility of

rising up to 10% of the total cost of the product (Rodrigue Jean-Paul, 2017), depending

on the context and some variables.

Page 32: Dissertação de Mestrado Mestrado em Estatística - UMinho

14

In-Season

Upon the product’s arrival at central warehouse and with the beginning of the season,

the in-season of the supply chain stage begins, represented by Figure 2.5.

The orientation towards the final consumer that retailers brought to the supply chain

has revealed flaws that until then had been undervalued in the supply chain. Since

1980, retailers have been struggling with the problem of the long lead-time (since the

conception of the article until its placement in the store) and the risk associated with

this. Aware of the current level of sophistication of the final consumer, its avidity for

novelty and progress, the retailers know that the answer to the consumer’s solicitations

is highly hampered with such a long lead-time which makes stocked articles become

rapidly obsolete throughout the chain.

Figure 2.5: In-Season of Supply Chain

Fashion products are segmented according to the class of products that the retail chain

focuses on. According to the different types of products that make up the retail chain,

the procedures for sending them will be different. There are mainly three types of

products:

Page 33: Dissertação de Mestrado Mestrado em Estatística - UMinho

15

• Fast fashion products – approximately 8-10 weeks of store life;

• Seasonal products – approximately 24 weeks of store life;

• Annual/permanent products – basic articles – available all year.

In a typical supply chain of the fashion industry, the in-season phase is divided into three

stages, adjustment of the previously defined plans in the pre-season stage, the allocation

process and lastly the replenishment process. Initially an adjustment to the planning

is elaborated in the sense that there is a great deal of time between its elaboration

(usually 9 months before) and the beginning of the season. During this time, there will

have been delays in the production of articles so that they are not yet available in the

central warehouse, or consumer preferences have changed, and consequently the order

of shipment of the products needs to be adapted. Still, at this stage, some targets are

adjusted that were also planned in the pre-season, such as future performance targets

for sales, stock, and other financial metrics.

With the advances in technology retailers started processing information coming from

their points of sale, realizing that the proposal of value made to the consumers should

not be restricted to price. Retailers realized that the consumer valued other factors

beyond price, normaly quality, level of service, uniqueness and novelty.

Allocation

With the elaboration of all the planning of the season done months before the begin-

ning of the season, which typically happens a few weeks before the beginning date of

the season this is revised proceeding to some changes, it gives if the next step, the

allocation. The allocation is the "first shipment of each product to each store". To do

this, the allocation takes into account the quantities of products in each store and will

then guide the delivery of the warehouse products to the stores according to the needs

of the products. Therefore, the allocation is about making product-specific shipping

decisions, that is, determining the products that need to be shipped to specific stores,

depending on their needs. Moreover, the exact quantities for the shipment need to be

calculated the just described supply chain process is the focus of this thesis.

There are two main strategies adopted globally by the retailer for the process of prod-

Page 34: Dissertação de Mestrado Mestrado em Estatística - UMinho

16

uct allocation to stores, Push and Pull distribution strategies, and the most applied

strategy in this process is the first. In the case of a Push distribution strategy, the prod-

ucts are allocated to the stores according to the historical demand for similar products,

the stock position of the warehouse and the stores. While the Pull strategy, product

shipping orders are generated at the store level based on the search data collected at

the point of sale terminals (Mcgraw, 2002).

As retail operations become more complex and information systems are more preva-

lent, retailers opt for the Pull distribution strategy instead of Push. However, the

Push strategy cannot be used by all retailers as it is difficult to predict sales for each

store. This can only be achieved by implementing sophisticated technology in terms

of software and hardware, which is often an expensive proposition. The retailer with

less sophisticated forecasting and information systems prefers to use the Push strategy

(Mcgraw, 2002).

The quantities allocated to each product and to each store are generally derived as

the difference between the permanent stock level and the current stock level. This is

termed as the amount retracted. The permanent stock level is the maximum stock level

that should be stored in the store for that product (Mcgraw, 2002).

Allocation transfers that typically occur between the warehouse and the stores are per-

formed on fixed days of the week or even during certain fixed periods. The delivery

frequency of the warehouse is also predetermined, which facilitates an effective under-

standing between the stores and the warehouse. Generally, products are shipped in

packs, set of sizes and colours of the product in question, to avoid the lack of availabil-

ity of the product to the preferences of the consumer.

The allocation process will still be explored in more depth in the next chapter.

Replenishment

After the products have been allocated to the stores, the Replenishment process comes

in. This process has an objective to maintain the stock position with frequent deliveries

of the warehouse to the stores until the end of the season, guaranteeing the level of

service to consumers. After the previous allocation process has taken place, it is pos-

sible to use more precise forecasts since the preferences of the consumers for season’s

Page 35: Dissertação de Mestrado Mestrado em Estatística - UMinho

17

products are already known. In this way, one may determine the correct replenishment

to maintain an optimum flow of stock between the warehouse and the stores.

In the stock planning process, the ideal stock levels are established, which must be

maintained to meet the expected levels of consumer services.

This process has two phases until a warehouse shipment to the warehouse is made.

First, the security amount of a particular product is defined by the retailer. This

amount allows the replenishment to be activated, in the event that the stock of a cer-

tain product is less than that quantity, it is activated the replenishment of that product

for the store in question. In a second phase, it is decided which is the maximum quan-

tity that the retailer intends to have in the store. Therefore, the order quantity of the

product will be the amount required for the current quantity to reach the maximum

quantity defined for that product.

It is of great importance to mention that for many supply chains during the in-season

phase one only gives the process of replenishment.

2.1.2 Supply Chain Management

The efficiency of a supply chain is a critical factor for the success of a company in the

fashion retail business. According to Hines (2004), in specialized global retain, the ones

that compete are the chains of supply and not the companies. Several approaches have

been made in a way to improve the supply chain, following concepts like Just-In-Time

(Bruce, Daly, and Towers, 2004), Agile Supply Chain (Christopher et al., 2004b; Bruce

et al., 2004) and Quick Response systems (Giunipero et al., 2001; Fernie and Azuma

2004). It matters to define and be able to tell apart each of these concepts.

The Just-In-Time (JIT) concept was born in the production area, more specifically

in the Toyota production line. This concept is also known as the Toyota Production

System, and aims to reduce costs related to the supply chain, by satisfying the demand

at any certain point, without creating an accumulation in stock at any point. It is a

concept which is guided for business and cost reduction and not for the final consumer

(Barnes and Lea-Greenwood, 2006). The implementation of the concept of JIT in a

supply chain makes it more lean, meaning, a chain developed while paying attention to

the reduction of any waste, including time and inventory, properly sized and levelled

Page 36: Dissertação de Mestrado Mestrado em Estatística - UMinho

18

in order to satisfy any existing needs.

The concept of Agile Supply Chain (ASC) is a concept that describes shorter supply

chains, more flexible, and guided towards real demand in detriment of the predicted

demand, typical approach of traditional supply chain management. An agile chain is

capable of using market knowledge, to process it internally and managing to explore

opportunities that may appear in volatile markets.

The Agile Supply Chains share updated information from the points of sale so that all

of it is synchronized and so there is no need for intermediary inventories (buffers). This

integration of information systems allows the shortening of response times and a more

efficient usage of resources all throughout the chain.

Lastly the Quick Response (QR) is born in a time when the traditional markets of

clothes production were losing margins and orders to emergent geographic regions with

cheaper labour and access to raw materials with cheaper prices. The Quick Response

has come to smooth the situation up, narrowing relationships between retailers and

manufacturers (Fernie and Azuma, 2004). It was considered by Christopher et al.

(2004a) as an evolution of the concepts of Supply Chain Management and Agile Supply

Chain. This tool defends the concept of minimizing de volume of pre-season purchases,

replacing them by more frequents points in times for making purchases, smaller in

volume and in in-season regime so that the domestic products may reacquire competi-

tiveness through the reduction of response times and flexibility.

Barnes and Lea-Greenwood (2006) defend that all three, JIT, ASC and QR, may be

considered supply chain management concepts guided towards business and not to-

wards the consumer.

Historically the power of the supply chain was found concentrated in the manufactur-

ers, and as production shifted towards distant geographic areas the JIT application

appeared with the purpose of alleviating inventories in the long supply chain. The

power has thusly converged around retailers, who now had a broader range of suppli-

ers for their sourcing. With the local manufacturers losing their order numbers AGS

and QR appeared, giving them a new competitive advantage based on the velocity of

response. Nowadays, the power of the supply chain has shifted towards consumers,

who have become more sophisticated and demanding, in constant change and progress.

Page 37: Dissertação de Mestrado Mestrado em Estatística - UMinho

19

Consequently, the concept of Fast Fashion has emerged, which does not preconize the

use of JIT, AGS or QR, but the simultaneous action of the three.

Thus, the cross between the long and inflexible supply chain of the apparel industry,

with the current state of the fashion sector, demanding with the rotation of store arti-

cles and oriented to the massive customization of its products. This led to a deadlock

leading to the creation of a series of different approaches focused on the management

of the supply chain and guided leading to the previously mentioned concept.

Fast Fashion is a business strategy that seeks to reduce the processes involved in the

purchase cycle and find inferior lead-times for the placing of articles in stores, in such

a way as to satisfy the demand by the consumer at its maximum point. Just like

Birtwistle et al. (2003) defends, the success of a company is dependent on its capability

for being flexible both in design and in production, allowing a reaction to the latest

trends.

According to Doeringer and Crean (2006), the key ingredients of the Fast Fashion con-

cept are the capability of quickly identifying fashion trends and identifying potentially

popular items, facilitating their availability in store before the appearance of other ten-

dencies and/or other potential items.

In this way, it becomes easy to associate any of the supply chain management con-

cepts previously covered (JIT, ASC, QR) to this business strategy. It is not intended

to have an inventory throughout the chain, since it needs to be free to process any

new items and any given time. We want the chain to be agile and free of waste and

that its responses are fast and flexible, with reduced lead-times in the fashion articles

which are clearly affected by the current tendencies. The concept of Fast Fashion is

guided towards the consumer and subjugates to it the concepts guided for business.

The impact on the supply chain of the Fast Fashion concept was identified by Barnes

and Lea-Greenwood (2006):

• The power of retailers – retailers are not willing to hold in their warehouse large

quantities of merchandise and demand that suppliers have deliveries that are more

frequent, faster and in smaller quantities. These orders come closer and closer to

the season to which they are concerned and the orders in season become more

frequent;

Page 38: Dissertação de Mestrado Mestrado em Estatística - UMinho

20

• A change in partnerships and connections – there are two concurring demands

imposed by retailers: there can be no delay from the supplier’s behalf and these

must be able to deliver a great and modern variety of styles;

• Manufacturers under rising pressure – the manufacturers are under a great deal

of pressure in order to become more flexible and fast when answering to the

volatility of demand;

• Elimination of some of the stages in the supply chain – the development of the

product and quality control suffer restructuring with the implementation of the

Fast Fashion concept, becoming less bureaucratic, with simpler and more agile

internal processes;

• The appearance of Fast Fashion manufacturing regions – according to a study car-

ried out by Barnes and Lea-Greenwood (2006). Turkey has emerged as a great

supplier of Fast Fashion due to its culture and due to its privileged geographic

position inside Europe. Its ease of access to raw materials has given it a compet-

itive advantage over the remaining geographic areas, as its lead-times are more

reduced and their costs are inferior;

• Smaller volumes of production – with the volatility of demand and the willingness

of retailers to refresh their stores with new products every week, large volume

orders have ceased to exist;

• Logistic changes – the pressure on articles to arrive as fast as possible to stores, in

parallel with technologic innovations and advances in transportation, have made

air transport a recurring reality and shorter expedition times by sea.

Page 39: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 3

Related Work

3.1 State of the Art

The fundamental decision to be taken into account in stock management systems with

two level (warehouse – stores) is the stock distribution of the central warehouse for

each of the retail stores. Making more stock available at stores gives a higher level of

service for customer demand, but this also raises the costs associated with logistics and

stock transportation and increases the maintenance cost of the stock in the stores due

to the shrinking of the available warehouse space, which is usually more expensive than

that at the warehouse. Higher costs resulting from the transportation of additional

items to the stores, may increase the value or reduce the margin. Furthermore, the

immediate distribution of a great proportion of stock towards the stores increases the

stock unbalance between stores, meaning, keeping additional stock in the warehouse

allows offering an advantage, because the stock can be directed to the stores that need

it the most. That may potentially reduce investments and overall stock costs. Mean-

while the delays in the shipping, due to waiting for the right moment, may negatively

affect the levels of service given to the client. These facts are horizontal to retail and

are both present in the allocation process (first shipment) and in the replenishment

process (later shipments).

The way in which stock is allocated in order to give response to the demand from the

stores is a critical determinant in the complexity of the inventory models. The process

of allocation affects the level of service and the stores costs, which places the challenge

21

Page 40: Dissertação de Mestrado Mestrado em Estatística - UMinho

22

on a high complexity level and most of the academic approaches uses approximations

or a relaxing of the problems restrictions in order to obtain limits over the true costs

and benefits, as will be presented next.

In 1984, Erkip argues that in a distribution chain inventory usually represents the

largest share of the investment, so it is often necessary to identify the costs associated

with it. He further went on to say that the solution involves minimizing costs through

operations that control inventory levels, stock rotations, various logistic cost plots, in-

ventory maintenance costs and the total cost of the logistics system. It also points

out that the main decisions to be taken in the system as regards the inventory are:

ordering, which refers to the replacement of a retailer’s warehouse from its suppliers;

and allocation, the division of any inventory (and sometimes timing) withdrawn from

the warehouse to the individual stores.

Most of the inventory control models are inspired by the work of Clark and Scarf, in

which they assume the existence of only one shipment to the stores of the entire ware-

house stock after shipment replacement, or then assume equal intervals of shipments

to stores, Clark and Scarf (1960).

The concept of "inventory balance" was first introduced by Clark and Scarf (1960). In

this work, the authors introduced the "echelon-stock" concept (sum of all the stocks of

the system) and characterized the ideal stock policies in a series of systems with finite

time horizon and without stock costs. The purpose of this work was to describe the

moment when all stock in retailers is in the same quantile of demand, that is, at the

same normalized level (the normalized retail inventory in a given interval is obtained

by dividing the difference between its inventory level, and the average search range,

by the standard deviation of the search range). In the case of identical retailers, the

perfect "balance" means that the inventory level of each retailer is the same. In addition

to their analysis of serial systems, Clark and Scarf also discussed how their approach

can be modified to deal with tree systems, multi-branched systems, roughly, using the

so-called "balance assumption".

The "balance assumption" assumes the authorization to the warehouse to make negative

stock allocations for the retailers. Consequently, the risk-pooling effect of keeping stock

in the central warehouse, for later use in stock equilibrium in retailers, disappears. This

Page 41: Dissertação de Mestrado Mestrado em Estatística - UMinho

23

also translates that only the sum of all stocks in the system, the "echelon stock", is of

interest for the decision making of the order in the warehouse. They further concede

that ideal stock allocation is achieved by examining the consequences only in the im-

mediate future.This perspective reduces the complex multi-period allocation problem

in a sequence of independent allocation problems over a single period. Later these

results were generalized to an infinite time horizon by Federgruen and Zipkin (1984),

for assembly systems by Rosling (1989), and for batch sorting by Chen (2000).

Later Axsäter, Marklund, and Silver (2002) show that "balance assumption" is less

adequate in situations of long order cycles and large differences between retailers in

terms of service requirements and demand characteristics, through the virtual alloca-

tion of orders for retailers, first suggested by Graves (1996). Most of the existing stock

allocation studies have assumed or examined types of balancing of the warehouse stock

division, for example, the allocation assumption in Eppen (1981) and Federgruen and

Zipkin (1984) and the run-out allocation rule in Jackson (1988).

The allocation assumption is based on the assumption that the order of entry is large

enough that the same probability of stock out of stock can be reached at each warehouse

or point of sale, ie in each allocation period there is enough stock to the warehouse dis-

tributes the product in sufficient quantity ensuring that the probability of stockout in

that same period is the same in all warehouses. For Eppen (1981), who instead of the

policies (s, S) or (Q, r), made a centralized study of systems of fixed length order cycles,

the warehouse order every m periods quantity enough to drive inventory position to a

level of y.The order policy, previously mentioned, is a policy (m, y), where the length of

the order cycle, m, is the focus of an optimization problem that includes consideration

of fixed order costs. Both Eppen (1981) and Eppen (1981) restrict the warehouse to the

role of a central ordering and shipping point with intermediaries, Central Warehouse

(CW).

In the multi-step literature (Schwarz, 1989) two reasons for the existence of CW are

identified. The first is transferring the risk of replacement time to the external sup-

plier and second is the shift of risk to retailers through periodic rebalancing of retail

inventories. Eppen (1981) call the first joint order effect ratio and the second deposit

effect. Since the former does not require the CW to hold stock, while the deposit effect

Page 42: Dissertação de Mestrado Mestrado em Estatística - UMinho

24

requires. These effects are later commented upon by Fine and Graves (1988).

Jackson (1988) considers an extension of the Eppen model (1981) where the warehouse

is allowed to have stock. The analysis focuses on a more elaborate allocation rule based

on the runout period of the warehouse stock, which explicitly takes into account the

risk-pooling benefits of keeping stock in the warehouse. This work proposes a policy

called Ship-up-to-S policy, where the warehouse makes shipments to reestablish the

stock position of each store to some pre-determined value, S, in each period for which

the warehouse has sufficient stock. Erkip (1984) considers the configuration of the same

problem as Jackson (1988) and proposes an alternative allocation scheme in which a

fraction of the warehouse inventory is allocated at the beginning of the order cycle and

the remainder is retained for an opportunity allocation at the end of the cycle.

Jönsson and Silver (1987) deal with a "depot less" system, analyzing the effect of allow-

ing redistribution of retailer inventories through side shipments shortly before the last

period of time in the order cycle. In a sense, its division of the entire order cycle into

two different time periods is similar to the "two-step allocation" heuristic of Axsäter

et al. (2002).

Two-step allocation policies are analyzed by Jackson and Muckstadt (1989) who ana-

lyze these policies for systems with similar N-retailers (they consider that retailers are

different, but still have identical cost parameters); with fixed intervals and the existence

of back ordering, relating to the thought of Erkip (1984).

Its purpose is to investigate analytically the risk-pooling effect of keeping stock in the

warehouse and conclude that given the balance of stock split that is withdrawn, the

expected inventory level of the second stock balancing interval decreases with increas-

ing of the amount of the first withdrawal interval.

With the Jönsson and Silver (1987) as a starting point, Schwarz (1989) analyzes the

value of risk-pooling on the lead time of the external supplier and concludes that the

value is particularly high in situations with significant variability of demand, long ware-

house lead time and short lead times for retailers.

This paper identifies two ways in which the warehouse can be used for risk centraliza-

tion. First, instead of allocating directly to individual retailers, the supplier sends the

unallocated stock to the warehouse, where it is later allocated to the retailers, thus

Page 43: Dissertação de Mestrado Mestrado em Estatística - UMinho

25

centralizing the risk on the supplier’s lead time. This type of risk centralization does

not require the warehouse to hold stocks. Second, the warehouse inventory can be

used among the reinforcements of the system to "rebalance" inventories of retailers that

may have become "unbalanced" due to variations in individual demand for them. This

repositioning of risk-pooling is examined in McGavin, Schwarz, and Ward (1993), who

consider a system with a warehouse and N-retailers identical and with lost sales.

McGavin et al. (1993) studied the decisions of warehouse shipments and store allocation

under the assumption of lost sales, that is, the approach that McGavin et al. (1993)

model is a distribution system with identical stores,matching times at zero for the or-

ders from the shops to the warehouse, a centralized control and periodic reposition in

the central warehouse facility. These authors consider that general allocation of stock is

based on four decisions: the number of shipments of stock from the warehouse (oppor-

tunity for stores to allocate stock), the time between shipments, the shipped quantities

and the division of stock sent to each store. The first three decisions are defined when

the warehouse is resupplied and the last one depends on the level of stocks present in

the stores. Their purpose was to minimize lost sales in store. The authors considered

an approaches that model two distinct allocation phases to store. They tried to deter-

mine the effective moment in time to allocate the stock, which would optimize their

goal of those two instances together with the quantities to be allocated, in a way as

to reach their goal. With the author’s developed work, they show that the best policy

for allocation is that which balances the stores stock (meaning, that maximizes the

store’s minimum stock). Still within this work, the authors have developed heuristic

policies under the assumptions that the number of stores be infinitely large and that

the store’s waiting time be zero. The heuristic policies that these consider for the

process of allocation was the heuristic 50/25. This heuristic consisted that at the first

instance 50% of the total stock is sent to the store and in the second shipment another

25% of the total stock is sent to the store. It must be pointed out that the 25% of the

stock remaining would serve for posterior shipments after allocation (first shipment),

meaning, “replenishment”. After the developed research, they came to the conclusion

that the choice of the amount of stock to send to each store and the distribution of the

stock are more important that the number of shipments.

Page 44: Dissertação de Mestrado Mestrado em Estatística - UMinho

26

In 1999, Ballou considered that much of the inventory management problems and,

consequently, allocation decision processing problems are due to perishable products,

because the demand for these is a sporadic event or a very unpredictable one. This

problem may cause the continuity of inventory from a “season” to another due to the

great uncertainty associated to the estimate of demand for these products. For such,

this author defends that when the demand and the “lead time” for supply cannot be

estimated with precision it is necessary to plan out a situation in which there will be

no available stock in order to satisfy the need of the store’s clients. In order to con-

trol these levels of stock, Ballou considered two systems for the first shipment process

(allocation): continuous review and periodic review. The first continuously monitors

the available quantities of each product, registering each transaction. The second one

checks the quantities periodically, in certain days of the week or days of the month

(Gonçalves, 2000).

In the continuous review system, also known as “Order Point” or “Level of Measure-

ment” (s,S), the existing quantity is reviewed constantly. Whenever the stock is under

a certain value s of a certain product category, an automatic shipment order is placed

in order to restock the necessary quantity of that product category until its maximum

value S. worth pointing out is that the shipment order is over new products referring

to that product category. This system can also be called Min-Max (Ballou, 1999).

For this type of policy, the risk is associated to the resupply period of the “lead-time”

(L), creating a small need for having a greater quantity of reserve stock so as to pre-

vent rupture while the resupply does not arrive (safety stock) (Gonçalves, 2000). The

continuous review system is illustrated by figure 3.1.

Page 45: Dissertação de Mestrado Mestrado em Estatística - UMinho

27

Figure 3.1: Continuous Review System

Source: Supply Chain Strategy & Analytics, 2016

As to the system of periodic review, also known as Revision Period or “Level of Mea-

surement” (R,s,S), this verifies the levels of stock at fixed time intervals, defined by R,

the review period. If at the end of a time period R the available quantity of a certain

category of products is inferior or equal to the minimum value s, the system generates

a new order of a sufficient quantity of new items in that category in order to replace

stock up to its filling level or maximum level S (Ballou, 1999), as is illustrated by figure

3.2.

Page 46: Dissertação de Mestrado Mestrado em Estatística - UMinho

28

Figure 3.2: Periodic Review System

Source: Supply Chain Strategy & Analytics, 2016

Ballou further points out that the choice between each of these systems depends of

various requirements. In the continuous review systems, it is necessary for the stock to

be continuously monitored, and that shipments can be placed at any time in order to

mitigate stock-outs. Since products usually have shipment order placement in different

time periods scale economies might be lost in the transportation, being so the author

defends that this system should be used solely for products of great value. As for the

periodic review system, this has advantages when there are a great number of products

to be controlled, because orders are all made at the same time, thus reducing the

shipment costs. Yet, there is a greater need for stock due to the fact that the stock

level is not being continuously monitored, making it necessary to protect not only the

supply lead time, but also the revision period in which orders are generated (Ballou,

1999).

For a better understanding of the literature review, the following table (table 3.1) was

elaborated with a summary of the works considered most important in the area. This

table was adapted from Caro and Gallien (2010).

Page 47: Dissertação de Mestrado Mestrado em Estatística - UMinho

29

Table 3.1: Summary Table

Decision Scope Time Horizon Shortage Model Stores

Ordering Withdrawal Allocation Finite Infinite Backorder Lost Sales Identical Non-identical

Clark and Scarf (1960) • • • •

Eppen (1981) • • • • •

Federgruen and Zipkin (1984) • • • • •

Jönsson and Silver (1987) • • • • •

Jackson (1988) • • • • •

Schwarz (1989) • • • • • •

McGavin et al. (1993) • • • • •

Graves (1996) • • • • • •

Axsäter et al. (2002) • • • • • •

The solution proposed in this thesis aims at differentiating itself from the previously

discussed ones by using at an optimization approach of multiple objectives (Multi Ob-

jective Optimization), supported on two basilar methodologies: a demand driven and a

business driven. Concerning the demand-driven method, and alike others, our proposal

includes, an effective forecast of demand that includes, not only an inference of demand

statistics, but its modelling through several steps, including the differentiated weight of

the weather, promotional and calendar impacts. For the business driven methodology,

it includes from the point of view of priority of items, restrictions, maximum devia-

tions, and other conditions that differ from the previously introduced. This is clearly a

framework that lacks tested and approved approaches in this ecosystem, notwithstand-

ing the other observed limitations from a point of view of simplification of the problem

that was witnessed in the remaining articles, and that is verified in the majority of the

remaining researched bibliography.

A large majority of alternative allocation solutions in the market focus on allocation

execution rather than prescriptive solutions of the nature in which it is inserted. This

limitation is particularly relevant in the Fashion retail segment, with its reality of short

product life cycles and strong demand variability, which tests the high differentiating

level of the proposed solution and its innovative character.

To sum up, the proposed solution will have the capability to optimize the initial allo-

cation plan by suggesting the rigth amount of product of product to push to the right

store and at the rigth time.

Page 48: Dissertação de Mestrado Mestrado em Estatística - UMinho

30

3.2 Business Process Model

In order to be easier to understand the approach taken to solve the problem, it is nec-

essary to know the fashion retail business adequately, thus presenting some concepts,

rules and methods of how this sector currently operates.

As already mentioned, the approach adopted led to the creation of an intelligent and

automated allocation system. For that, mathematical algorithms were used to tailor

store-to-store allocation, channelling the stock flow according to factors such as avail-

able sales space and store sales performance to ensure the right combination of items,

styles, colours and sizes, avoiding overstock and avoiding the existence of stockout.

By reducing costs and aligning stocks with the opportunity to sell each store, a retail

allocation solution minimizes the time it takes to accurately allocate the product and

maximize profit. To drastically reduce rebates and stock maintenance costs, the allo-

cation system must provide forecasting and demand delivery methods across a wide

variety of product types, using multiple sets of allocation and logistics rules. The sys-

tem must employ analytical methods that can automate the allocation effort based on

the available sales histories, the potential performance of individual stores, and the

direct channels to consumers.

Knowing the rules of logistics and business that define a retail chain is extremely impor-

tant, thus allowing to define choices and restrictions for the allocation process. These

rules allow the capture and retention of customers and markets, the efficiency of op-

erations management and the profitability of companies, resulting in large part from

the combination of some factors that have been taking the competitive environment of

companies much more demanding and complex.

Allocation decisions of a retailer are adjusted according to certain rules defined by

them. Following are some business rules typically defined by retailers in the fashion

industry:

• Restrict suggested quantities based on product and store properties;

For example: block the sending of basic clothes to stores in the district of Porto

or block the sending of shorts to stores located in the region of Guarda.

• Restrict suggested quantities based on price information and product margin;

Page 49: Dissertação de Mestrado Mestrado em Estatística - UMinho

31

For example: block the shipment of products with a margin lower than 30 % to

stores located in the Azores or Madeira archipelago.

• Restrict suggested quantities based on store-space metrics;

For example: block the sending of more than 500 underwear units to stores with

an area of less than 60 m2.

• Restrict suggested quantities based on weather facts;

For example: block the sending of knitwear with outside temperatures higher

than 20ºC.

In addition to business rules, there are also logistic rules that condition the allocation,

such as not sending more than 5000 units per week to a certain store. It should be

noted that most of these rules will be taken into account and will form an integral part

of the proposed solution, which will be discussed later in the next chapter. Figure 3.3

shows a diagram of the allocation process.

Figure 3.3: Allocation Scheme

The large of products is one of the characteristics important for the fashion retail sector.

In fact, as fashion trends involve various styles and colours, and combinations of those,

with variations in sizes, the range of products becomes enormous. Thus, the retailers

typically develop a merchandizing structure in order to organize the entire range of

products marketed. Figure 3.4 represents a typical retail structure of a fashion retailer.

It should be noted that the market structure can differ from retailer to retailer, with

Page 50: Dissertação de Mestrado Mestrado em Estatística - UMinho

32

more or fewer levels.

Figure 3.4: Product Structure

In the product structure considered for the development of the solution, the department

assumes the top of the same. This distinguishes the business that the retail chain is

focused on, for example, food department, textile department, electronic department,

among others. Since this project focuses on the fashion retail, the department of the

retail chain is the textile department. The level of the marketing structure that is

below the Department is the Business Unit. The business unit consists of the patent

brand of the stores. Depending on the size of the retailer, the retailer may have several

distinct brands operating in the same department. Next, is the Category, the product

category of the store can be separated, for example, into women’s clothing, men’s

clothing, among others. Subsequently, the next level of the article is the Typology, as

illustrated by Figure 3.4. The product typology consists, for example, of the knitted

sweater, among others. Next, ones finds the Base Unit level. The base unit of the

product further characterizes the product, i.e. blue knitted sweater. Next is the Style

level. Here is specified the colour, the texture, among others. Finally, the lowest level

of the market structure is the SKU (Stock Keeping Unity). At this level, the size of

Page 51: Dissertação de Mestrado Mestrado em Estatística - UMinho

33

the product is specified in the context of fashion flap.

In the context of fashion, is added the concept of Pack is also important. The pack

consists of the aggregation of SKU’s, for example, 1 article of size S; 2 articles of size

M and 1 item of size L.

Typically, in fashion retail, the collection of a season is fragmented into small collections,

which are sent to the stores at different times throughout the season. For example, in

the spring-summer season 2018, the entire collection is fragmented into small collections

based on the plans defined 9 months before the seasons, with some logic on the part of

the retailer.

Still, in fashion retail, the season is divided into phases. Generally, the period of these is

monthly and may vary from retailer to retailer. In each phase, the retailer still divides

it into review cycles, with the usual duration of are week, as we can see in the example

presented in Figure 3.5.

Figure 3.5: Season in Fashion Retail

In the illustrated example, the season in question has four phases, with each phase

having four review cycles with the exception of the last phase that has only three

review cycles.

Page 52: Dissertação de Mestrado Mestrado em Estatística - UMinho

34

Page 53: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 4

Allocation Plan - An Optimization

Approach

As can be seen in previous chapters, product allocation is a key competence for any

retailer, particularly for the fashion industry. Thus, it is of great importance to create

an intelligent and, most importantly, automated allocation system. This chapter will

focus on developing a solution for the optimization of the initial allocation plan.

This thesis aims to develop a specific model of automatic recommendations for stock

allocations of new products to the stores of a fashion retailer. This allows you to bal-

ance the planning component of the fashion collection with the consumer expectations

of each of the product categories, combined with specific constraints or motivations of

the business management teams. Only in this way it will be possible to build a solution

with the flexibility to adapt to different realities of several customers in the area of

fashion retail. The challenge in question assumes the use of optimization techniques

(e.g. Integer Programming) in which sales expectations will be developed. Further-

more, the identification of stock imbalances and, consequently, recommendations of the

volume and nature of the allocations needs to be performed. From the point of view

of the development of this component, we will elaborate the algorithm in R language

because of the high flexibility and adaptability to the data science in general, and the

optimization in particular.

Table 4.1 shows the main tasks that were part of the development of the proposed

solution.

35

Page 54: Dissertação de Mestrado Mestrado em Estatística - UMinho

36

Table 4.1: Development Activities

ActivityActivity Main Tasks

Effort

(%)

1 Preliminary

studies

Study and identification of Allocation processes

in the retail area.25 %

Preliminary study of Allocation model in the retail

area.

2

Specification

of the Solution and

Prototyping

of Models

Functional specification of the Stock Intelligence

solution in the retail area of fashion.25 %

Specification, prototyping and development in R

language of the algorithm components, namely in

Allocation models.

3

Application

Development

Integrated

Integration of algorithm components developed in

R language into Inovretail product.50 %

Development of evaluation metrics for the allocation

algorithm.

We now present the approaches performed in each of the activities described in Table

4.1 and respective deviations.

Preliminary Study

In the preliminary study phase, it is highlighted that through the wide range of In-

ovretail retail customers, it was possible to evaluate in detail the main functional and

technical specifications that a Stock Intelligence solution must respond in the fashion

retail environment. An essential point in this component was the survey of require-

ments that retailers look for in stock intelligence solutions, which were translated into

a set of functionalities base of the solution to be developed. Components such as the

identification of stakeholders in the stock management process, as well as the respective

mapping of decision processes in the allocation process can be highlighted here. Still

in the preliminary study component, but directed to the models, an academic review

of the existing use cases was made, but due to the lack of existing bibliography, the

result of this action was little relevant for the development of the solution.

Page 55: Dissertação de Mestrado Mestrado em Estatística - UMinho

37

Specification of the Solution and Prototyping of Models

The specification and development activities are performed cyclically by each of the

main functionalities. In the component of identification of data sources, Extract, Trans-

form and Load (ETL) and construction of the data models, different structures were

identified and developed to support the solution, namely information related to the

products (hierarchy, technical characteristics, functional groups, etc.), stores, transac-

tional information (actual purchasing data, stock, transfers, etc.), among others.

Application Development Integrated

In the algorithm component associated with the Allocation, the developed solution

consists of

1. Demand calculation module that combines a sales forecast (which includes inputs

from promotional campaigns, calendar and meteorology, as well as the modeling

of natural demand) with an evaluation of the stocks (physical and inbound /

outbound flow);

2. Integer programming techniques are used to allow a correct sequencing, appor-

tionment and distribution of the various products available to allocate to each of

the previously identified needs, taking into account the different rules of priority

and limitation;

3. The algorithm needs to include business constraints that can be configured to

restrict the proposed allocation, such as the possibility of considering mandatory

submissions, differentiating flexibilities of product shipments between stores, or

even whitelist or blacklist products that may be temporarily mandatory or ex-

cluded from the allocation process.

This task is being applied to a retail company in the fashion sector and, therefore,

there are still some limitations not fully accommodated by our algorithm, namely in

situations of limiting need and supply possibilities (ex: apportionment in the absence

of sufficient stock for cover minimum requirements of the plan, or supply behavior in

situations of execution outside the plan considered) that are still being analyzed and

Page 56: Dissertação de Mestrado Mestrado em Estatística - UMinho

38

that should be included in the algorithm soon.

In addition, in this Development Activity, a Business Rules component is also high-

lighted, which will allow users to create a diversity of business rules that allow the

solution to become more business driven. An example of a business rule can be to

block or to force a product into some store. Furthermore it was detected that some

business restrictions could and should be automatically included in the algorithm’s sug-

gestion.

Lastly, we created and developed a set of evaluation metrics to assess the performance

of the algorithm, that is, some measures for the allocation process, that will allow users

to keep track of the process over the course of the season.

In the following sections, we present a brief description of all the work developed. We

will initially introduce in this chapter some concepts and notions of the retail business.

Further on, we will present the algorithm approach and development. And, finally, we

will present the metrics that were developed.

4.1 Solution Description

In this section, the functional description of the algorithm will be presented, including

details on the data and the approach.

Before proceeding to the approach/solution method, it is necessary to introduce the

data involved in the process. For the scope of this algorithm, let us consider the

following datasets:

• Allocation Plan - Initial plan that indicates the lifetime of each product in the

new collection;

• Allocation - Execution of actual allocation (product shipment to stores) against

the initial allocation plan;

• Assortment - A list of products available for sale at each store;

• Forecast - Sales forecast for each store and each product;

• Stock (SIT + SOO + SOH) - SIT (stock in transit), SOO (stock on order) and

SOH (stock on hand);

Page 57: Dissertação de Mestrado Mestrado em Estatística - UMinho

39

• Products - Listing of all products;

• Products Pack - Pack composition.

A diagram of the allocation processes is presented in Figure 4.1, in which the developed

solution is divided into the following phases: Inputs, Process and Output.

Figure 4.1: Process Diagram Allocation

Inputs • Load Inputs

• Load Parametrizations

Process

• Data Transformation

• Calculate the needs of the stores

• Calculate the quantity allocation for each store

• Calculate the "best" moment for the allocation

Output• Suggestation Allocation Data

Several retailers execute the process of allocating products to the stores based on

the plans defined months before the start of the season. However, every week dozens

of market analysts and managers carry out revisions of the plan making only a few

adjustments to them. These adjustments are made based on their experience and

sensory aptitude. Therefore, in this chapter, as mentioned before, we aim to create an

intelligent and efficient process that will determine the ideal quantity to allocate at the

right time for a given store taking into account the needs of the stores. In addition,

applying this process will allow you to review the predefined plan more efficiently,

automated, and faster than the traditional process.

In order to fully understand the approach to the allocation solution and once this

strategy is applied in development, the default complexities of the algorithm will be

presented. For this, three use cases are presented.

Page 58: Dissertação de Mestrado Mestrado em Estatística - UMinho

40

4.1.1 Use Case 1: Baseline

This UC (Use Case) is our default case. It aims to suggest the best review cycle in the

phase of the season in which the product is planned. This use case strictly follows the

defined plan, such that it allocates the entire available allocation stock to the stores in

the same phase.

It considers the following assumptions: allocating the product in the phase in which it

is planned and sending it only to the stores that were defined in the assortment. The

algorithm thus calculates the ideal quantities to be sent to each store by assigning a

store ranking according to the needs of the stores by the product.

Let us exemplify this UC with an example: suppose we have a product of a given

category which is defined as having to be sent in phase 0, according to the plan. The

allocation process will specify the quantities to be sent to all the stores and determine

the best moment of the phase to do it. Figure 4.2 shows consumer demand in a

particular store for a specific category of products. Suppose this store has 80 units in

stock in this category. According to consumer preferences, the store will need it during

revision cycle 2. To be able to cover consumer needs, the product will be allocated in

the previous revision cycle, revision cycle 1, for the store in question.

Figure 4.2: Use Case 1 - Forecast

Concerning the calculation of the quantity to be sent to assortment stores, this is

Page 59: Dissertação de Mestrado Mestrado em Estatística - UMinho

41

based on a ranking of the needs of the store, so that stores that have greater needs

will receive more quantity of product. It should be noted that the quantities are

distributed so as to satisfy and safeguard the inventory to cover the needs of the

product category in each store during phase 0. In summary of this UC, the prod-

uct of the category in question will be allocated in the review cycle 1 for the given shop.

4.1.2 Use case 2: Shipment at different phases than planned

We now illustrate an increase in complexity compared to the previous use case, in

which it becomes possible to suggest the allocation of a product in a phase different

from the planned one. The use case 2 considers the following assumption: if the

product is allocated in a phase for a given assortment store, then it will be allocated

to the remaining assortment stores, even if they do not need the product category in

question. It should be noted that under this assumption, the additional complexity

allows allocating the product during any phase of the season.

As illustrated in Figure 4.3, suppose that a given product is defined in the plan as

having to be sent in phase 0 and it is also assumed that there are 50 stock units for a

particular store in the category products in which it is planned to be sent to it.

Figure 4.3: Use Case 2 - Forecast

Through the graphic visualization, it is possible to verify that for this store there will

Page 60: Dissertação de Mestrado Mestrado em Estatística - UMinho

42

be a need for this products category in phase 1, more specifically during review cycle

1. In this way, this product will be sent in review cycle 0 of phase 1. It should be

noted that, in the remaining stores that do not have needs until the end of phase 1,

the product will be allocated to them at the end of phase 1.

4.1.3 Use Case 3: Shipping Gap

Our last use case is characterized by the additional flexibility in sending the products.

In this use case, as in use case 2, it is possible to allocate the products in phases different

from those planned and, moreover, a product might be sent to stores of the assortment

in different phases, unlike previous use cases. In this case, we include the complexity of

the final delivery Gap according to the plan information, which provides the lifecycle

period for each product. As the end of the product lifecycle approaches, the gap will

force the shipment of products even if they do not have stock needs.

Let us analyse the example presented in Figure 4.4. We will assume that store 1 has 50

stock units of a given product category and that a given product of the same category

is defined in the shipment plan in phase 0 and we will also assume that it has 3 phases

life cycle (from phase 0 to phase 2).

Figure 4.4: Use Case 3 - Forecast [Store 1]

Through the graphical visualization, it is possible to verify that this category will lack

stock during review cycle 1 of phase 1 and, in this way, the product in question will be

Page 61: Dissertação de Mestrado Mestrado em Estatística - UMinho

43

allocated in the previous review cycle, review cycle 0 of phase 1.

Now we present an example of a second store in order to apply the final delivery gap,

according to consumer preferences given in Figure 4.5. We will assume that in this

store there are 80 stock products in the category of product in question. As can be

seen from the consumer preference, at the end of the product life cycle, this shop will

not need the product category, and consequently will not need the product, however,

due to the existence of the final shipping gap, the product will be forced to allocate in

review cycle 1 of phase 2.

Figure 4.5: Use Case 3 - Forecast [Store 2]

4.2 Additional Complexities of the Algorithm

Having presented the allocation process in a more simple way, as exemplified by the

previous use cases, it was then decided to include more complexity into the algorithm in

order to increase its robustness and to improve its performance. In addition, increasing

the complexity of the algorithm will also give us more flexibility in order to eliminate

certain limitations that initially existed.

The increase in complexity was intended to satisfy retailers in tinding a solution in the

allocation process for their retail chains. Although this solution was created and tested

having in mind the fashion retail chain, it was actually designed to be applied to any

Page 62: Dissertação de Mestrado Mestrado em Estatística - UMinho

44

sector.

4.2.1 Coverage Stock

Stock coverage is a very common feature for many retailers today. This consists in

determining how long a store continues to sell some product group.

As it is depicted in the previous use cases, a stock needs calculation was made that

combines the sales forecast with an evaluation of the stock in stock. For this calculation,

we need a parameter to be defined by the retailer that specifies a stock coverage value.

This parameter will be used to ensure that the allocated stock meets the store’s needs

during the coverage period. For example, if the stock coverage is 60 days, the calculation

of the needs and the quantity to be allocated meanwhile have to guarantee that there

is no expected stockout in the next 60 days. This example is illustrated in Figure 4.6.

Figure 4.6: Stock Coverage

4.2.2 Product Sequencing

Product sequencing is a feature very used in fashion retail. This sequencing will allow

choosing the products of a given product group that will be sent to each store when

needed. Product sequencing may be based on various strategies of the retailer, such as:

• Based on the First In First Out (FIFO) Plan - The product that is defined first

in the plan will be shipped first. For example, in Figure 4.7 is exemplified two

products of a certain product group. In the case of necessity of this product

group, the first product to be sent will be product 1, in which its life cycle starts

Page 63: Dissertação de Mestrado Mestrado em Estatística - UMinho

45

in the first place.

Figure 4.7: Complexity Product Sequencing (Plan FIFO - First In First Out)

• Based on Price - The highest priced product is shipped with priority;

• Margin Based - The product with the highest profit margin is shipped first in

comparison to lower margin products;

• Based on Stock Warehouse - The product with the highest stock in the warehouse

will be chosen in preference to products that have a lower stock.

4.2.3 Stock Reserves

After deciding the products to be shipped according to the sequencing rules pre-

established, it was necessary to develop the complexity of reserving stock for stores

that have not yet received a certain product. The need for this feature is important

since it is necessary to avoid sending the total available inventory of a certain product

by storing at a single store, that is, you should reserve stock for the remaining stores

defined in the assortment. This complexity is called inventory reserve for planned

allocations.

4.2.4 Assortment Overwrite

Assortment Overwrite Allocation functionality allows you to take into account the stock

that was allocated to stores that were not defined in Assortment. For example, let’s

look at a product 1 of type A in which it was defined only for store 1 and store 2. In

Page 64: Dissertação de Mestrado Mestrado em Estatística - UMinho

46

case there is an allocation of product 1 to store 3, the algorithm needs to be taken into

account. By applying this functionality, the algorithm will allow access to the quantity

of product that was sent out of Assortment.

4.2.5 Quantities Cap

After calculating the ideal quantities that each store will receive at the ideal time, the

retailer can restrict or better limit the amount of each product the store can receive.

This feature is called Cap. It should be noted that with the restriction of the quantity

of the products, the need for the typology in the store in question is satisfied beyond

upper limit of the quantities. Let us see the example of Figure 4.8, assuming a Cap of

3 units for all products of type A.

Figure 4.8: Complexity Cap in store 1

4.2.6 Gap-End

The complexity of the proximity of the end-phase was also developed. Our goal is to

force shipping and not wait until the end of the product lifecycle. This complexity will

force the submission weeks before the end of the product lifecycle. An example of this

complexity is illustrated in Figure 4.9. The product in question will not be allocated

when there are less than 4 weeks to the end of its life cycle, or rather, the product has

already been allotted at that moment.

Page 65: Dissertação de Mestrado Mestrado em Estatística - UMinho

47

Figure 4.9: Complexity Force Allocation in End-Phase Proximity

4.2.7 Gap-Begin

Another of the complexities added to the algorithm was the Gap-Begin. Before this

implementation, the algorithm was able to allocate products in any phase of the season

depending on the store needs. Due to the high interest of retailers in the matter of not

allocating product in advance, a new feature of the algorithm was created to allocate

from a distance from the beginning of the life cycle, this functionality is called Gap-

Begin. Let us look at the example illustrated in Figure 4.10. Assuming a Gap-Begin

of 4 weeks, this means that one can only send the product when one still has 4 weeks

up to the beginning of the life cycle.

Figure 4.10: Complexity Allocation in Begin-Phase Proximity

Page 66: Dissertação de Mestrado Mestrado em Estatística - UMinho

48

4.2.8 Allocation on Low Refreshness

One of the complexities also added to the allocation process was the ability of the

retailer to allocate a quantity greater than one level, ie each allocation of products

in stores should be greater than or equal to a value stipulated by the retailer. For

example in Figure 4.11, assuming that the retailer has defined the maximum of 10

packs for typology A, then this amount should be allocated each time to the store.

So, as initially only a total of 8 packs were allocated, the algorithm considering this

functionality allows the allocation of 10 packs and thus it adds automatically two more

products.

Figure 4.11: Force Allocation on Low Refreshness

Table 4.2 summarizes all features developed in this thesis, which we made available

to be integrated into the baseline allocation algorithm. It should be noted that these

complexities are configured by the retailer and thus they are also known as Strategies.

Page 67: Dissertação de Mestrado Mestrado em Estatística - UMinho

49

Table 4.2: Summary table of the Complexities/Parametrizations

Type Complexity DefinitionPossible

Values

BaseVe

rsion Revision Period

This parameter represents the period interval

between runs of the algorithm.

"week" or

"month"

Stock Reserves of

Planned Allocations

This parameter represents saving stock

for future allocations."Yes" or "No"

Force Allocation on

End-Phase Proximity

This parameter forces the sending when

it reaches a distance from the end of the

lifecycle if it has not yet been allocated.

Integer

Target Stock Coverage

This parameter represents the stock coverage

of the execution of the algorithm,

ensuring that there is no stock break

during that period of interest.

> 0

Com

plexities Product Sequencing

This parameter represents the priority

of the choice of products.

"Plan FIFO" or

"Price" or

"Margin" or

"Stock

Warehouse"

Assortment Overwrite

Allocations

This parameter allows the algorithm to

take account of non-Assortment allocations."Yes" or "No"

Permission Allocation on

Begin-Phase Proximity

This parameter allows the allocation at a

distance from the beginning of the lifecycle.Integer

Allocation Maximum

Quantity Limit (Cap)

This parameter represents the maximum

quantity of a certain product that can be

allocated in each execution.

> 0

Force Assortment on

Low Refreshness

This parameter represents the minimum

quantity that should be allocated in each

execution

> 0

4.3 Metrics

All retail chains, regardless of type of industry, have their own set of metrics for two

main reasons, helps the business to constantly evaluate its performance and allows

it to be a benchmark against the competition. The metrics are a set of measures

Page 68: Dissertação de Mestrado Mestrado em Estatística - UMinho

50

that evaluate retail chain results in relation to some type of reference, such as past

performance, store objectives, industry averages, among others. Metrics are also known

as "key performance indicators".

Defining metrics to measure retail success is not easy due to the prevalence of various

retail formats (such as supermarkets, hypermarkets, mono-brand retailers, department

stores, speciality chains, among others).

Competition in fashion retail is persistent and customers in stores expect immediate

availability of the products. But many retailers find it difficult to maintain accurate

inventory and proper allocation with typical retail stock accuracy rates ranging from

60% to 70%. Low levels of accuracy lead to loss of sales and consumer deception.

The Metrics solution transforms product data in real-time into analytical insight by

showing what is happening in business operations. This allows retailers to sell more

products and reduce the costs of their operations at the same time. As products move

from the supply chain to the store warehouse, store to shelf, shelf to the taster, and

finally to purchase, the identity and location of all products are constantly monitored,

providing to retailers visibility at the level of the final product and real-time analysis.

With the application of the metrics, it is possible to take comprehensive, real-time

analyzes that provide full in-store business transparency to key departments throughout

the retail organization, avoiding lost sales opportunities and dissatisfied consumers. In

summary, the development of the metrics has as main objectives:

• Optimize the supply chain - makes it possible to track each product with granular-

ity at the product level and transform demand visibility into allocation efficiency;

• The predisposition of the store - maintain an inventory accuracy of close to 100%,

avoid the lack of stock with fast replenishment, instantly locate lost products and

avoid losing them;

• Optimize the store - gain insight into visual merchandising, replace manual stock

counts to make your workforce more efficient, optimize the time and depth of

rebates, and make smarter stock-allocation decisions;

• Consistency - keep the same set of assets active, ship to stores based on current

inventory and use inventory visibility to improve marketing activities.

Page 69: Dissertação de Mestrado Mestrado em Estatística - UMinho

51

In summary, metrics enable retailers to have the ability to visualize the elasticity of de-

mand for products, gain store-level profitability information, increase sales by reducing

stockouts, and improving margins. The metrics developed were directed to the alloca-

tion process, where they can be distinguished into three groups, metrics for business

performance, stock and sales.

In the development of the thesis, a total of 11 metrics were implemented:

• Stock

– Stock Age

– Stock Refreshness

– Assortment Refreshness

• Sales

– Time to Sell Out

• Business Performance

– Percentage of Allocation

– Difference between Initial Plan and Allocation Executation

– Difference between Initial Plan and Algorithm Suggestion

– Percentage of Difference between Initial Plan and Allocation Executation

– Percentage of Difference between Initial Plan and Algorithm Suggestion

– Difference between Allocation Executation and Algorithm Suggestion

– Range Max

We now present in detail of each of the metrics described above, accompanied by an

explanation of the approach adopted.

4.3.1 Stock Metrics

Stock Age

Stock Age is a metric that represents the average time spent on store stock allocation.

The unit of measure for this metric is in days and it can be calculated in two dimensions,

Page 70: Dissertação de Mestrado Mestrado em Estatística - UMinho

52

per store or per product. With the use of this metric, it will be possible to evaluate

the average age of the allocated stock and thus arrive at the conclusion that a certain

product is obsolete or which stores are more efficient.

Stock Refreshness

As for the Stock Refreshness metric, this represents the percentage of stock ratios of

newly allocated products in each store, and thus the size of this metric is per store only.

This metric has as main objective to evaluate the refreshment of the stores, that is, to

quantify the new product stock in comparison with the total stock.

Assortment Refreshness

This metric represents the amount of new products that each store received as an

allocation in a given period of interest. This metric makes it possible to verify during

the period of interest the refreshment of the store by the new product. Currently, the

need for refreshment of the store is very recurrent in retailers, because the greater the

novelties in store, the greater the flow of consumers, that is, those visitors the stores

more regularly. This metric only involves the store size and it requires two parameters,

the period of interest ("day", "week" or "month") and the period (value of interest).

In case the retailer intends to measure the refresh level of the last week of a given

store, we have the period of interest parameter defined as "week" and the value is 1.

Assuming the current date of 15/02/2018, then the period of interest is (08/02/2018 -

15/02/2018). For example, according to the allocation of store 2 represented in Table

4.3, we conclude that three new products have arrived at the store in the previous week.

Page 71: Dissertação de Mestrado Mestrado em Estatística - UMinho

53

Table 4.3: Allocation - Assortment Refreshness

Store Product Date

2 40 04/01/2018

2 41 09/01/2018

2 42 09/01/2018

2 43 13/01/2018

2 45 03/02/2018

2 46 04/02/2018

2 47 10/02/2018

2 48 11/02/2018

2 49 12/02/2018

4.3.2 Sales Metrics

Time to Sell Out

The metric represents the average time of the total sale of the products allocated. By

applying this metric, one is allowed to assess whether certain products are obsolete or,

in the case of stores, if certain stores are more efficient at selling novelty products. The

dimensions of this metric are per store or per product. The unit of this metric is in

days.

4.3.3 Business Performance

Percentage of Allocation

The percentage of allocation plan allows to have the real-time tracking of the allocation

of the new products within a given season. By applying this metric it is possible to

have the actual percentage of allocation over any store or over any product and thereby

make decisions about the allocation process. It will be possible to check which stores

have the highest percentage of execution. To execute this metric, you need the dataset

Assortment and the dataset Allocation.

For example, if we fix store 1 and assume the current date as 05/02/2018.

Page 72: Dissertação de Mestrado Mestrado em Estatística - UMinho

54

Table 4.4: Input’s - Percentage of Executation Plan Metric (Store)

(a) Assortment - Store 1

Store Product

1 45

1 46

1 47

1 48

(b) Allocation - Store 1

Store Product Date

1 47 01/02/2018

Through the Table 4.4 it is verified that the execution percentage of the plan of store

1 is 25%. If we now fix product 45 and same current date.

Table 4.5: Input’s - Percentage of Executation Plan Metric (Product)

(a) Assortment - Product 45

Store Product

1 45

2 45

3 45

4 45

(b) Allocation - Product 45

Store Product Date

1 45 01/02/2018

3 45 15/01/2018

Based on information given in Table 4.5, it is possible to calculate the percentage of

the plan executed against product 45, the percentage is 50%. The retailer will have the

perception of the percentage execution metric per store and per product (any level),

through the visualization of Table 4.6.

Page 73: Dissertação de Mestrado Mestrado em Estatística - UMinho

55

Table 4.6: Output Executation Plan Metrics

Store Percentage Executation Plan

1 25%

... ...

Product Percentage Executation Plan

45 50%

... ...

Difference between Initial Plan and Allocation Execution

This metric represents the mean and median of the difference of the deviations between

the initial allocation plan and the executed one. This metric aims to evaluate the initial

plan, taking into account that this was done months before the start of the season. To

perform this metric, one needs the Initial Plan dataset and the Allocation Executed

dataset. To better understand this metric, we will divide it into three example cases.

The first case represents the situation when the allocation occurs before the planned

start date; the second case represents the situation when the allocation takes place

within the product life cycle; and finally the third case happens when the allocation

takes place after the planned end date. All three cases are illustrated by the scheme in

Figure 4.12.

Page 74: Dissertação de Mestrado Mestrado em Estatística - UMinho

56

Figure 4.12: Cases - Difference between Initial Plan and Executed Plan

For case 1, we will consider the information represented in Table 4.7.

Table 4.7: Input’s - Difference between Initial Plan and Allocation Executated - Case

1

(a) Initial Plan

Product Start Date End Date

45 10/05/2018 31/05/2018

(b) Allocation Executed

Store Product Date

1 45 01/05/2018

2 45 24/04/2018

The differences between the two plans of the previous allocations are represented in

Table 4.8.

Table 4.8: Difference between Initial Plan and Allocation Executed - Case 1

Store Product Difference

1 45 - 9 days

2 45 - 15 days

For case 2, we will consider the information represented in Table 4.9.

Page 75: Dissertação de Mestrado Mestrado em Estatística - UMinho

57

Table 4.9: Input’s - Difference between Initial Plan and Allocation Executed - Case 2

(a) Initial Plan

Product Start Date End Date

46 10/05/2018 24/05/2018

(b) Allocation Executed

Store Product Date

1 46 19/05/2018

2 46 15/05/2018

Thus, the differences between the two plans is given in Table 4.10.

Table 4.10: Difference between Initial Plan and Executed Plan - Case 2

Store Product Difference

1 46 0 days

2 46 0 days

Finally, we consider the Initial Plan and Allocation Executed represented in Table 4.11.

Table 4.11: Input’s - Difference between Initial Plan and Allocation Executed - Case 3

(a) Initial Plan

Product Start Date End Date

47 10/05/2018 27/05/2018

(b) Allocation Executed

Store Product Date

1 47 14/06/2018

2 47 17/06/2018

The differences between the two plans are calculated and the results given in Table

4.12.

Table 4.12: Difference between Initial Plan and Executed Plan - Case 3

Store Product Difference

1 47 18 days

2 47 21 days

Then, the output of the Difference Metric between the Initial Plan and the Allocation

Executed is summarized in Table 4.13.

Page 76: Dissertação de Mestrado Mestrado em Estatística - UMinho

58

Table 4.13: Output Difference between Initial Plan and Allocation Executed Metrics

Store Mean Median

1 3 days 0 days

2 2 days 0 days

... ... ...

Product Mean Median

45 -12 days -12 days

46 0 days 0 days

47 19.5 days 19.5 days

... ... ...

Difference between Initial Plan and Algorithm Suggestion

This metric represents the mean and median of the difference of the deviations between

the initial plan and the plan suggested by the algorithm. The approach adopted in

this metric was the same as that chosen in the previous metric, in which instead

of considering the execution plan, it is considered the suggestion given by the algorithm.

Percentage of Difference between Initial Plan and Allocation Execution

This metric represents the percentage of the difference in the deviations between the

initial plan and the executed allocation. The dimensions of this metric are two, being

per store or per product (Product Group).

Let us analyse at the example in Table 4.14, which represents the initial plan, the

assortment and the executed plan.

Page 77: Dissertação de Mestrado Mestrado em Estatística - UMinho

59

Table 4.14: Input’s - Percentage of Difference between Initial Plan and Allocation

Execution

(a) Dataset - Initial Plan

Product Start Date End Date

45 10/05/2018 27/05/2018

46 01/05/2018 15/05/2018

47 12/05/2018 27/05/2018

48 10/05/2018 27/05/2018

(b) Dataset - Assortment

Store Product

1 45

1 46

1 47

1 48

2 45

2 46

2 47

2 48

(c) Dataset - Assortment

Store Product Date

1 45 15/05/2018

1 46 20/05/2018

1 47 09/05/2018

1 48 14/05/2018

2 45 15/05/2018

2 47 19/05/2018

2 48 20/05/2018

In case the product is allocated according to the plan the difference is 1 (by planned).

Whereas, if the product is allocated off-plan, the difference is 0 (not by planned). And

if the product has not been allocated to the store in question, the difference is 2 (not

yet happened).

Assuming that the current date is 28/05/2018, the differences between the two plans

are given by Table 4.15.

Page 78: Dissertação de Mestrado Mestrado em Estatística - UMinho

60

Table 4.15: Dataset - Difference Plan

Store Product Date Start Date End Date Difference

1 45 15/05/2018 10/05/2018 27/05/2018 1

1 46 20/05/2018 01/05/2018 15/05/2018 0

1 47 09/05/2018 12/05/2018 27/05/2018 0

1 48 14/05/2018 10/05/2018 27/05/2018 1

2 45 15/05/2018 11/05/2018 27/05/2018 1

2 46 12/05/2018 30/05/2018 2

2 47 19/05/2018 10/05/2018 27/05/2018 1

2 48 20/05/2018 10/05/2018 27/05/2018 1

Thus, the differences between the plans of each store and each product are now given

in Table 4.16.

Table 4.16: Dataset - Difference Plan Store

Store Percentage By Planned Percentage Not By Planned Percentage Not Yet Happened

1 50 % 50% 0%

2 75 % 0% 25%

Product Percentage By Planned Percentage Not By Planned Percentage Not Yet Happened

45 100% 0% 0%

46 0% 50% 50%

47 50% 50% 0%

48 100% 0% 0%

Percentage of Difference between Initial Plan and Algorithm Suggestion

This metric represents the percentage of the difference of the deviations between the

initial plan and the plan suggested by the algorithm. The dimensions of this metric

are two: per store and/or per product (Product Group). The approach taken in this

metric is similar to the previous metric, the percentage of the difference between the

initial plan and the allocation execution metrics.

Page 79: Dissertação de Mestrado Mestrado em Estatística - UMinho

61

Difference between Allocation Execution and Algorithm Suggestion

This metric represents the difference of the deviations between the executed plan and

the suggestion given by the algorithm. By applying this metric it will be possible to

verify and identify errors in the current allocation process. The unit of measure for

this metric is in days. This metric has two dimensions: per store and/or per product.

We consider the example in Table 4.17, that represents the executed plan of products

1 and 2. While Table 4.18 represents the plan suggested by the algorithm.

Table 4.17: Dataset - Executed Plan

Store Product Date

1 40 07/04/2018

1 41 07/04/2018

1 44 07/04/2018

2 40 16/04/2018

2 42 16/04/2018

2 44 07/04/2018

Table 4.18: Dataset - Suggested Plan

Store Product Date

1 40 10/04/2018

1 41 10/04/2018

1 42 10/04/2018

2 40 22/04/2018

2 42 22/04/2018

2 44 10/04/2018

Thus, one can calculate the difference between the plan executed and the suggested

one which is given in the Table 4.19.

Page 80: Dissertação de Mestrado Mestrado em Estatística - UMinho

62

Table 4.19: Output - Difference between Executed Plan and Suggested Plan

Product Product Difference

1 40 3 days

1 41 3 days

2 40 6 days

2 42 6 days

2 44 3 days

It should be noted that the differences presented in Table 4.19 are positive. This leads

us to conclude that the algorithm suggests that the date is being executed.

Range Max

This metric represents the time interval between the first store and the last store in

which a particular product was allocated. The unit of measure of this metric is in days

and its size is at the product level. In case it has not yet been allocated to all the

stores that were defined in the assortment, the range max is given by the time interval

between the first store and the current date.

For example, if we fix two products, product 45 and product 46, we can calculate

the max range of both products through the allocation information and through the

assortment information of the two products (see Table 4.20).

Page 81: Dissertação de Mestrado Mestrado em Estatística - UMinho

63

Table 4.20: Input’s - Range Max

(a) Allocation - Product 45 and Product 46

Product Store Date

45 1 02/01/2018

45 2 02/01/2018

45 3 15/01/2018

45 4 16/01/2018

45 5 31/01/2018

45 6 28/01/2018

46 1 15/01/2018

46 2 15/01/2018

46 3 -

46 4 20/01/2018

46 5 -

46 6 25/01/2018

(b) Assortment - Product 45 and Product 46

Product Store

45 1

45 2

45 3

45 4

45 5

45 6

46 1

46 2

46 3

46 4

46 5

46 6

If we assume that the current date is 20/02/2018 based on information given in Table

4.20, we can conclude that the product 45 has already been allocated to all the stores

that were defined while the product 46 is still missing two stores (Store 3 and Store

4) to be allocated. Thus, the maximum range for product 45 is 29 days, while the

maximum range for product 46 is 32 days, as shown in Table 4.21.

Table 4.21: Output Range Max

Product Range Max

45 29 days

46 32 days

Page 82: Dissertação de Mestrado Mestrado em Estatística - UMinho

64

Page 83: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 5

Results

Finally, we will conclude with the performance analysis of the developed algorithm,

comparing it with the allocation made by the retailer in the spring-summer season 2018.

Through real data provided by an international retailer of fashion, it was possible to

analyze and evaluate the solution, namely a validation of performance of the processes

developed.

5.1 Exploratory Analysis

We carry out a analysis of the data through historical data provided by the retailer (of

the sales history, inventory history, elaborated plans, among others).

Initially, we proceeded with the selection of four product categories with the purpose of

performing a preliminary analysis of these specific data. These categories of products

were chosen because they represent different business components and because they

have several characteristics and behaviours. Behaviours from inventory problems, sea-

sonality problems, products with relevant impacts in relation to meteorology, among

others. The four product categories chosen were Shorts, Polos, Dresses and Pajamas.

When we review the sales history (01/01/2015 - 31/12/2017) of the product categories

described above represented in Figure 5.1 and in Figure 5.2, it is possible to find dif-

ferent characteristics between them.

65

Page 84: Dissertação de Mestrado Mestrado em Estatística - UMinho

66

Figure 5.1: Monthly Sales History (Dresses & Shorts)

Figure 5.2: Monthly Sales History (Polos & Pajamas)

The Pijamas category is considered by the retailer as a strong category, since, on

average, one sells about 4895 units per month in all stores, as can be seen in Table 5.1.

They also consider the categories Dresses and Shorts as medium categories, in that they

sell, on average, about 3380 and 3684 units monthly, respectively. As we have seen in

Page 85: Dissertação de Mestrado Mestrado em Estatística - UMinho

67

Figure 5.1, over the three years of historic, sales in the Shorts category are lower than

in previous categories and in which the retailer regards it as a weak category, of 1766

units monthly. Table 5.1 represents a summary of the main statistics in relation to the

sales history of the four product categories.

Table 5.1: Summary of Monthly Sales History

min 1Q median (η) mean (µ) 3Q max σ

Dresses 1630 2388 2830 3380 4274 6233 1284

Shorts 367 671 1532 3684 7389 11820 3778

Polos 715 1228 1714 1766 2183 4335 754

Pajamas 1941 3325 4299 4895 5148 13300 2547

Still, based on the analysis of Figures 5.1 and 5.2, we conclude that the sales history

represented in time series Dresses and Polos might be considered stationary in the

mean, that is, the expected value is constant, or rather does not depend on time. It

can also be observed that both series do not present great variability on the data. It

should also be noted that the Dresses product category presents seasonality, unlike the

Polos product category. Still, it is notable that the Pajamas category is not a stationary

time series since it presents a tendency (linearly increasing) on time. In the category

of Shorts, there are indications that this series is not stationary (expected value is not

constant). It should also be noted that the categories of pajamas and shorts present

great variability on the data. Finally, we highlight that the pajamas categories present

seasonality, unlike the product category Shorts.

Returning now to a stock-based analysis, through the analysis of Figure 5.3, and based

on inference tests carried out, it is possible to conclude that there are significant differ-

ences between the weekly stock level present in the stores in the Spring-Summer 2017

season and the Spring-Summer 2016 in all product categories analyzed. The exception

is the Shorts product category, for which there is statistical evidence that the average

weekly stock between the two seasons might be assumed as the same. On the contrary,

the three other product categories, there is statistical evidence that the weekly stock

level of the stores in the Spring-Summer 2017 season is higher than the Spring-Summer

2016 season, according to tests carried out.

Page 86: Dissertação de Mestrado Mestrado em Estatística - UMinho

68

Figure 5.3: Weekly Stock History (SS16 (Spring-Summer 2016) - SS17 (Spring-Summer

2017))

In Table 5.2 one can verify the occurrence of stockouts of the four categories of products

in the two seasons.

Table 5.2: Summary Stocks

Sales SS16 Stockout’s SS16 Sales SS17 Stockout’s SS17

Dresses 5694 195 8158 254

Shorts 33750 562 40143 516

Polos 11571 188 15899 247

Pajamas 19961 561 28165 498

Through statistical tests carried out, we can conclude that there was a generalized

increase in the demand for the four categories of products, which consequently re-

sulted in an increase in sales. In terms of stockout’s occurrences, there were significant

differences between the weekly stockout occurrences of the two seasons in the product

categories Shorts and Pajamas. Unlike the product categories Polos and Dresses, where

it was concluded that there were no significant differences between the occurrences of

stockout’s weekly between the two seasons.

In Table 5.3 is represented a summary of four analyzed product categories.

Page 87: Dissertação de Mestrado Mestrado em Estatística - UMinho

69

Table 5.3: Summary about Product Category

Product Category Sales Behaviour Seasonality Observation

Dresses Medium Seasonal Standard behaviour

Shorts Medium No seasonal Seasonal Problem

Pajamas Strong Seasonal Stock problems

Polos Weak No seasonal Relevant Weather Impact

5.2 Algorithm Results

In this section, we aim to study the parameters of one possible configuration of the

developed solution, as well as to carry out evaluations of the optimization of the initial

plan of allocation.

5.2.1 Parametrizations

Since the developed solution allows different strategies depending on the retailers’ ob-

jectives, it is important to study the effects of alternative strategies that may be by the

retailer as function of certain parameters.

Stock coverage is one of the parameters that requires a lot of attention from retailers,

since it ensures that there is no stockout in stores during this coverage period. The

solution developed allows defining different stock coverages for different product cate-

gories according to the commercial objectives of the retailers. In Figure 5.4, we are able

to observe the allocation suggestion given by the algorithm assuming different stock

coverage values.

Page 88: Dissertação de Mestrado Mestrado em Estatística - UMinho

70

Figure 5.4: Coverage of Stock

As we can be see in Figure 5.4, the allocation amounts differ significantly with changes

in the coverage parameter. As expected, with shorter coverage days, the amount to

be allocated will be lower compared to higher coverage days, since the amount to be

allocated will only need to cover a shorter range of days. It can be concluded that the

closer to the beginning of the season, the more differences are felt with the changes of

this parameter because the preference of consumers is not precise and the availability

of products to be allocated is greater. Still, we can see in Figure 5.4 that in release 1

the allocation differences between the stock are thousands, whereas for release 3 there

is a convergence of the allocation quantity for the different coverage values of stock.

Turning now to the study of the Gap-End parameter (see Section 4.2.6 for more detail),

we have distinct scenarios. Since our proposed solution allows to define different values

of gap-end of different categories, scenarios were performed with gap-end off, gap-end

equal to zero weeks and gap-end equal to four weeks. The results are summarized in

Figure 5.5. It should be noted that the retailer’s choice of the best configuration for

certain product categories is a complex problem.

Page 89: Dissertação de Mestrado Mestrado em Estatística - UMinho

71

Figure 5.5: Gap - End

As we can see in Figure 5.5, the suggestion given by the algorithm is further reduced

when the gap-end parameter is turned off. Moreover, there is a significant increase in

the allocation suggestion when the gap-end is connected and this value is greater than

zero. This increase is due to the fact that this parameter forces to send of products

when they are close to the end of their life cycle, in other words, this parameter turns

these products into mandatory products. Thus, the amount of allocation stock given

by the algorithm when the Gap End parameter is off only allows to cover the needs in

stores. Whereas, the products that were added after changing the parameter will not

be needed in store, thus leading to an overstock of the stores.

5.2.2 Algorithm Performance

After the development of the algorithm and its tests in the previous season (SS 17),

weekly follow-up executions of the season SS 18 were carried out. This monitoring

began in the week 19 of the year 2018. Thus, after running the algorithm, the results

were sent to the retailer. In this way, the retailer could gain confidence in the solution

developed by analyzing the potential of the presented solution and performing internal

tests of the robustness and performance of the solution.

But prior to the start of the season follow-up, an analysis was made of the state of

the seasonal SS 18 allocations. With this analysis, it was possible to verify that only

6 % of product and store category combinations have optimal stock while 23% of the

combinations of product categories and store has excess stock. This analysis made it

possible to verify that 12% of combinations of product categories and stores have a

Page 90: Dissertação de Mestrado Mestrado em Estatística - UMinho

72

stock less than 45 days of coverage, while 6 % have a stock greater than 60 days of

coverage.

The first execution of the season provided to the retailer was dated the week 19 of the

year 2018. At the date of writing this thesis, six executions were carried out and sent to

the retailer. In this section, we will calculate the gains/losses with the application of the

suggestions given by the algorithm rather than the execution made by the retailer.Table

5.4 represents the dates of the six executions.

Table 5.4: Execution realized

Executions Date Week

Release 1 (R1) 07/05/2018 19

Release 2 (R2) 14/05/2018 20

Release 3 (R3) 21/05/2018 21

Release 4 (R4) 28/05/2018 22

Release 5 (R5) 04/06/2018 23

Release 6 (R6) 11/06/2018 24

Through the six executions, it was possible to calculate considerable gains with the

application of the solution. If we focus on the Shorts product category, in Figure 5.6,

we depict the execution of the retailer in comparison to the suggestion of the algorithm

together with the needs of all stores.

Figure 5.6: Allocation in SS 18 - Products Category Shorts

Page 91: Dissertação de Mestrado Mestrado em Estatística - UMinho

73

Through the graphical visualization, it is possible to verify that the suggestion given

by the algorithm together with the actual execution of the retailer, satisfies the needs

of all the stores in the six executions. However, it is also verified that the retailer’s

delivery is well above the needs of the stores and thus, allocating too much product.

Overloading product in stores can lead to two major problems, the inability of the

store to receive other products due to overstock sense in the store’s warehouse and the

inability of other stores to receive product and consequently lost sales occurrence.

Also in this figure, it is possible to verify that the algorithm suggestion satisfies the

needs of the stores in the six executions and sends a smaller quantity significant in

comparison with the executions of the retailer. Table 5.5 represents the allocation

quantities given by the algorithm and the execution of the retailer together with the

needs of the stores for the product category Shorts.

Table 5.5: Comparison of allocation in SS 18 - Products Category Shorts

Retailer Algorithm Needs 4 (Retailer - Algorithm)

Release 1 (R1) 394 301 239 93

Release 2 (R2) 168 72 55 96

Release 3 (R3) 42 48 15 -6

Release 4 (R4) 74 52 49 22

Release 5 (R5) 54 52 51 2

Release 6 (R6) 0 0 0 0

Total 732 525 409 207

Instead of the retailer’s allocation, the algorithm’s allocation suggestion allowed a ware-

house stock retention gain of about € 1000 in week 19 only for the Shorts product

category (calculation given by the average RRP of the product category and by the

difference between the execution of the retailer and the suggestion of the algorithm in

Table 5.5). In the 6 executions, the suggestion given by the algorithm allowed a gain

of € 2500 for the product category Shorts. If we calculate the gains for all categories

of products, we conclude that our proposal allowed a potential for the first execution

(R1) of about 25k€. Now, by setting a store and comparing the actual allocation and

the allocation suggested by the algorithm, represented in Figure 5.7, it can be seen that

Page 92: Dissertação de Mestrado Mestrado em Estatística - UMinho

74

the retailer allocates a large number of products compared to the suggestion given by

algorithm. It should be noted that the algorithm allocates quantities lower than the

actual allocation while safeguarding the coverage of the needs.

Figure 5.7: Allocation in SS 18 - Store

Table 5.6 shows the allocated executed by the actual allocation and the quantities

suggested by the algorithm together with the requirements for each release.

Table 5.6: Comparison of allocation in SS 18 - Store

Retailer Algorithm Needs 4 (Retailer - Algorithm)

Release 1 (R1) 104 44 41 60

Release 2 (R2) 108 48 40 60

Release 3 (R3) 36 22 9 14

Release 4 (R4) 76 36 34 40

Release 5 (R5) 0 0 0 0

Release 6 (R6) 0 0 0 0

Total 324 150 124 174

Through the computation previously explained, with the application of the algorithm

suggestion there would be a retention of stock of 174 units in the warehouse, this

retention would allow the retailer to save around 1500 € (average RRP of all product

categories) in the six executions.

Page 93: Dissertação de Mestrado Mestrado em Estatística - UMinho

75

5.3 Business Perfomance Metrics

In this section we will present some business metrics to follow the season’s allocation

process, from the implementation of the allocation throughout the season, the

differences between the initial plan and the allocation execution, the number of new

products that the stores receive throughout the season, among others.

The graphics presented in this section only illustrate the season SS 18 from the

beginning of January 2018, ie starting from week 1 of the year 2018.

Assortment Refreshness

Interesting metrics for the business is the new product refresh in store (assortment

refreshness is a metric inserted in the metrics stocks, see the section 4.3.1 for more

details). Thus, the retailer allows you to track the amount of new product in the last

week in each store. Fixing the same stores of the Figure 5.9, we have represented in

Figure 5.8 the quantity of new product that each store received in the previous week

along the season. This shows that the plan is not always followed because it was done

9 months earlier.

Figure 5.8: Assortment Refreshness - Season SS 18

As we can see graphically, there is a great push from the beginning of the season to

send the new product to stores, where naturally this shipment is reduced throughout

the season. Still, in the 5.8 chart, store 4 has a high amount of product allocation at

week 17, since it would be close to its opening.

Page 94: Dissertação de Mestrado Mestrado em Estatística - UMinho

76

Percentage of Execution

We should remember that the execution percentage is a business performance metric

(see the section 4.3.3 for more detail). Figure 5.9 represents the execution of the

spring-summer 2018 season allocation plan for four stores. In order to understand

different behaviors of stores, we proceeded to choose different types of stores. Store 1

is a store with very strong profiles because it has a large sales volume and it is a store

with large areas. While, store 2 is considered an average store, with an average volume

of sales. On the other hand, store 3 is a weak store that has a lower sales volume than

the previous two stores. And lastly, store 4 is a store that opened for the first time

during the SS 18 season.

Figure 5.9: Allocation Execution By Store - Season SS 18

As we can see in Figure 5.9, the current stock allocation is differentiated depending

on the stores, that is, stronger stores (store 1 and 2) present greater needs than the

previous ones due to the greater demand in them and, in this sense, the percentages of

allocation in these stores are higher throughout the whole season.

We focus now on the implementation of the allocation plan of the season SS 18 by

category of products as it is represented in Figure 5.10. In this figure, it is possible to

observe that throughout the season certain categories of products (mainly Polos and

Shorts) remain constant during some weeks and suddenly there is a high amount of

stock allocation to the stores. This shows that these categories strongly change over

time, as they are influenced by the weather.

Page 95: Dissertação de Mestrado Mestrado em Estatística - UMinho

77

Figure 5.10: Allocation Execution By Product Category - Season SS 18

Difference between Initial Plan and Allocation Execution

The differences between actual allocation and the initial plans is important to evaluate

both to fully understand the differences he made before the initial plan being imple-

mented, metric inserted in the performance metrics of the business (see the section

4.3.3 for more detail). By way of example, if we fix the same categories of products,

it can be seen in Table 5.7 that the products of the Dresses category are shipped to

stores on average 2 days in advance. On the contrary, the product category Pajamas

are shipped to stores on average 0.10 days later compared to the initial plan.

Table 5.7: Difference between initial plan and allocation executed

Product Category Difference

Dresses - 2.00 days

Shorts 0.00 days

Pajamas 0.10 days

Polos - 0.65 days

Range Max

Other metric of great interest to the retailer is the one that allows to measure the time

interval between the first and last store that received a certain product (see the section

4.3.3 for more details). This metric makes it possible to detect that a certain product is

left in the warehouse or that demographically close stores may have different product

ranges in store, which is to be avoided. For this, at week 23 of the year 2018, the

maximum range of the four product categories is presented in Table 5.8.

Page 96: Dissertação de Mestrado Mestrado em Estatística - UMinho

78

Table 5.8: Range Max - Season SS 18

Product Category Range Max

Dresses 82.35 days

Shorts 73.43 days

Pajamas 83.79 days

Polos 115.25 days

As seen in the previous table, the product category with the lowest range max is the

Shorts category, meaning the average difference between the first store and the last

store for all products in this category is 73.43 days.

Page 97: Dissertação de Mestrado Mestrado em Estatística - UMinho

Chapter 6

Conclusion and Future Work

There has never been so much talk on the future as now in the retail industry. The

introduction of concepts in our day-to-day life as artificial intelligence or crypto-coins

allow us to conclude that technology will dictate the deepest changes in how people

relate to the world and how they acquire their assets. Accompanying this forecast with

deep attention, in particular in the retail sector that is now experiencing a paradigm

shift. The CRBE research team focused on "O Futuro do Retalho | 2030" with the

aim of encouraging reflections on the different scenarios that retailers will face and

understand how they can prepare for the future (Carlos Récio, 2017).

This team anticipates that automation, customization, technological innovation and big

data will be the key to driving growth in the retail industry. Fact is that on this two-way

road, from retailers and consumers, both benefit: the advanced integration of trillions

of connected sensors will provide powerful insight into people’s consumption patterns

and habits enabling retailers to anticipate their customers’ requirements - before they

realize it themselves - making the distribution chain more accurate and efficient.

In terms of this work, and from a global point of view, this process demonstrated a

strong performance in terms of efficiency, being expected to achieve very substantial

productivity gains in a business process that is often wasteful at retailers. Initial tests

point to reductions of up to 10% of in-store stockouts associated with reductions of

up to 2 FTE’s in the planning and stock control teams at a retailer with the annual

turnover of € 20-50m.

However, along the elaboration of this work, there was a lack of volume and quality of

79

Page 98: Dissertação de Mestrado Mestrado em Estatística - UMinho

80

input data that resulted in a limitation of the solution developed. In these situations, it

may be important to evaluate an alternative to use specific modelling techniques that

present a higher performance with a lower dataset requirement. It is also important to

highlight the inability to achieve a solution with daily processing, which may make it

impossible to apply it in real operational contexts of the retail business, in particular,

this might be a problem in the beginning and end periods of the collection (balances),

during which the dynamism is very high and business decisions entail a short period

of application. Lastly, it was also difficult to achieve adequate levels of accuracy in

the retail environment that allow users to trust the use of Automatica Allocations

algorithms. Typically, this represents a very high effort on the configuration process

that goes beyond the effort of manually finding a marginally less optimized solution.

Page 99: Dissertação de Mestrado Mestrado em Estatística - UMinho

Bibliography

ARTSOFT. Retalho | Software de Gestão, n.d. URL https://www.artsoft.pt/

software-gestao/solucoes-verticais/retalho.

Sven Axsäter, Johan Marklund, and Edward A Silver. Heuristic methods for centralized

control of one-warehouse, n-retailer inventory systems. Manufacturing & Service

Operations Management, 4(1):75–97, 2002.

Ronald H Ballou. Business Logistics Management: Planning, Organizing and Control-

ling the Supply Chain. Instructor’s Manual. Prentice Hall, 1999.

Banco de Portugal. Boletim Estatístico | Statistical Bulletin 4. page 370, 2018.

Liz Barnes and Gaynor Lea-Greenwood. Fast fashioning the supply chain: shaping the

research agenda. Journal of Fashion Marketing and Management: An International

Journal, 10(3):259–271, 2006.

Grete Birtwistle, Noreen Siddiqui, and Susan S Fiorito. Quick response: perceptions

of uk fashion retailers. International Journal of Retail & Distribution Management,

31(2):118–128, 2003.

Margaret Bruce, Lucy Daly, and Neil Towers. Lean or agile: a solution for supply

chain management in the textiles and clothing industry? International journal of

operations & production management, 24(2):151–170, 2004.

Carlos Récio. O futuro do retalho em 2030, 2017. URL

https://www.jornaldenegocios.pt/opiniao/colunistas/detalhe/

o-futuro-do-retalho-em-2030.

81

Page 100: Dissertação de Mestrado Mestrado em Estatística - UMinho

82

Felipe Caro and Jérémie Gallien. Inventory management of a fast-fashion retail network.

Operations Research, 58(2):257–273, 2010.

Fangruo Chen. Optimal policies for multi-echelon inventory problems with batch or-

dering. Operations research, 48(3):376–389, 2000.

Martin Christopher, Robert Lowson, and Helen Peck. Creating agile supply chains in

the fashion industry. International Journal of Retail & Distribution Management,

32(8):367–376, 2004a. ISSN 0959-0552. doi: 10.1108/09590550410546188. URL

http://www.emeraldinsight.com/doi/10.1108/09590550410546188.

Martin Christopher, Robert Lowson, and Helen Peck. Creating agile supply chains in

the fashion industry. International Journal of Retail & Distribution Management, 32

(8):367–376, 2004b.

Andrew J Clark and Herbert Scarf. Optimal policies for a multi-echelon inventory

problem. Management science, 6(4):475–490, 1960.

Informa DB. Barómetro | Abril 2018. Technical report, 2018.

Peter Doeringer and Sarah Crean. Can fast fashion save the us apparel industry?

Socio-Economic Review, 4(3):353–377, 2006.

Gary Eppen. Centralized ordering policies in a multi-warehouse system with lead times

and random demand. Multi-Level Production/Inventory Control Systems, pages 51–

67, 1981.

N Erkip. A restricted class of allocation policies in a two-echelon inventory system. In

Technical Report No. 628. School of Operations Research and Industrial Engineering,

Cornell University, 1984.

Awi Federgruen and Paul Zipkin. Approximations of dynamic, multilocation production

and inventory problems. Management Science, 30(1):69–84, 1984.

John Fernie and Nobukazu Azuma. The changing nature of japanese fashion: Can

quick response improve supply chain efficiency? European Journal of Marketing, 38

(7):790–808, 2004.

Page 101: Dissertação de Mestrado Mestrado em Estatística - UMinho

83

Charles H Fine and Stephen C Graves. A Typical Planning Model for Manufacturing

Subcomponents of Mainframe Computers. Alfred P. Sloan School of Management,

Massachusetts Institute of Technology, 1988.

Rafaela Frankenthal. Como a tecnologia está causando mudanças no com-

portamento do consumidor, 2017. URL https://mindminers.com/consumo/

tecnologia-e-comportamento.

Gary Gereffi and Olga Memedovic. The Global Apparel Value Chain : What Prospects

for Upgrading by Developing Countries ? page 36, 2003.

Larry C Giunipero, Susan S Fiorito, Dawn H Pearcy, and Lisa Dandeo. The impact of

vendor incentives on quick response. The International Review of Retail, Distribution

and Consumer Research, 11(4):359–376, 2001.

José Fernandes Gonçalves. Gestão de aprovisionamentos. 2000.

Stephen C Graves. A multiechelon inventory model with fixed replenishment intervals.

Management Science, 42(1):1–18, 1996.

Tony Hines. The emergence of supply chain management as a critical success factor for

retail organisations. International Retail Marketing, page 108, 2004.

Alexander Hübner. Retail category management: decision support systems for assort-

ment, shelf space, inventory and price planning, volume 656. Springer Science &

Business Media, 2011.

INESC-TEC. Retalho e Logistica: Por detrás das prateleiras, 2016. URL https:

//www.youtube.com/watch?v=iw1V6FycDBA&t=53s.

Peter L Jackson. Stock allocation in a two-echelon distribution system or what to do

until your ship comes in. Management Science, 34(7):880–895, 1988.

Peter L Jackson and John A Muckstadt. Risk pooling in a two-period, two-echelon

inventory stocking and allocation problem. Naval Research Logistics (NRL), 36(1):

1–26, 1989.

Page 102: Dissertação de Mestrado Mestrado em Estatística - UMinho

84

T. Jackson and D. Shaw. Mastering Fashion Buying and Merchandising Manage-

ment. Palgrave Master Series. Palgrave Macmillan, 2000. ISBN 9780230365155.

URL https://books.google.pt/books?id=SZBoxlTKt18C.

Henrik Jönsson and Edward A Silver. Stock allocation among a central warehouse

and identical regional warehouses in a particular push inventory control system.

International Journal of Production Research, 25(2):191–205, 1987.

R. Kaplinsky. Globalization, Poverty and Inequality: Between a Rock and a Hard

Place. Wiley, 2005. ISBN 9780745635538. URL https://books.google.pt/books?

id=EwanZofrY1YC.

Edward J McGavin, Leroy B Schwarz, and James E Ward. Two-interval inventory-

allocation policies in a one-warehouse n-identical-retailer distribution system. Man-

agement Science, 39(9):1092–1107, 1993.

Tata Mcgraw. The art of retailing. Hill Education, 2002.

Amorim Pedro. Transport, Logistics Classes. PhD thesis, FEUP, 2015.

Rodrigue Jean-Paul. The Geography of Transport Systems. PhD thesis, Hofstra Uni-

versity, 2017. URL https://transportgeography.org/.

Kaj Rosling. Optimal inventory policies for assembly systems under random demands.

Operations Research, 37(4):565–579, 1989.

Leroy B Schwarz. A model for assessing the value of warehouse risk-pooling: risk-

pooling over outside-supplier leadtimes. Management Science, 35(7):828–842, 1989.

Keah Choon Tan. A framework of supply chain management literature. European

Journal of Purchasing & Supply Management, 7(1):39–48, 2001.

Cushman & Wakefield. Portugal Retalho | Março 2018. 2018.

C Donald J Waters. Global logistics and distribution planning: strategies for manage-

ment. Kogan Page Publishers, 2003.

Joachim Zentes, Dirk Morschett, and Hanna Schramm-Klein. Strategic retail manage-

ment. Springer, 2007.