INCREASING THE COMPETITIVENESS OF MEXX B.V. AT SIZE ALLOCATION Size level availability - the Achilles heel of apparel retailing

1

INCREASING THE COMPETITIVENESS OF MEXX B.V. AT SIZE ALLOCATION Size level availability - the Achilles heel of apparel retailing

Master thesis, MscBA, specialization Operations and Supply Chains University of Groningen, Faculty of Operations

Submission date:

27-08-2012

Judit Osika

Student number: s2021668 Aquamarijnstraat 221.

9743 PE Groningen tel.: +31 681459822 e-mail: j.osika@student.rug.nl

Supervisors/ university Dr. T. Bodea

Dr. X. Zhu Supervisor/ field of study

L. Kay

Mexx B.V., Amsterdam

2 Abstract

Matching supply and demand has never been an easy task, especially not in the retail industry. The life cycles of fashion products is shortening while the variety of products offered are increasing and customers are becoming more and more demanding. Before products can be found at any store, an extensive planning process, the merchandise planning takes place. It has to be decided when, where and how much of the products to supply so that the right products are available at the right sizes, colors etc.

When products are not available i.e. they are out of stock (OOS) when demand occurs for them, retailers do not only lose sales and margins but they may also face dissatisfied customers and lose market share in the long term.

In general, forecasting demand is based on extrapolating historic sales data. However, this may result in biased forecasts: sales data only represent partial demand and not the demand which was not satisfied due to the unavailability or stockouts. Therefore sales data is considered to be constrained when the sales are limited by an OOS event.

A sample of 12 styles of merchandise was taken to observe the OOS phenomenon and its associated costs at Mexx B.V. It was found that OOS ratio on average is already at the 4th week after a product was introduced is 27% and increases up to 64% by the 16th week. This could imply that unconstraining methods could foster a better matching of demand on the size level and enable the retailer to have more precise information about demand surrounding a store.

Keywords: merchandise (demand) planning, out-of-stock rates, unconstraining sales data

3 Table of contents

1. Introduction ...4

1.1. Topic area ...4

1.2. Problem statement and research question ...6

2. Literature review ...7

2.1. Revenue management ...7

2.2. Unconstraining methods ...7

2.3. Stock-out in the apparel retail industry ...8

3. Methodology...9

4. MEXX B.V. ...11

4.1. The merchandise hierarchy ...11

4.2. The merchandize planning process ...12

4.2.1. The corporate strategy and the financial planning ...12

4.2.2. Product and Store Planning ...13

4.2.3. Range planning ...14

4.2.4. Assortment planning ...14

4.2.5.Allocation & Replenishment ...14

4.2.7. Size profiles ...15

5. RESULTS ...16

5.2. Out-of- stock rates ...16

5.2 Availability of all sizes on the store levels ...18

5.3. OOS on the store- style- size level ...19

5.4. Quantifying the size level OOS costs ...21

6. DISCUSSION AND CONCLUSION ...24

7. REFERENCES ...26

8. APPENDICES ... Error! Bookmark not defined.

4 1. Introduction

1.1. Topic area

Matching supply and demand has never been an easy task, especially not in the retail industry. Making the right products and services available at the right time, at the right place and in the right quantity is a common manifesto for the industry (Kumar and Banga, 2007). To fulfill this goal, retailers have to manage a series of challenging tasks such as predicting the desires of customers, buying and allocating the right set of merchandise to their stores, setting prices and effective promotions (Kumar and Banga, 2007). These tasks are even more challenging in these present days as product life cycles tend to be shortening while the variety of products offered are increasing, and customers are becoming more and more demanding (Fernie and Corcoran, 2011). As a result the retail supply chain is characterized by inefficiencies such as high inventories, stockouts, steep markdowns, poor assortment, inconsistent pricing and ineffective promotions which result in a negative effect on retailer’s profits (Vinod, 2005).

Although improved distribution systems and technologies are of great help in this challenging business environment, retailers still have a hard time to match supply with demand (Friend and Walker, 2001). Sales are often limited by the supply of goods and services. It has been estimated that 8% of items that customers intend to buy are out of stock, and that one-third of goods are sold at markdowned prices (Friend and Walker, 2001). Both occasions harm the revenues of retailers. If retailers are not able to make the products available when demand occurs for them, they do not only lose sales and margins but they may also face dissatisfied customers and lose market share in the long term (Zinn and Liu, 2001). In the grocery industry it has been estimated that when customers are faced with a stockout for the first time, 31% of the customers leave the store without purchasing.

When it happens for the third time, the percentage of leaving customers without buying anything is as high as 69% (ECR Europe 2003). Retailers are able to satisfy customers when supply matches or exceeds demand. With excess supply however, customer retention is high but retailers are confronted with the costs of holding inventory. They need to use markdown sales to clear their excess inventory which means that they also lose profits as they have to sell their products at a lower price than initially intended. It has been shown that department store markdowns increased from 8% to 37% between 1971 and 1997, and that this trend continues to increase. Fisher et al. (1994) indicated that in highly volatile markets such as the fashion retail industry, the costs of markdowns and stockouts could actually exceed the total cost of manufacturing. Therefore managing the trade-off between the costs of excess inventory and the costs of insufficient inventory has a very important effect on the long-term profitability of retailers.

Developing and implementing a merchandising plan is crucial to overcome the above-mentioned difficulties and to achieve improved margins, inventory turns, comparable store sales and earnings (Friend and Walker, 2001).

Due to outsourced production, lead times in the fashion retail industry can take up to six months. Consequently, there is only a limited possibility for reordering and as fashion items become quickly obsolete, the merchandise have to be sold within couple of weeks or months ((Levy and Weitz, 2008). This means that retailers have maximum 2 or 3 months to generate profit on a product as later it will be not fashionable anymore and it will not sell and generate profit. The merchandise planning process aims to define when, where and how much of the merchandise to sell so that revenue targets are supported. Decisions about merchandising planning, buying, allocation, replenishment, pricing and promotion are all primarily based on historic sales data (Friend and Walker, 2001).

As the product variety offered by the retailers increases year by year, the stock keeping units (SKUs) increase as

well. A stock keeping unit is the lowest level of the merchandize hierarchy and it represents a certain style (i.e. a

variant), color and size variation. For example if two styles of blouses are available (one style is with long

5 sleeves and the other with short sleeve) in two colors (red and white) and five sizes (XS,S,M,L,XL), they represent altogether 20 SKUs. To facilitate the overview of the styles (and corresponding SKUs), retailers use the tool of merchandise hierarchy. The merchandise hierarchy is a kind of taxonomy of the products offered and it is specific to every company. At Mexx B.V. the highest level of the hierarchy is the product group. A product group depends on their target market of the retailers. For example at Mexx B.V. there are the product groups of women, men, youth, accessories and footwear. A product group can be further broken down into classes. A class is a unit in the merchandise hierarchy in which merchandise is carried that a customer may view complementary (Hart and Rafiq, 2006). Mexx B.V. offers generally 10 classes such as blouses, blazers, t-shirts, skirt etc. Within every class there are the styles. The styles are different from each other in material, tailoring and patterns.

Depending on how many colors they are available they are further broken down into a style/color level which is again broken down into a style/color/size level, the SKUs. The merchandise hierarchy does not only to give an overview of the products offered but it is also reflected in the organizational structure. Another complex decision is where, at which locations to offer the products. To continue with the previous example that a retailer wants to allocate 20 SKUs (2styles*2colors*5 sizes) to its 100 stores, it would mean that 2000 decisions have to be made about how much and where to allocate. Therefore to facilitate the planning process, stores of similar sizes are classified together into so-called clusters. With the help of the clustering it can be decided easier which cluster receives which styles. Usually the higher clusters (the bigger stores) receive more styles of class than smaller stores.

In general, forecasting demand is based on extrapolating historic sales data. However, this may result in biased forecasts: sales data only represent partial demand and not the demand which was not satisfied due to the unavailability or stockouts. After products are sold out, demand cannot be observed as “attempted sales” is difficult to record (Talluri and van Ryzin, 2004). Historic sales data embodies only observed demand hence often termed as constrained or truncated demand. Talluri and van Ryzin (2004) illustrate the effect of forecasting based on historic sales data with the example of a product which has been consistently closed (i.e. not open to sale) in the past. Its observed demand would be zero, and a forecast based on this data would forecast demand as zero. Leaving sales data constrained or truncated can result in a negative spiral-down effect on revenue management (Cooper, Homem-de-Mello and Kleywegt, 2006).

Unconstraining stands for the application of statistical methods in order to correct for the incomplete sales data and estimate how much could have been sold if the right quantity of products had been available (Friend and Walker (2001). Thus, observed demand has to be combined with the estimated latent demand (Zhu, 2005).

Although there is no universal solution to unconstrain sales data, Weatherford and Pölt (2002) argue that switching to a better method can increase revenues with 0.5-1 % in the airline industry.

Besides using historic sales data, another weak point of apparel retailers may be that assortment planning relies on aggregate measures such as the clustering of stores mentioned before (Grewal et al, 1999). Mantrala et al.

(2009) suggests that retailers should plan their assortments at individual store levels rather than at a national level. The reason for this is to better match the local taste of customers (Mantrala et al., 2009). Friend and Walker (2001) also express that the most challenging task for retailers is allocating the right mix of merchandise to the stores. They focus on the size level availability and they point out that every store faces different demand with respect to sizes, therefore setting a single size profile for the entre supply chain is ineffective. Moreover if size level out-of-stocks (OOSs) remain unconstrained on the store level, the true size profiles are not revealed and a biased size profile is used. They further propose that each store should have their specific size profiles;

however retailers usually do not consider the allocation at this level because of the associated costs.

The proposition of Friend and Walker (2001) is also underpinned by the emergence of the increasing number of

merchandise optimization software packages such as those provided by JDA, Retek and SAP that model

6 demand at the level of individual stores and individual stock-keeping units (SKUs). With the use of these software packages some retailers reached a 20% gain in productivity (Friend and Walker, 2001). However research on unconstraining methods in the retail industry has not received the same attention as in the airline or hotel industry. This thesis is organized as follows. In the next section the problem and the research question are stated along with the relevance of the study. In chapter 2, the relevant literature is discussed. Chapter 3 introduces the methodology and chapter 4 includes the description of the case study and the analysis. In chapter 5 the results and findings are discussed. Chapter 6 discusses the conclusion of the thesis and chapter 7 discusses further implications and further research areas.

1.2. Problem statement and research question

The aim of this research is to investigate the size level out-of-stock phenomenon in the apparel retail industry and to investigate the feasibility of unconstraining size level sales data. As discussed in the introduction using constrained sales data for planning purposes can harm both the current and future revenues of the retailer.

Therefore quantifying the lost sales does not only mean that the current but also the future revenues of retailers are constrained. Revealing size level OOS rates for individual stores could help in creating size profiles for stores which better match customer demand. Better matching customer demand in turn increases revenues and customer satisfactions levels.

The research questions are: what are the size level OOS rates within the locations and what costs they are associated with?

Before answering the research question the current practice of Mexx B.V. is discussed with regards to the

merchandise planning decisions.

7 2. Literature review

The aim of the literature review to introduce the “science” of revenue management. Besides that another goal is to show where the method of uncontraining is currently staying. It is also the intention of the literature review to demonstrate what revenue management practices have been used so far in the apparel retailing. And finally to show what studies have been made in the retail industry on lost sales ideas.

2.1. Revenue management

Talluri and van Ryzing (2004) and Phillips (2005) provide a thorough introduction to the field of revenue management. Talluri and van Ryzin (2004) refer to revenue management as the complement of supply chain management. Supply chain management addresses the supply side decisions along with the processes of the firm with the objective of lowering costs while revenue management deals with the firm’s connection to the market with the goal of increasing revenues. Thus revenue management concerns decisions about demand management and it is an interdisciplinary science, rooted in operations research methods and management science practices.

In their book they identify three areas of revenue management: structural decisions, price decisions and quantity decisions. The relevance and the timescale of these three areas are dependent on the business environment the firm operates in. Structural decisions concern strategic decisions about the selling of products. Pricing and quantity decisions are about setting prices or deploying capacities in advance. Depending on the choice, the ability to adjust the price or the quantities on the tactical level is limited.

For example retailers usually commit to quantities, in the form of initial stocking decisions. Therefore they have more flexibility to adjust prices over time. However if they decide to keep a certain amount of stock at the warehouse and make a replenishment decision rather than committing all their stock in advance is a mixture of price and quantity based revenue management. The decision is dependent on the ability of the firm to react to the market changes by altering prices or quantities. In apparel retailing generally companies commit to order quantitates well in advance of the selling season, in some cases even to certain stocking levels in each store.

Often it is very costly or impossible to reorder stock or reallocate inventory from one store to another. At the same time, it is easier for most retailers to change prices as this may only require changing signage and making data entries into a point of sale system. For online retailers, changing signage of prices is almost costless therefore they benefit from price-flexibility.

Furthermore, they discuss the common elements of price- and quantity- based revenue management, such as forecasting. For retailers forecasting usually concentrates on forecasting demand as a function of price and promotion. Forecasting often faces the problem that data are partially observable or missing. Incomplete data occurs when sales and no-sales are not directly observable. This makes it difficult to obtain information on customer purchase behavior. As mentioned before revenue management is rooted in the airline reservation context, hence most of the unconstraining methods are applied in that industry.

2.2. Unconstraining methods

Before any successful revenue management optimization could take place, accurate demand information is necessary to be available (Lan et al., 2008).

Unconstraining sales data to obtain more precise demand information can have significant effects on revenue management, however the research on this method is scarce compared to setting and adjusting booking limits (Queenan et al., 2007). Unconstraining sales data is relevant to capacity or resource constrained companies.

There are two ways to approach data unconstraining: direct observation of unmet demand or the use of statistical

methods. Observing unsatisfied demand means recording latent demand which occurs when the resource or the

capacity constraint is reached. Direct observation of latent demand in some industries is simply not possible

because of the used distribution channels (Queenan et al., 2007). In other industries such as the hotel industry,

observation of unmet demand is recorded in the form of request turndowns. However not all the information on

8 turndowns represent latent demand as requests can be also rejected because of the price considerations of customers. Obtaining quality demand data from turndowns is becoming even more difficult with the appearance of booking sites which do not record turndowns. Therefore there is an increasing need for robust statistical methods to unconstrain data (Queenan et al., 2007). Unconstraining methods pursue a common methodology:

demand over time is classified as being constrained or unconstrained.

The most widely used unconstraining method is the expectation-maximization (EM) method was introduced in 1977 by the study of Dempster, Laird and Rubin. Although studies on unconstraining airline demand data began in the mid-1990s (such as Skwarek, 1996 and Shaleh, 1997) under the aegis of AGIFORMS, the application of the EM method was first considered in the airline context by Pölt (2000) and Weatherford (2000). In their subsequent study, Weatherford and Pölt (2002) compare six unconstraining methods to see which one of them is the most robust in the contex of airline reservation systems. They test three naïve averaging methods, the booking profile method, the projection detruncation method and the EM method. Using simulation, they created scenarios where the fractions of total constrained observations vary between 0 and 98 %. A 0 % censoring means that data represent the true demand as the booking limit is not reached; whereas a 98% truncation means that 98% of the data represent demand constrained by the booking limit. The authors find that the first three methods provide better results when there are lower levels of censoring; but as censoring levels increase, methods based on projection de-truncation and expectation maximization are superior. They suggest that using expectation maximization algorithm, 2-12 per cent of improvement in revenue management can be achieved. Although the expectation maximization method is cited as the most accurate method (Ratliff et al., 2007), it also has its limitation as being time and computation-intensive.

Talluri and van Ryzin (2004) describe five methods to unconstrain data: the EM method, Gibbs sampling, Kaplan-Mier Product Limit Estimator, plotting procedures and the projection detruncation method. Queenan et al. (2007) propose the use of double exponential smoothing (DES) to forecast the constrained values of a data set. They compared this method to one averaging method, the EM, the PD method and the so called life table (LT) method which was used previously in medical and reliability engineering experiments. Ratliff et al. (2008) provide an extensive overview of single class, multi class and multi-flight methods in the airline context. The most recently published study about demand unconstraining is the article by Vulcano, van Ryzin and Ratliff (2012). They propose a method for estimating substitute and lost demand when only sales and product availability is observable.

2.3. Stock-out in the apparel retail industry

Although out-of-stock events have received attention in the retail environment, the majority of the studies focus solely on grocery retailers (Grant and Fernie, 2008). The first study which directly aimed at quantifying the cost of OOS events in the fashion industry was done at the House of Fraser by Carey and Staniforth (2007). Their exit survey has shown that 36% of customers who visited the House of Fraser store and planned to purchase did not, mainly because of the non-availability of size and color (Carey and Staniforth, 2007). They estimated that if only half of those customers have bought what they planned to buy, sales could have increased by £63 million.

Fisher and Rajaram (2000) discuss the retail testing method to improve forecasts of fashion merchandise. Retail testing means that new products are introduced at a sample of stores prior to the selling season and from their observed sales, demand for the entire chain is forecasted. They also investigate the clustering of the stores to identify the similar common factors that explain similar sales patterns and they find that store size and location have little impact on sales patterns.

Most of the research in the apparel industry focuses on identifying the causes of OOSs (such as the study Fernie

and Corcoran, 2011) or focuses on understanding the behavior of customers responses to OOSs (such as the

study of Carey and Staniforth, 2007). Unconstraining sales data in the apparel industry received much less

attention from researchers in the past years. It is difficult to tell why it is the norm in the airline and hotel

9 industry and not in the apparel industry. A possible explanation could be that to unconstrain sales data the lowest levels of aggregation has to be considered while retailers tend to use aggregate levels of sales data. With thousands of SKUs and hundreds of locations, the reason why retailers stayed away from unconstraining methods so far could be the associated costs and time (Friend and Walker, 2001).

3. Methodology

This research is conducted within a case study at Mexx B.V. Amsterdam. Interviews are conducted with the manager and the members of the merchandise planning department to get a clear overview of the process flow of planning. The overall retail strategy, merchandise planning, buying, allocation and pricing are described with respect to sizing. Next to the general description of these processes, feedback loops between the processes are also described (if they exist). The most crucial point of the descriptive part is to point out if size level sales data is unconstrained and whether or not it is fed back into the planning processes.

In the second part, the thesis aims to provide actual proof that unconstraining methods are necessary to use. For this, size level sales and inventory data will be analyzed across several stores. The time period chosen for the analysis is the season of fall/ winter of 2011. This season covers the months from June until November.

The study is targeted at the retail stores as the author’s internship takes place at the department where the planning and allocation decisions are handled for them. These retail stores sell a variety of clothing within the product groups of women, men, youth, accessories and footwear. Depending on which product groups the stores offer, retail stores can be classified into family stores (where all the five product groups are available), lifestyle stores (where woman and men product groups are available) and women stores.

Within the retail stores, the family stores are predominant (32 out of 75). The other two store formats, lifestyle and woman stores are not taken into account in the analysis. The overview of the family stores can be found in the appendix. As mentioned in the introduction, to facilitate decision making, based on their size stores are grouped into clusters. The lower the number a cluster is ranked, the bigger the stores it comprises. For example cluster 1 contains the biggest stores while cluster 7 contains the smallest store.

By the end of the season, out of the 32 stores, 4 were closed down. From the 28 stores, 6 were located in the Netherlands. Because of the limitations of the study, only 5 of the stores (Maastricht, Den Haag, Eindhoven, Rotterdam and Breda) were chosen for further analysis, excluding the smallest store, Almere. Maastricht and Den Haag are chosen from the first cluster, Eindhoven and Rotterdam from the second, and Breda from the fourth.

After the choosing the locations, the styles and colors of merchandise had to be selected. For this research the interesting items are the style color variations that are sold in high quantities and/or at a high value. These items can hurt significantly the revenues of the company if they are not available at the right time, right places and the right quantity. The best seller reports of the merchandise planning department were reviewed and 12 styles were chosen. The 12 styles belong to the categories of blazers, blouses, dresses, pants, skirts and outerwear. The description of the styles is found in the Appendix.

In order to analyze the demand, the supply and the size level OOS rates, a rolling time window of 16 weeks was

chosen. For the 16 week period, the weekly sales data, on-hand inventory, returns, original and markdown

prices and initial base stock levels were collected across all the style/colors/size levels. The 12 styles (on average

with 6 sizes) represent 72 SKUs. When the weekly OOS were calculated, it was counted with the help of an

excel function how many of the 72 SKUs were not available in a given week. It was not possible to reveal the

inventory receipts from the database used by MEXX B.V. Therefore where needed it was calculated by

subtracting the difference between the on-hand inventory of the previous week and sales units of the current

week from on-hand inventory of the current week. In some cases, the receipts occurred as certain SKUs were

returned to the store. In the other cases when it was calculated, it is not known whether it was replenishment

from the warehouse or it was transferred from another store.

10 At MEXX B.V. the current practice is that styles are not markdowned until the 8th week after the styles were introduced in the stores. Therefore the time unit chosen is 16 weeks so that 8 weeks of the full price performance and 8 weeks of the markdowned selling performance can be observed.

In the end, with the help of the descriptive part, the results of the analysis are interpreted. The last chapter of the

results section is about quantifying the size level OOS rates. There some assumptions were made about how

much could have been sold without using unconstraining methods.

11 4. MEXX B.V.

4.1. The merchandise hierarchy

The MEXX approach to retailing is very diverse. Their products are sold in 115 company-owned stores, 168 concession and franchise stores, premium department stores, and 700 shop-in-shops, as well as through mail order catalogs and the e-shop. In all, the company has close to 9000 selling points worldwide.

Figure 1 The merchadise hierarchy of MEXX B.V.

The merchandise hierarchy is depicted in Figure 1. The five product groups (woman, man, youth, accessories and footwear) are divided into departments. For instance, women apparel is divided into the departments of casual and metropolitan collections. This distinction is very recent as it is only used from July 2012 on. Before, there was one single department within the woman merchandise group. The distinction was made to separate casual and metropolitan merchandise in order to simplify the planning process and create a better overview of the processes. The difference between casual and metropolitan collections is that the focus of metropolitan collections is on offering more urban and business style merchandise.

The next level in the hierarchy is the level of classes. Generally for both casual and metropolitan women’s department there are ten classes (such as blazers, blouses, pants etc.).

Within each class there are several styles. Each style is identified by an 8 digit code (e.g. N1ME3042). A style represents at least one and at most five-six color variants; this next level is referred to as style-color level. There are three-digit codes used to describe the colors of styles. Therefore when referring to a style/color variation, the code is 11 digit. This 11 digit code can be found on the price tag of clothes. Finally the lowest level is the style/color/size level. Classes have a different size range. For example skirts are available in sizes from 34 until 42, while pants are ranged from size 32 until 42.

Apart from the merchandise hierarchy, it is also important to denote the difference between fashion and staple products. The reason for this is that demand for certain products are stable whereas for others demand is very seasonal and dependent on current fashion trends. The former classification is often referred to as staple merchandise or basic merchandise and the latter is as fashion merchandise (Levy and Weitz, 2008). Staple merchandise is purchased in larger batches and sold during the complete season, and fashion merchandise is trendier, produced in smaller batches. For instance a basic white T-shirt is considered as staple merchandise while T-shirts in seasonal colors with special prints as fashion merchandise.

Merchandise group

WOMEN

Department CASUAL METROPOLITAN

Class BLAZERS BLOUSES DRESSES PANTS …. OUTERWEAR

Style N1ME3042 N1ME3522 …

Color MIDHNIGHT BLUE WHITE/NATURAL …

Size range XS-XL 34-44

12 The main difference between these two types is that the life cycle of the staple merchandise is significantly longer compared to fashion merchandise which is only available for a relatively short period of time, mostly few weeks or months. As fashion trends change quickly, styles offered by the retailer have to be sold within a short period of time when demand occurs, otherwise it becomes obsolete. Forecasting fashion merchandise is associated with higher risk as often there is no historical sales data about a certain style. When errors are made, the excess inventory has to be cleared by markdowns. The size and timing of markdowns is critical—a too small markdown too late in the season may result in leftover merchandise; whereas too high markdown too early in the season and the retailer sacrifices gross margins (Grewal et al., 1999).

4.2. The merchandize planning process

The merchandise planning process at Mexx B.V. entails both short and medium term planning and reflects both top-down and bottom-up approaches to planning. The planning process starts with generating forecasts based on historic sales at the level of classes. The main goal of merchandise planning is to plan the timing of the introduction of styles and decide where and how much to be made available in order to achieve the overall financial plan. The overview of the planning process is shown in Figure 2. Generally the planning process can be split into the pre-season and in-season phases. The pre-season planning involves all planning phases up to the point when the merchandise arrives at the warehouse. From that point in time, the processes and decisions are part of the in-season planning.

4.2.1. The corporate strategy and the financial planning

The planning process starts with the creation of a financial plan. The financial plan reflects the corporate strategy of the company. The financial targets are expressed as high-level forecasts of sales, target intake margins and market share projections. The financial plan starts with analyzing historic sales performance on the group level. The goals are influenced by the current business environment and the historic performance of the company. Based on these, a budget for the consecutive financial year is created.

At Mexx B.V. there are two product seasons, the spring-summer (SS) and the fall-winter (FW) seasons. Each season consists of six collection months when new products are introduced. The FW season lasts from June until

Corporate Strategy

Financial Plan

Range Planning

Assortment Planning Product Planning Store Planning

Allocation &

Replenishment

Pre-season planning

In-season planning

Figure 2 The merchandise planning process

13 November and the SS season from December until May. The time span of product seasons does not exactly correspond to the financial seasonality as there is some overlap between them. Figure 3 shows the difference between the two kinds of seasons.

Figure 3 The product and the financial season

The changing point of the product seasons is one month earlier than that of the calendar year, therefore during one financial year; three product seasons can be differentiated. Next to these three product seasons, the company also offers so-called NOOS (never-out-of-stock) products. NOOS products represent a continuous season as it is planned for the whole year compared to the seasonal products which are planned for six months. To sum up, within one financial year, products of three seasons and the NOOS products have to be planned.

4.2.2. Product and Store Planning

The product and store planning are the next steps in the planning process after the budgets are generated. Both plans have the same starting point with a different focus. In this phase the financial plans are allocated across the product classes and stores. Figure 4 shows the different levels of focus in product and store planning.

Due to the seasonality of demand, the share of every class is different of the budget throughout a season because of the seasonality of demand. For example, within woman casual department, blazers are sold throughout the whole season but a smaller percentage of the budget is spent on blazers in July than in September. The aim of the product plan is to project the amount of units needed to achieve the financial plans for classes for both on the level of seasons and the total financial year.

The importance of store planning lies in the fact that not all the stores are subject to the same growth. The company however has to make sure that its stores are operating at a profit and it has to provide enough merchandise to meet the sales and growth targets of all the stores. A store plan is a high-level projection of sales on the group level for one financial year.

Figure 4 Product and Store Planning

Months of a calendar

year Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Financial

year Product

season 2013 Spring-Summer

Financial Year 2012/1 Financial Year 2012/2

2011 Spring-Summer 2012 Fall-Winter

Product Planning Store Planning

Time unit week week

Location all retail stores together individual retail stores

Product Hierarchy class level group level

Seasonality within a season for each full year, no seasons

collection month

14 4.2.3. Range planning

Every store faces a different class level demand. Blazers may sell good in one store but less good at another store. Planning the range of classes for each store separately would be a very time consuming activity. Therefore to simplify the process, the stores are grouped according to two attributes: size and turnover. Firstly, they are clustered based on their size. Usually the stores are divided into eight clusters. The bigger the store space is, the higher the ranking assigned to the cluster. The composition of the stores within clusters change with the opening, closing and reconstruction of stores, and is reviewed in every season.

Secondly, within each cluster the stores are further classified into grades based on the selling performance of classes. As there are generally ten classes, accordingly ten grading plans are created. Within a grading plan there are four or five rankings used. The higher the planned class level sales of the store, the lower grade it is assigned to. As a result the option plan is created which contains the ranking of stores by store space and turnover in a descending order. In the option plan for each cluster/grade (each rank) the number of sites is indicated and the monthly budget for a class is split between the rankings.

The planned budget of the stores is summed at the level of classes and ranks. In this phase the buyer and the planner together decides which options of the class to offer in each rank. For example if there are five style variants in a class, the higher ranked may receive mostly all the styles and the lower ranked less styles. The depth of the classes however is intended to be consistent within all the grades. Based on the option plan the buyers decide what styles to buy. There will be a basic collection which is available in every store thus represents all the ranks and there will be options which only the bigger and more popular ranks receive.

4.2.4. Assortment planning

The higher the ranking is in the hierarchy, the wider the assortment is. The grade will influence the number of units of each class a rank receives. There is a clear trade-off the buyer and the planners have to manage together.

The depth and the width of classes are constrained by the available budget and store space.

In the assortment planning, with each collection month, the width of each style within each class is determined for the available rankings. Within the assortment planning, it is also reviewed whether certain styles and colors are to be sold in specific countries or not. As a result of the assortment planning the Final Fabric Production Plan is created. The purchasing consolidates all the material requirements from the wholesale and retail departments and when they don’t reach the minimum levels; they can drop certain styles or add more quantities to certain styles to reach that minimum amount. The Final Fabric Production plan only includes the quantities based on the style-color level and not on the size level. The reason for this is that all the sizes are regarded to have more or less the same material requirements. For the purchase order country-level size profiles for each class are applied.

After the final order is sent, there can be still changes in the mix of classes and ordered amount due to the minimal amounts required by the producers. When it is set the merchandise is produced and one month before it reaches the stores, the allocation stage starts.

4.2.5.Allocation & Replenishment

The in-season planning starts with the allocation of merchandise to the stores. During the season the financial goals and planned sales set in the previous phase are monitored to ensure that they are achieved. The current policy of Mexx is to allocate 80-85 percent of the total stock of a certain style-color to the stores and keep the rest as replenishment. The allocation procedure starts approximately 5 months after the purchase order was generated and it should be ready 2-5 weeks prior to the merchandise is sent to the shop floor. The allocation process takes generally up to 2 weeks. The allocators first check whether there has been any store opening or closing the past 5 months and they review the last 4-5 weeks overall performance of the stores for each classes.

As every month there is a new collection introduced, the performance measurement is always based on the

previous collection month. This means that when allocating the collection of September, the performance of the

August collection is reviewed. Due to changes in performance (i.e. higher or lower store turnover), stores can be

15 up or downgraded. This changes the composition of stores which make up the cluster/grade ranking. Stores with better performance end up with less weeks of supply (WOS) therefore stores with the lower WOS will get a higher ranking. When a store first was planned into the first grade but then it has high weeks of supply (around 10-12 weeks) they are degraded as the display of merchandise is limited. Subsequently it may happen that the planned quantities of classes of rankings exceed or is less than the current quantities planned.

After the regarding is finished, the base stock model is created. Another name for this is the par level. The par level shows the quantities per style/color and size allocated to each store. The software used generates size- profiles based on historic sales data and allocates the quantities per style/color for stores (the size profiles are discusses in 4.2.7.). As mentioned before 80-85% of the total amount of stock is allocated, which can be changed manually in the software. Afterwards it is calculated again how much stock the stores receive. This data is then submitted to the warehouse management system, from where the goods are picked and packed and finally they reach the stores.

After the goods were distributed, the last step left is the replenishment. As only 20% of the stock is held for replenishment, not all stores can be replenished. A so-called priority list establishes the ranking of stores which are to be replenished. This list again is maintained manually by the allocators

The replenishment report is run every night and the system would replenish the stores as long as there is stock in the central warehouse. However there is also manual setting allowed here, as every week the allocators look at the availability SKUs at the stores and the stock available at the distribution center. They can manually push some replenishment to the stores, and add quantities to the base stock model. Pushing replenishment or decreasing the base stock levels can be initiated by the store itself through a request. Every second week there is a store transfer. Stores can contact allocators directly for more units of merchandise. When a query is received the allocator checks the availability of merchandise in the warehouse, if there is no merchandise left in the warehouse, they check with the wholesale department. The third option is to transfer units between stores.

4.2.6. The markdown process

Currently markdowns are planned for all the retail stores in Europe. However, in the near future the company is preparing for the installation of new markdown optimization software. This would base the markdowns on the country levels instead of the aggregate level of retail stores. Budgets are determined in the financial planning phase and are planned for a whole product season. Currently, the markdown budget of Mexx B.V. is around 30- 35%. Every season there are two sales, the mid-season and end of the season sales.

On the style/color level, the policy of the company is that no markdowns can be taken for the first 8 weeks. The style/color performance is reviewed on a weekly basis and if there is high supply of a particular style/color combination, markdowns are taken. Firstly 20-30% and later depending on how the product performs 1-2 weeks later additional markdowns can be taken.

4.2.7. Size profiles

As mentioned before there are two seasons in a year at Mexx B.V. After each season, country level size profiles are created based on the performance of all the classes in the previous equivalent season (i.e.SS2011 for SS2012). The size profiles are created for all the countries where Mexx B.V. is present (such as Germany, Belgium and France) The resulting country/class level size profile are used for the purchase orders. The class level size profiles are based on the first 3-4 weeks of the sales history of all the style/color combinations. The reason for this is twofold. Firstly, the first 3-4 weeks are considered to show the true size ratios, as the probability of a stock-out is assumed to be low. Secondly, it is considered to show the full price demand.

Separate store size profiles are not taken into account because the average single size profile is supposed to make

up for the differences between stores. The size profiles reflect the distribution of sizes and they are created by

16 reviewing the following aggregate indicators for each class and each size: the percentage of units sold, the percentage of initial stock and the sales thru ratios on the country level.

As the purchase order is received, the store size profiles are used to allocate the merchandise. The store-size profiles are based on the full price historic sales of classes of the last corresponding season (e.g. for SS2012, SS2011 is the bases).

5. RESULTS

5.2. Out-of- stock rates

The sample taken (12 styles from 6 classes) resulted in a total number of 72 stock-keeping units (on average each style/color is available in six sizes). A stock-keeping-unit (SKU) reflects a style/color/size combination.

When calculating the OOS rates, only the availability or the non-availability of the SKUs is taken into account.

This is shown by the left graph of Figure 5 where the cumulative OOS ratios of the total sample are illustrated.

The OOS ratio of the total stores in week 4 is 27 % and in week 16, 64%. In other words, by week 4 27% of the SKUs and by week 16, 64% of the SKUs are not available. In other terms, in week 4 the OOS rate equals 20 SKUs and in week 16, 47 SKUs.

As the inventory of the SKUs decreases and the cumulative sales increases, the OOS ratio tends to be the smallest in the beginning of the observed period and increasing toward the end of the period (when there is no replenishment). The left graph of Figure 5 shows the individual and the average OOS ratios with respect to the five locations. The right part of the figure shows the OOS ratios in proportion to the average OOS ratio. The highest OOS rate occurs for Breda and the lowest for Maastricht. To explain the differences in these ratios, both the supply and demand of the stores have to be reviewed as and OOS event is dependent on their interaction.

Figure 5 OOS rates and share\

The left graph of Figure 6 shows the individual inventory movement of each of the five stores. The total

inventory of each store decreases until the end of the observed period. Within the observed period there are

weeks when the inventory is increasing and consequently the pattern of the graphs is increasing (e.g. between

week 6 and 8 the inventory of the store of Maastricht increases from 55% to 65%). The reason for the rise in

inventories is that stores received extra replenishment. Unfortunately inventory receipts cannot be revealed

directly from the database. However, there is one store which was closed down in the observed period and as

explained by the allocators the purchased inventory of the closed store was divided between the rest of the

stores.

17 The right graph of Figure 6 shows the inventories of the stores by their share of the total inventory. The percentages along the X axis indicate the share of a store of the total inventory. The initial percentages are primarily dependent on the store size and the selling performance of the given class (e.g. blazers, skirts) in a given store. For example when demand for blazers in Maastricht is low, its inventory will not deplete as quick as expected and the inventories will be high. When new styles of blazers are introduced to the stores, the inventory levels are reviewed and Maastricht will receive lower quantities than what was initially planned, in order to avoid overstocking.

It can be observed that Breda, the store with the highest OOS rate, has the smallest inventory share. This could explain its high OOS ratios, as the lower inventory levels stock out earlier. As mentioned before the OOS depends on both the demand and the supply of a product. In the case of Breda, it is shown later on in Figure 7 that next to the smallest initial inventory share, the store of Breda faces a similar demand pattern as the other stores.

This figure also shows that at the end of this 16 week period, the five stores end up with an excess inventory, which is on average 27% of the total inventory. From Figure 5, it is shown that by the end of Week 16, on average 15 SKUs are available out of the 72 SKUs. Therefore, the 27% of the total inventory at the end of the 16 week period is held in 15 SKUs. Excess inventory in the end of the season could be sold at further markdowns which may spur the demand for a product and please customers. However when the size required by the customer is not available when the product is available at a reduced price, it could also lead to a negative customer experience.

Figure 6 On-hand inventory

18 Figure 7 shows the weekly sales and the cumulative weekly sales as a share of the total sales of the 16 week period. The left graph shows that the weekly sales graphs reach their peak around week 4 and after that they are decreasing. This can be attributed to the seasonal pattern of demand; however the right curve also shows that until week 4, 45% of all sales are completed. The right graphs also show that sales are strongly increasing until week 6 and afterwards it is increasing with a slower pace. The reason for this could be that on average the OOS rate is around 30% meaning that on average 22 SKUs are not available. At the same time, the on-hand inventory is still 55%. This may mean that although certain styles are very popular, the required sizes are not available of them and therefore sales level off. The availability of the full size ranges is discussed in chapter x.

These three graphs were introduced to show the relationship between them as they are in correlation. If we look at the three graphs together, the general nature of the OOS ratios can be explained; however this unit of analysis is still too aggregate. We could see before that the OOS ratios are varying with respect to locations. However from this level it cannot be seen yet if OOS rates are also varying with respect to sizes at the stores.

5.2 Availability of all sizes on the store levels

The following five graphs depict the availability of the styles when all the sizes are available. The following graphs show for every store the number of weeks when all the sizes are available for each of the 12 styles. The store average is also indicated on each of the graphs. It can be seen that the store average of the availability of the full size range is decreasing as the store size is decreasing. In the stores of Maastricht and Den Haag on average all the sizes of a style are available for 5 weeks, while in Breda on average all the sizes are available for less than 3 weeks.

Every style of the sample sells differently in all the five locations. This is caused by the nature of demand for the styles and also the difference in the supply of the styles. In other words, it could happen that blazers are more popular within a selling season than dresses (which again can be attributable to many factors). The number of the purchased units of the styles also determines the selling performance of the styles.

There are styles that already stock out of certain sizes in the first week of the selling period such as Blazer 1 does in Maastricht, Den Haag and Eindhoven. In these locations the full size range of Blazer 1 is only available for a couple of days. The consequence is that already at the first week due to the non-availability of certain sizes, demand becomes constrained and the sales data does not represent anymore the true size level demand for this style.

Some other styles such as Blazer 2 in Breda start to stock out within the first week while at the rest of the stores the full size range is available until the sixth week. The sizes are available for the longest period in Maastricht and Den Haag, where the sizes start to stock out at week 9, after the markdowns started. In this case (Breda being the smallest store) received fewer initial inventory which stocked out earlier. In Maastricht and Den Haag

Figure 7 Weekly and cumulative sales

19 the customers can choose from all the sizes of Blazer 2 even when the markdowns start. This could increase customer satisfaction and the sales in these two stores. However at the same time if some inventory has been transferred to Breda it could have been sold at a full price there.

Out of the 12 styles and the 5 locations there is only one case when all the sizes are available during the whole 16 week period. It is the outerwear 1 in Maastricht. As for the observed period there is no size level out-of-stock for this style, the sales data represent the true size level demand. The rest of the location-style variations, the sales data are truncated.

To sum up, looking at the sales data without looking at the availability of inventory can cause biases in understanding the selling performance of styles.

5.3. OOS on the store- style- size level

After a certain point in time, as demand reaches its peak, fragmentation of merchandise starts, and some sizes are not be available. In the following paragraphs the size level OOS ratios of the five locations are analyzed and

Figure 6 Availability of the full size range

20 discussed. The discussion will go into detail about two styles, Blazer 1 and Blazer 2. The rest of the styles can be found in the Appendix.

Figure x the OOS ratios of Blazer 1 on the store and size levels. The vertical dotted line at week 8 signs the beginning of the markdown period. The blue dots on the figures indicate when an OOS occurred of a certain size. Out of the 12 styles, Blazer 1 was allocated in the smallest amount compared to the other styles (42 pieces compared to the average of 79). It can be observed that OOS events occur already in the first six weeks for most of the SKUs. Because of this, sales and the corresponding revenues were not maximized. Although most of the sizes of this style are sold out very quickly, two SKUs in four locations are available until the end of the observed period. One of the SKUs, size 44 is available in Maastricht, Rotterdam and Eindhoven until week 16.

The same SKUs are out of stock in Den Haag and in Breda after some weeks. If these units (namely 4 pieces) were allocated to either Breda or Den Haag, they could have been sold even at the full price. In Rotterdam size 44 is not available from the 2

until the 4

week. On the 5

week it is again available, however it is not sold within the observed period. The other SKU is the size 40 SKU. In Eindhoven and Breda, they are out of stock at the third week but the following week they are available again. In Breda the reason for the increase in the SKUS is that two items of size 40 were returned. In case of Eindhoven (and the other locations) there were no returns recorded in the database. Thus, when a certain SKU is out of stock for a certain number of weeks and then becomes available again (such as in Maastricht size 36) is due to replenishment. It is probable that only some sizes were available in the warehouse as not all the sizes were replenished in the stores.

However the replenishment could have happened too late as there was no demand for the SKU even though they

became available again. Had these items been transferred to another store or had they been available when

needed, they could have been sold at a full price.

21 Figure 9 shows the store level and size level OOS ratios with respect to Blazer 2. It can be seen that this style performs differently than Blazer 1. Blazer 2 was purchased in a higher quantity than Blazer 1. Most of the size levels OOSs occur after the markdown period started which could imply that this style did not sell too well at the full price but rather at markdowned prices. It is interesting to point out that is the smaller stores which stock out earlier. The current allocation of quantities is dependent on the store size, however for example this does not mean that Breda faces a significantly lower demand than Maastricht.

The intention of this subchapter was to show that the size level OOS rates are dependent on both the stores and the styles. Popular stores and popular styles sell out earlier than other stores and other styles.

5.4. Quantifying the size level OOS costs

Without going into the details of the unconstraining methods, this subchapter aims to quantify the costs associated with the size level OOSs. From the above analysis it can be concluded from the five locations Breda faces the highest OOS rates from the five stores and from the twelve styles Blazer 1. Therefore it is expected that this store-style variation can face the highest costs associated with the size level OOSs.

Figure 10 Blazer 2 OOS rates

22 Figure 11 shows that only one size out of the six sizes is available at the store of Breda until the end of the sixteen weeks. However the sales (demand) data is constrained for that size level SKU as for one week it was out of stock. By the fourth week, all the sizes are out of stock in Breda except for size 40. In the following chart the costs of size level OOS are calculated based on the simple projection of the sales graphs.

Breda Blazer 1

34 36 38 40 42 44

lost sales units (fullprice) 1 1 1 1 1 1

in EUR 103 103 103 103 103 103

lost sales units (markdowned price) 1 1 1 0 1 1

in EUR 83 83 83 83 83 83

Total lost sales in EUR 1116

full price EUR 129 markdown price EUR 109 cost price EUR 26

Sizes Figure 11 Size level OOS of Blazer 1 in Breda

Figure 12 Size level OOS costs

23 Figure 12 shows some simple calculations which are based on the assumption that sales could have increased if the supply was not constrained. It can be seen that if there was enough supply of the sizes of Blazer 1 in Breda, the company could have increased its revenues by 1116 EUR.

Although the method used was very simple, it illustrates the importance of unconstraining size level sales data.

All the styles were reviewed in the five locations. The lost sales were calculated by reviewing the size level OOS rates of each store of each style. When the OOS events occurred after the markdowns started, it was calculated how the company could have earned if one more unit of the SKU is available. When the OOS units occurred in the beginning of the observed period, it was calculated how much revenue was lost if one extra unit could have been sold at full price and another one at the markdown price.

Figure 13 summarizes the costs associated with the size level OOSs. For the five store, the cost of lost sales is

17892 EUR.

24 6. DISCUSSION AND CONCLUSION

It is generally noted that the costs of acquiring a new customer exceeds retaining an existing one (Matzler and Hinterhuber, 1998). Therefore to remain profitable it is very important for retailers to keep their customers loyal.

If retailers fail to provide the right items at the right store, it can not only harm the revenues of the store but also customer satisfaction, brand loyalty and store image (Aastrup & Kotzab, 2010). Other long term negative impacts were found to be loss of market share, negative word of mouth, loss of patronage (Fitzsimmons 2000, Zinn and Liu 2001).

The thesis aimed at identifying the size level OOSs and the associated costs. It was found that on average already by the fourth week after the styles were introduced to the shops, almost one-third of the sizes are out of stock. By week 8 this ratios is around 45%. Currently Mexx B.V. uses the sales data of the four and the eight week period in order to plan the purchase orders and the size profiles of stores. However such high percentages of OOS rates mean that in the 8 week period, already 45% of the sales data is truncated. OOS events vary across locations, styles and sizes, in order to capture the real demand these OOS events should be analyzed on the levels of stores and sizes.

The on-hand inventory analysis reveals that the five stores together carry one –third of the total allocated inventory at week 16. As the items become obsolete and usually after 8-9 weeks they are markdowned, it is likely that those items have to be sold at an even lower price (maybe lower than their cost price). However when correct measures are used to forecast size-level store demand, the high inventory at the end of the season could be avoided. As the styles included in the sample were bestseller and there was high demand for them, the high on-hand inventory can be due to the fact that one-third of the SKUs were allocated to a store where they represented a size for which there was low or no demand. If some of those sizes could have been transferred to another store (or were initially allocated to another store), they could have been sold.

OOS events are not necessary a bad condition as it means that a style was popular and it was sold out. However in the sample it was observed that size level OOSs occur already in the first 3 or 4 weeks of the selling period.

As styles are just introduced there, it is crucial that the full size ranges are available at the full price selling period.

It was mentioned in the introduction that in order to simplify the allocation decisions to stores, based on their size the stores are clustered. However in the sample, Breda is clustered as the smallest store, therefore it receives fewer inventories. If the corresponding demand around Breda would be also smaller, the OOS rates would be more or less the same as for the other stores. However in this sample, Breda faced the highest OOS ratios, which could be due to the fact that it is underallocated and there is more demand than what is right now supplied. This is in line with the proposition of Grewal et al. (1999) who state that comparing performance of stores based on aggregate measures can provide false conclusions about the real performance of stores.

As markdown budgets are increasing in order to meet target margins, Mexx B.V. purchases more units as it is

expected that 30% of the merchandise have to be sold for a reduced price. With unconstraining sales data, more

precise information about demand can be learnt and demand can be better matched. This could mean the need to

purchase more units in order to meet target revenues could become unnecessary. In other words, this could mean

that the retailer does not have to purchase 130.000 units of women’s group merchandise (as it expects 30% to be

markdowned), but it is enough to allocate 100.000 units which are sold during the selling period.

25 The study has several limitations. Firstly, there are a limited number of styles studied in a single period. Future research could reveal more information about demand around stores by comparing multiple seasons. The styles are from different months introduced but they were analyzed without further observation of seasonality effects.

Because of the POS scanners, there are great amounts of data available. However if the data are only used on an

aggregate level, many possibilities remain unlocked. Analyzing data on a disaggregate level does not only help

in creating the size profiles but several other analyzes can be conducted with regards to the colors, materials and

other attributes.

26 7. REFERENCES

Carey, A. and Staniforth, J. (2007), “Improving availability at House of Fraser; availability and demand planning”, paper presented at ECR UK Conference, 21 March, London.

Cooper, W. L., Homem-de-Mello, T. and A. J. Kleywegt. (2006). Models of the spiral-down effect in revenue management. Operations Research, 54 (5) 968–987.

Dempster, A. P., N. M. Laird, and D. B. Rubin. (1977). Maximum likelihood from incomplete data via the EM algorithm. J. Roy. Statist. Soc. B 39, 1–38.

Fernie, J., and Corcoran, L. (2011). Responses to out-of-stocks and on-shelf availability in UK fashion retailing.

International Review of Retail, Distribution & Consumer Research,21, 309-322.

Fisher, M. L., Raman A., Hammond J H., and Obermeyer, W. R. (1994). Making Supply Meet Demand in an Uncertain World. Harvard Business Review, 72 (3), 83-93.

Friend, S.C., and P.H. Walker. (2001). Welcome to the new world of merchandising, Harvard Business Review 75,133-139.

Grewal, D., Levy, M., Mehrotra A. and Sharma, A (1999). Planning Merchandising Decisions to Account for Regional and Product Assortment Differences. Journal of Retailing, 75 (3), 405-424.

Keaveney, S. M. (1995) Customer Switching Behavior in Service Industries: An Exploratory Study, Journal of Marketing, 59(2), p.71

Kumar, B., and Banga, G. (2007). Merchandise Planning: An Indispensable Component of Retailing. ICFAI Journal of Management Research, 6( 11),7-19.

Lan Y., Gao, H., Ball,M.O., and Karaesmen, I. (2008). Revenue management with limited demand information, Management Science, 54 (9), 1594–1609.

Levy, M., and Weitz, B.A. (2008). Retailing Management McGraw-Hill Higher Education

Mantrala, M.K, Levy, M., Kahn, B.E., Fox, E.J., Gaidarev, P., Dankworth, B., et al.(2009). Why is assortment planning so difficult for retailers? Journal of Retailing, 23(1),71-83.

Matzler, K., and Hinterhuber, H.H. (1998). How to make product development projects more successful by integrating Kano’s model of customer satisfaction into quality function development. Technovation, 18 (1), 25–

38.

Queenan, C. C., M. Ferguson, J. Higbie, R. Kapoor. (2007). A comparison of unconstraining methods to improve revenue management systems. Production Operations Management, 16(6), 729–746.

Vinod, B. (2005). Retail revenue management and the new paradigm of merchandise optimization. Journal of Revenue and Pricing Management, 3(4), 358-368.

Zhu, J. (2006). Using turndowns to estimate the latent demand in a car rental unconstrained demand forecast.

Journal of Revenue and Pricing Management, 4(4), 344-353.

Polt, S. (2000). From Bookings to Demand: The Process of Unconstraining, in Proceedings of AGIFORS Reservations and Yield Management Study Group, AGIFORS, New York.

Phillips, R.L. (2005). Pricing and Revenue Optimization. Stanford University Press