study” “ Designing a variant of the can-order replenishment policy: A case-based simulation

(1)

MASTER THESIS TECHNOLOGY & OPERATIONS MANAGEMENT AND SUPPLY CHAIN MANAGEMENT

“Designing a variant of the can-order

replenishment policy: A case-based

simulation study”

by

ROBIN MEULENBROEK

University of Groningen Faculty of Economics and Business

Primary supervisor: dr. J.A.C. Bokhorst

Secondary supervisor: dr. S.A. de Blok

r.j.meulenbroek@student.rug.nl Student number: 2043459

Abstract

(2)

2

Preface

First I would like to thank my primary supervisor, dr. J.A.C. Bokhorst for his support and commitment during this project. Our meetings have been most helpful in achieving progress throughout the project. Moreover, I would like to thank my secondary supervisor, dr. S.A. de Blok for her feedback and her input regarding the overlap of the thesis with the Supply Chain Management Master.

(3)

3

1 Introduction

The grocery retail industry is characterized by large product volumes, low margins and fierce competition and is therefore constantly seeking efficiency improvements in its supply chain (Saghir & Jönson, 2001). The trend towards more product variety and short response times calls for smooth and efficient logistic operations (Rouwenhorst et al., 2000). Moreover, retail chains are still losing potential revenue by not delivering the right goods, to the right places, at the right time, despite large investments in infrastructure and IT (Hübner, Kuhn, & Sternbeck, 2013).

One source of inefficiency in supply chains is workload variations (Sainathuni et al., 2014). Large variations in daily workload in a distribution centre (DC) can have a detrimental effect on the entire supply chain and could cost a company significant amounts of money annually (Sainathuni et al., 2014). For instance by not being able to fulfil orders at the specified time and by having trucks waiting at the DC to be loaded and unloaded. This might affect the competitiveness of the company or even the entire supply chain (Rouwenhorst et al., 2000), due to an increase in transportation costs caused by these idle trucks. A major cause of workload variations in the food industry are consumer promotions (Fisher, 1997), which is also the main cause in the case company in this study. Consumer promotions turn simple predictable demand into a chaotic series of spikes, which only increases costs (Fisher, 1997). Promotions are an undesired strategy from an operations perspective, as they are notorious for creating disruptions in logistical systems (van den Berg, 2007). But they are often a desired strategy from a marketing perspective as they are generally liked by customers (Blattberg et al., 1995).

One way to deal with workload imbalances, which is also the current practise at the case company in this study, is by ensuring a sufficiently high capacity to cover the peak demands. This results in under-utilization of resources during non-peak times, which is essentially unused capacity and results in excess logistics costs (Harmon, 1993). It furthermore leads to requiring high flexibility in the workforce, and potentially to high amounts of overtime. Another way of dealing with workload imbalances is by postponing orders in a busy period, thereby reducing the need for (expensive) overtime and temporary staff (van den Berg, 2007). This is however often not possible, as orders cannot always wait to be fulfilled. Preparing orders in advance during non-peak times is furthermore a strategy to deal with workload imbalances (van den Berg, 2007). This strategy is essentially the opposite of postponing orders. However, this generally requires significant amounts of space available to store the orders prepared in advance. Since space is often limited in warehouses, a different method for balancing the workload in a DC should be found.

(6)

6

replenished by the next delivery, the ‘must-orders’, but also determines the products that already have sufficient space on the shelves available to restock, the ‘can-orders’. Both order types are replenished till the ‘order-up-to-level’ (Johansen & Melchiors, 2003). In the periodic replenishment policy, after each period, every item is replenished till the ‘order-up-to-level’ (Melchiors, 2002). Section 2 will elaborate on these models. The shortcomings of these models is that they assume the perspective of the buyer (retailer), instead of the perspective of the supplier (DC). By taking the perspective of the retailer, the focus is solely on achieving the best results for that retailer. By including the perspective of the DC, the focus can be on achieving the best results for the entire supply chain. By satisfying the demand of the retailers while achieving more favourable results for the DC.

The aim of this study is to develop a model that achieves better balanced workloads, while taking downstream effects into account. This will be achieved by designing a variant of the JRP (the can-order policy) and test this with a discrete event simulation model that, based on real-world demand data of a grocery retail chain, simulates the customer demands and store replenishments. The performance criteria in this model are not the traditional inventory and transportation costs, but the balance in workload, the number of out-of-stocks (OOS) and the number of products stored in the backroom of the stores.

This study contributes to the theory by developing a model of the JRP which can be applied by the supplier (the DC) in association with the buyers (retailers). This can be enabled through closer cooperation within the supply chain thanks to Collaborative Planning, Forecasting and Replenishment (CPFR) tools. While applying the JRP in this way might lead to significant benefits throughout the supply chain, it has not been examined yet in the literature. It furthermore delivers a managerial contribution by proposing recommendations to companies that want to increase their logistical efficiency, but are facing problems with imbalance of workload in their DC and idle capacity at some days, while having insufficient capacity on others. The findings of this study can help managers facing similar problems to overcome them. This will be achieved through a case-based simulation study to answer the following question;

How can the joint replenishment problem contribute to a better supply chain performance?

(7)

7

2 Theory

2.1 Characteristics of the grocery retail industry

In grocery retail it is often common practice to determine fixed delivery days for each store, combined with an automated inventory replenishment system to determine the number and type of products to order (van Donselaar, et al., 2010). This holds especially for retailers with smaller stores, which do not need to be replenished every day. By determining fixed delivery moments, routing decisions can be fixed for a certain period, eliminating the (expensive) need to make new routes every day. It furthermore provides consistency to retailers by knowing when to schedule sufficient personnel to unload the truck and fill the store shelves.

Supply chain management (SCM) is an important topic in the grocery retail industry because of low margins and fierce competition (Saghir & Jönson, 2001). According to Fearne (1996), SCM ‘seeks to achieve a relationship of mutual benefit by defining the organisational structures and contractual relationships between buyer and seller, which up to now have been classified as adversarial’. SCM seeks to achieve higher service levels while at the same time substantially reduce costs (Fearne, 1996). One powerful way to achieve good SCM is the integration of business logistics systems with logistics systems of suppliers and customers using collaboration methods like Collaborative Planning, Forecasting and Replenishment (CPFR) (Branska & Lostakova, 2011). Three levels of CPFR can be distinguished, based on the level of integration and the extent of collaboration (Skjoett-Larsen, Thernøe, & Andresen, 2003). When only limited integration with trading partners exist, for instance when solely stock level data is exchanged between supplier and retailer, basic CPFR is applied. Developed CPFR is applied when an increased level of integration is achieved, for instance by exchanging stock level data and forecast data, a good example is that retailers hand over replenishments to the supplier. Advanced CPFR takes it even further by co-ordinating processes within forecasting, replenishment and planning. Advanced CPFR might for instance also include planning processes related to production, product development, transport and marketing (Skjoett-Larsen, Thernøe, & Andresen, 2003). A large difference between the three levels of CPFR collaboration is the type of relationship between the supply chain partners. It moves from ‘transactional’, to ‘information sharing’, to ‘mutual learning’ when moving from basic, to developed, to advanced CPFR. It is argued that to be able to implement a system which utilizes stock levels of multiple supply chain partners to achieve better performance throughout the supply chain, the supply chain partners need to be at least in a developed CPFR-relationship.

2.2 Distribution and replenishment optimization strategies

This part covers part of the literature on distribution and replenishment optimization models. It is split up into two parts. The first part covers routing models (2.2.1), followed by the JRP (2.2.2).

2.2.1 Routing models

(8)

8

literature. While each of these models take the perspective of the supplier (DC) in determining optimal distribution and replenishment strategies in a supply chain, all of them have their shortcomings. Table 2.1 summarises the pros and cons of each model reviewed. The most relevant models to this research are the ones by Gaur and Fisher (2004) and Ronen and Goodhart (2008). Although they achieve better balanced workloads in the DC over the days, they might both not be the best applicable model to balance the workload in the DC, while also taking the rest of the supply chain into account. The workload in the DC might become well-balanced, but because of the capacity constraints on the DC, the downstream part of the supply chain might experience detrimental effects when these models are applied. The next section covers the JRP.

2.2.2 Joint Replenishment Problem

(9)

9

including capital tied up in inventory, taxes and insurance. By coordination of replenishment orders, considerable cost savings can be achieved by reduced ordering costs, reduced freight rates, reduced handling costs or quantity discounts (van Eijs, 1994).

The literature is mostly divided in two classes of policies, the continuous ‘can-order’ policy and the ‘periodic replenishment’ policy (Johansen & Melchiors, 2003). Balintfy (1964) was the first to introduce the can-order policy. In the can-order policy, the products that need to be replenished by the next delivery, the ‘must-orders’, are determined. Next to that, the products that already have sufficient space on the shelves available to restock, the ‘can-orders’, are determined. Can-order policies are characterized by three parameters (Si,ci,si) for each item i

(Khouja & Goyal, 2008). The levels of inventory are continuously monitored. When the inventory position of item i is at or below the ‘must-order level’ si, it will trigger a replenishment

order. Other items are inspected, and any item j at or below the ‘can-order level’ cj is included in

the joint replenishment. The inventory position of all items in the order are raised up to its ‘order-up-to-level’ Si (Khouja & Goyal, 2008). The height of the order-up-to level is in the grocery retail

industry most often determined by the shelf capacity (Trauzettel, 2014). The actual number of products that are replenished might furthermore depend on whether the product can be replenished based on individual units or on case packs. For instance, if the case size is 6 and the maximum number of products that fit onto a shelf is 8 (shelf capacity); if there are 2 units left on the shelf, 1 case of 6 units is ordered. However, if there are still 3 units left on the shelf, no case will be ordered (Trauzettel, 2014).

Figure 2.1 shows an example of the behaviour of a two-product system under the can-order policy. The figure shows that when product 1 has reached its ‘must-can-order’ level at ‘t1’,

product 2 is below its ‘can-order’ level. Product 2 is included in the replenishment, even though product 2 has not reached its ‘must-order’ level yet. Together with product 1, product 2 is restocked till its ‘order-up-to’ level. Figure 2.1 furthermore shows that when product 2 reaches its ‘must-order’ level at ‘t2’, product 1 is not included in the replenishment order because it has not

(10)

10

note that although the access to information may be continuous, there is often a limited number of replenishment opportunities, a few times a week for instance.

The periodic replenishment policy was introduced by Atkins and Iyogun (1988) and at its base is a (R,T) type policy. Which means that every T periods, item i is raised to Ri. This entails

that a retailer orders products on fixed intervals, at which every product is replenished till its order-up-to level. Viswanathan (1997) proposed a different periodic replenishment policy, a P(s,S) policy, which outperforms the standard periodic replenishment policy. In the periodic review (si,Si) policy, the inventory of every item i is ordered up to the level Si, if its inventory

position is less than or equal to si at the time of review (Viswanathan, 1997). The interesting

thing about the periodic replenishment policy is the fact that replenishment orders occur at fixed intervals, which is often common practice in the grocery retail industry (van Donselaar et al., 2007). The disadvantage of the current periodic replenishment models is that there is no differentiation between products that really have to be reordered, and products that could wait till the next delivery. In every order, each product, or each product below a certain level (si) is

restocked till its order-up-to level. By combining the can-order policy with the periodic review policy, the best of both models is incorporated into a single model. This can either be achieved by determining the can-order policy for periodic replenishments, like Johanson and Melchiors (2003). Or it can be achieved by adding a parameter to the periodic replenishment policy, say mi,

to represent the ‘minimum si’ or ‘must-order level’. The chosen terminology for the model

developed in this study is a variant of the can-order policy, while it is actually a combination of the can-order policy and the periodic replenishment policy.

(11)

11

backordered in these models. In the case of fixed deliveries, an individual product is generally not backordered between two consecutive replenishments, OOS’s are simply replenished with the next delivery. This does not mean that the JRP could not contribute to a better performance in this type of supply chain, it simply means that the current JRP models in the literature are not directly applicable.

Over the years various variations on both policies have been proposed, mainly trying to determine the optimal solution for the JRP, which entails the lowest combination of transportation, set-up and backorder costs. However, as Khouja and Goyal (2008: 14) state; ‘the

quest for the algorithm guaranteeing the optimal solution of the classic JRP has passed the saturation point and the time has come when we should concentrate on developing applicable models of the JRP for the real life inventory problems’. Therefore, to apply the can-order policy

(12)

12

This leads to the following research question;

How can the workload in a distribution centre for non-perishable groceries be better balanced through applying a variant of the can-order replenishment policy throughout the supply chain? While taking the downstream effects into account.

The following sub-questions help to answer the main research question;

A. How does the height of the ‘can-order’ level influence the performance of the can-order policy in terms of the three performance measures?

B. How does the height of the ‘must-order’ level influence the performance of the can-order policy in terms of the three performance measures?

(13)

13

3 Problem definition

This chapter first introduces the case company Coop (3.1). Subsequently, the application of the can-order policy to Coop is covered (3.2). And finally, consumer promotions are discussed (3.3).

3.1 Supply chain network Coop

The distribution centre of the supermarket chain Coop in Gieten (referred to as DC Gieten for the remainder of this report), is the main DC of Coop for dry groceries (non-perishables). They serve each of the 254 Coop stores in the Netherlands, alongside the DC in Deventer, which handles all perishable and frozen groceries (see figure 1 in appendix 2). A total of 162 full-time employees work at DC Gieten, supplemented by a varying amount of flex workers (190 in total). Replenishments are performed from Monday through Saturday and multiple stores can be replenished on one route. DC Gieten originated from a merger of two DC’s of Coop and was taken into operation in September 2014. The company’s head office is located in Velp (see figure 3.1 for the supply chain). Practically every store is replenished every day by DC Deventer (except on Sundays). DC Deventer is not facing large workload variations over the days because of the regularity in delivery frequency (daily). DC Gieten however is facing large variations in workload over the days because DC Gieten only replenishes the stores one to six times a week, depending on the demand of the stores. The average replenishment frequency from DC Gieten is three times a week (see table 3.1). This entails that the stores have just a limited number of replenishment opportunities, which makes a periodic review policy more applicable than a continuous review policy. Coop employs two different ‘formulas’ of stores as they call it (1) Coop supermarkets, which are the regular supermarkets owned by Coop, and (2) CoopCompact supermarkets, which are smaller supermarkets or ‘convenience stores’ which are run by independent entrepreneurs on a franchising basis. DC Gieten can be marked as a ‘Distribution warehouse’ with a manual warehousing system, the picking activities in the DC, the process of

retrieving products from storage in response to a specific customer request (de Koster, Le-Duc &

(14)

14

Each Coop store places their orders at both DC’s based on the days they have been assigned and are allowed to order on. A growing number of Coop stores are connected to SAP Forecasting and Replenishment (F&R). At the moment of writing, 214 stores are connected, with the intention of connecting each of the 254 Coop stores by mid-2016. F&R recommends the order quantities for the stores, the store managers do have the possibility to bypass the system and make adjustments if they feel is necessary. The SAP F&R tool is a good example of CPFR, by implementing this, Coop has made a step closer to increased supply chain integration. Coop can be characterized as having a developed CPFR-relationship with a large number of the retailers, because stock level data and forecast data is exchanged, and the replenishment is performed mostly by the head office. The first stores that employed this system achieved an increase in service level (fill rate) of over 2%, a decrease of in-store inventory of 20% and a decrease in OOS’s of 50%1, compared to without using the SAP F&R system. A parameter of SAP F&R is the number of replenishments per week, which are determined by rules of thumb. Each store is replenished on one or more prescribed days of the week. The delivery frequencies obtained from the rules of thumb are dependent on the average number of ‘case packs’ a store requires each week. A case pack contains a number of units of the same product, the quantity of products and the size of the case pack differ per product (Waller, Heintz Tangari, & Williams, 2008). A ‘case pack’ is a standard unit of measure and could take up significant space (like toilet paper or bags of crisps), or take up little space (like small roll-on deodorants). In the DC, the case packs are manually packed on roll trolleys which can contain an average of 35 case packs. These roll trolleys are loaded into trucks and delivered to the retailers. Table 3.2 shows the delivery frequency rules of thumb measured in the average number of case packs a store orders each week at DC Gieten.

1

(15)

15

The main problem DC Gieten faces is an imbalance in workload amongst the days. The workload in the warehouse is quantified by the ‘number of case packs picked per day’. The day with the highest workload for the DC is Friday, with an average of 125.744 case packs picked. While the least busy day, Monday, only has an average of 72.592 case packs picked a day (see figure 3.2). Table 3.3 provides an overview of the number of stores that are replenished by DC Gieten per day of the week, it furthermore shows for how many stores the orders are picked each day, as not all orders are picked and shipped the same day. As mentioned in the introduction, the current practice to ensure the demand of every store is met, is by ensuring a sufficiently high capacity to cover the peak demands. Which leads to excess logistics costs (Harmon, 1993), high flexibility requirements in the workforce and to high amounts of overtime. Moreover, because the workload in the warehouse is not well balanced over the days, congestions in the warehouse occur. Order pickers are waiting on each other to pass or to pick products. These workflow congestions are a major concern because, next to lower picker efficiency, it can result in immediate safety problems: damaged uprights, damaged vehicles and damaged workers (Tompkins et al., 2003). Although DC Gieten was built for growth, the DC is currently barely able to meet the demand. Because the demand and even the number of stores are steadily growing, the DC will likely not remain able to meet the demand in the future. It furthermore occurs that trucks of suppliers are having to wait at the DC to unload their cargo, because the employees in the warehouse are too busy with outbound order preparation. Which leads to dissatisfied suppliers of the DC and to a decrease in performance of the supply chain.

3.2 Application to Coop

(16)

16

image requirements (SIR). This is the minimal number of products the store owners want on their shelves to ensure the shelves do not look empty, as empty shelves deter customers. Order forecasts are made by SAP F&R. During the week, the program knows how many of which products will likely be sold, based on past demand data. Therefore it knows when stock levels of products will likely drop below the SIR level. The stock level should never reach below this level. When this will occur between successive replenishments, the program will include these products in the replenishment order. The program furthermore has a ‘balancing’ function, which essentially tries to balance the store replenishments by trying to create replenishments of even sizes. When each store receives equally large replenishments over the week, the workload in DC Gieten should in theory be balanced over the days. Figure 3.2 shows however that this is clearly not sufficient.

The goal is to search for products that do not have to be replenished yet, but have sufficient space on the shelves available, and include these in replenishment orders on less busy days. The result of this is that these products do not have to be included in replenishments on busier days. Which essentially moves a part of the workload towards less busy days, and thereby relieving busier days of a part of the workload. This entails that the policy determines which items need to be re-ordered, the ‘must-order’ products, and which products can be re-ordered with it, the ‘can-order’ products. A number of scenarios will be created to show the potential benefit of this policy, together with the effects of different parameter values of the can- and must-order levels and of the shelf capacity.

3.3 Consumer promotions

(17)

(18)

18

4 Methodology

This section describes the choice of the research method and tools, the simulation model itself and the analyses performed on the outcomes of the experiments.

4.1 Simulation

This research provides a case-based simulation study within the case company Coop. A simulation model was used to test whether a can-order replenishment policy could be a better alternative in the non-perishable grocery industry than the current replenishment policy, in terms of warehouse workload variation. The designation ‘non-perishable grocery industry’ is an important one to make, since applying a can-order policy to perishable groceries will be significantly more limited because of the limited storage life of these products. Simulation was chosen, since it can deal with variability, interconnectedness and complexity (Robinson, 2014), all of which were present in this case. Moreover, with simulation it is possible to conduct experiments with the model, to evaluate the effect of various strategies on system performance (Shannon, 1975). This creates the opportunity to create a number of scenarios to determine the most appropriate strategy. This is also confirmed by Terzi and Cavalieri (2004), who argue that simulation is a powerful tool supporting a multi-decisional context, as a supply chain is. Logistics related problems are often analysed by the use of simulation, while the JRP is mostly researched through mathematical modelling and heuristics (Khouja & Goyal, 2008). Simulation can provide a good way to overcome the restrictive assumptions in mathematical inventory models (Köchel & Nieländer, 2005). The last advantage of simulation, is that it provides more transparency to its stakeholders than for instance mathematical modelling (Robinson, 2014: 15). This helps to communicate the outcomes of the model and to convince stakeholders of the benefits of the solution.

The simulation model has been built with Tecnomatix Plant Simulation 12.02 from Siemens. Which uses a discrete-event simulation approach based on SimTalk. Plant Simulation provides the necessary functionality to model, analyse, and maintain large and complex systems in an efficient way (Bangsow, 2010). The main reasons for selecting this software package were, the suitability of experimenting with different parameters and alternatives, prior knowledge of the program and the availability of a license.

4.2 Data collection

Data requirements in any modelling exercise can be split up into three types of data according to Robinson (2014). (1) Contextual data, data that are required for developing a thorough understanding of the problem situation. (2) Data required for realizing the model, for instance detailed data on activity times and demand patterns. And finally (3) data used for validation.

(19)

19

clarify the processes, and verify the problems within the DC (Voss, 2009). Due to the limited time available, this research mainly made use of category A data, data that are available (Robinson, 2014). Category B data, not available but collectable data, was treated as category C data, not available and not collectable (Robinson, 2014), or an assumption was made, since collecting this would consume too much time. Data required for realizing the model were mainly gathered from the SAP system and the Access database of Coop regarding the ordering and delivery schedule. The collection of this type of data also required face-to-face semi-structured interviews, since not all data were readily accessible. The advantage of doing this face-to-face is that the interviewee can receive greater detail about which data are required (Voss, 2009). This takes away confusion and makes it more likely the right data is provided. A disadvantage is that receiving an appointment for these interviews can be time-consuming (Voss, 2009), therefore these interviews have been semi-structured to ensure no data requests were forgotten. Data collection for validation is discussed in section 4.6.

4.3 Data analyses

Daily demand information of the last year was retrieved from the SAP system from 17 different stores of 20 different products of the planogram ‘soda’. The planogram ‘soda’ has been chosen because of the diversity of products, from fast-movers like Coca Cola, to slow-movers like the store brand ice tea. This ensures the most realistic scenario was simulated, and might give the possibility to analyse differences amongst the different types of products. The 17 stores have thereafter been chosen because these stores all shared the composition of products in their planogram. Other stores did not have the same composition of products and shelf capacities, which would increase the difficulty in modelling. Other information like the ordering schedules and dates of the promotion weeks were retrieved from the Access database. The large variation in the ordering schedules contributed to a more realistic model as well (see appendix 3 for the ordering schedules). The dataset was limited to 20 products because of the time-consuming task of the replenishment specialist to retrieve all the data from the systems. Because for some products only data of a few weeks or months were available, and because of polluted data, only 12 products have been included in the simulation model. These products did still vary from fast-to slow-movers. As a consequence of the limited subset of all sfast-tores and products, the results have limited generalizability, but this composition of 17 stores and 12 products were deemed sufficient for the exploratory nature of this study.

(20)

20

MATLAB, see appendix 4 for details. The following sections discuss the assumptions made and a description of the actual simulation model.

4.4 Assumptions

A number of assumptions had to be made to simplify the situation and to focus on the most important aspects of the problem.

 Deterioration and obsolescence were assumed not to be of influence (not occurring), since this research is conducted on a dry-grocery DC.

 Every time an order was placed, those products arrived on the next delivery. Meaning that the DC always had every product in stock, and all orders were picked without errors. By incorporating stock-outs at the DC, the model would be subject to unnecessary variability, which deflects the focus of the most important aspects.

 Backorders were not taken into account. Backorders only occur within Coop when the DC made a mistake with the replenishment order. Since this only happens sporadically, it was assumed not to be of influence on the normal daily operations.

 Imperfect quality was not modelled in the simulation. It might occur that products do not safely make it to the stores shelves, this was not taken into account, since this will not be influenced by the outcomes of this research and will not have an influence on the outcomes of this research.

 OOS’s were determined based on the modelled customer demand and availability of that product in the store. Differences in theoretical OOS and actual OOS, which might be caused by theft, by dropping items in the stores or by miscalculations during inventorying, were not taken into account. Insufficient data were available to correctly model this.

 Heterogeneity of truck fleet, the fact that not every store can be replenished by every truck type mostly because of space limitations, was not taken into account, since this is not changed by this replenishment policy. Applying this policy will probably change the number of products per replenishment, which might imply that on certain days additional trailers will be required while on other days less trailers are required. But the routes of these trucks will not change, so it adds no value to include truck heterogeneity.

4.5 Simulation model

(21)

21

the standard deviation in workload, the number of OOS’s in the stores, the number of products that need to be stored in the backrooms of the stores, and the average daily workload per experiment. The standard deviation indicates whether the workload is balanced over the days of the week. The lower the standard deviation, the better balanced the workload is and vice versa. Appendix 5 contains the stores and the products included in the simulation. Appendix 6 contains the conceptual model of the simulation, including the experimental factors and the outputs.

(22)

(23)

23 4.6 Model verification and validation

Before the model development began, interview sessions with various replenishment employees and managers were conducted to ensure conceptual model validation. The main theme was the scope of the simulation model, and which data would be possible to retrieve from the information systems. To ensure data-validity, all data were retrieved by the replenishment specialist and checked for inconsistencies. Regular discussions mainly with the replenishment specialist ensured that the model was validated throughout the life-cycle of the simulation study (2014: 255). Further verification and white-box validation were mostly performed throughout the project by checking the code, visual checks and inspecting output reports (Robinson, 2014). Adjustments were made based on this.

Ideally the Black-Box validation of the model, ensuring that the overall model represents the real world accurate enough for the purpose at hand (Robinson, 2014: 254), would have been done by comparing the model’s behaviour to actual historical data. Unfortunately, the actual forecasted data is only stored for one month. Therefore the validation is mostly performed through discussions with replenishment experts within Coop. One of these discussions has led to quite a large change in the model. From the discussion it became evident that the shelf capacities were not deemed most important in the replenishments. The main determinant of the quantity of products to replenish is the expected demand. If the forecast says that the store is going to sell more of a type of product than its shelf capacity, the store will receive more products than fit on the shelves. The model was adapted to incorporate this, by replenishing the stores with the forecasted amount, even if this is higher than the shelf capacity.

The actual SAP F&R system utilizes advanced forecasting techniques to include effects of the weather, holidays, special events and much more on the demand of products. Because of the limited insights into the working of these ‘demand influencing factors’ (DIFs) within SAP, a simpler forecasting technique is utilized in the simulation model. The forecasted demand is simply determined by the empirical distributions of the demand of the products. Therefore the current situation is not modelled entirely accurate, but the model is deemed accurate enough for the purpose at hand because of the exploratory research goals of this study. Experimentation validation will be discussed in section 4.8.

4.7 Experimental design

The experiments of the simulation model were divided into three sets of experiments. The first two sets were used to answer the sub-research questions. The configuration of the last set was based on the results of the first two sets and was used to answer the main research question. The following sections elaborate on these experiments.

4.7.1 Set 1: Different heights of the can-order and must-order levels

(24)

24

number of case packs per product. The size of the case packs differ per product, an overview of the case pack sizes and shelf capacities can be found in appendix 7. The higher the experiment level, the higher the can-order level was set. Which entails that level 7 represented the ‘shelf capacity – 1 case pack’, level 6 ‘shelf capacity – 2 case packs’ etc. The can-order level was lowered from level 7-1 with one case pack per level. The height of the must-order level was firstly (level 0) based on the current reorder levels, which are derived from the store image requirements (SIR). The must-order levels for the further experiments were based on a percentage of the shelf capacity, from level 1-6 it was increased with each experiment from 10%-60% of the shelf capacity (see first three rows of table 4.1).

(25)

25 4.7.2 Set 2: Increasing shelf capacities

(26)

26

times that the demand on a day was higher than the shelf capacity. This means that a number of products had to be stored in the backroom of that store. It was expected that, with increased shelf capacity, the number of products that need to be stored in the backroom would also decrease, thereby improving the performance of the can-order policy.

The base case used the current shelf capacities with the current reorder levels derived from the SIR (level 0 in set 1), which represented the must-order levels. The shelf capacity was increased with 10% at every successive experiment. The increase in shelf capacity was simulated both with and without the can-order policy applied. The can-order level is indicated by ‘not applied’ and ‘applied’ respectively (table 4.2). As can-order level, ‘shelf capacity – 1 case pack’ was chosen because with this level, most products are included in the can-order policy. The specific heights of the can- and must-order levels were not the main concern for this experiment, the goal was to show the effect of increasing the shelf capacity. Table 4.2 shows the 2x11 full factorial design of the experiment set.

4.7.3 Set 3: Practical application

(27)

27 4.7.4 Promotional demand

To most accurately model the real world situation, and simulate the effect of the promotional products on the workload, the demand data of the products were based on non-action weeks only. The demand of these products during action weeks were filtered out. The demand data were from one year, from that data, only a few weeks (0-6 weeks) of promotional demand data per product were available. This is too little data to accurately make theoretical distributions of. Furthermore, this would create an unrealistic scenario that not every week would contain promotions. As the goal was to make a realistic scenario based on these 12 products, the number of promotional products picked in the DC was based on averages from past year, and added to the daily workload without promotions.

An analysis of the number of promotional products picked in DC Gieten each day revealed the percentages in table 4.4. So on average the number of promotional products picked on Monday for instance, represent around 1.0% of the average number of products picked per week. The number of products based on these percentages have been added to the workload on each day. This created a more realistic scenario of the workload, as each week now contained promotions. Moreover, since the promotional products were added based on the average weekly workload, they were commensurate with the 12 products.

4.8 Experimental settings

(28)

28

that the model did need a short warm-up period, a warm-up period of 3 weeks (21 days) was chosen to be certain the steady state was reached. The run length should be at least ten times the length of the warm-up period, according to the rule of thumb of Banks et al. (2009). To be certain an adequate run length was chosen, it was set at 1 year (365 days). This entails that the total run length was set at 386 days.

The confidence interval method as described by Robinson (2014: 184-186) was used to determine the number of replications. The method was used for the primary output; the standard deviation of the daily workload. Figure 2 and table 1 in appendix 8 show that for the standard deviation of daily workload, a confidence interval with less than 5% deviation was already achieved at 6 or more replications. A rule of thumb by VanVoorhis and Morgan (2007) however state that you should have at least 7 observations, but preferable 30 when using ANOVA tests, as this should lead to a strong effect size of .80 (80% power). Therefore 30 has been chosen as the number of replications.

4.9 Output analyses

The output of the simulation model was fourfold; the standard deviation in workload (StDevs), the number of out-of-stocks (OOS) and number of products stored in the backroom during the simulation run (Backroom), and to be able to give an overview, the average daily workload per simulation run. This resulted in 30 StDevs, 30 OOS and 30 Backroom output per experiment because of the 30 replications. The StDevs was the primary performance measure, as the goal was to balance the workload over the days. The OOS and Backroom were the secondary performance measures.

The StDevs, OOS and Backroom were all analysed with a two-way ANOVA (analysis of variance). The dependent variables for all tests were these three performance measures. The independent variables in the first and third set were the heights of the can- and must-order levels. In the second set, the can-order applied/not applied groups and shelf capacity groups were the independent variables. As post hoc test, the Tukey test was performed.

4.9.1 ANOVA assumptions

(29)

29

(30)

30

5 Results

5.1 Results set 1: Different heights of the can-order and must-order levels

A two-way ANOVA was conducted on the influence of two independent variables (height of the can-order level, height of the must-order level) on the standard deviation of weekly workload (StDevs), on the number of out-of-stocks (OOS), and on the number of products stored in the backroom (Backroom). Table 5.1 shows the results of all three ANOVA tests. For the StDevs, both main effects and the interaction effect showed a statistically significant difference (p-value<0.001).

(31)

31

(32)

32

(33)

33

(34)

34

The F-values of all three variables suggest that the effect can be attributed most to the can-order level. However, while for OOS and Backroom the difference between level 0 and the rest is evident, the means plots show no large differences between can-order levels 1-7. This is confirmed by the Tukey post hoc tests. While for the StDevs, the differences between almost each level was significant for both the can- and must-order levels, this is not the case for OOS and Backroom. For both variables, the differences between can-order level 0 (not applied) and the rest of the can-order levels (applied) are all statistically significant (p-value<0.05), which can clearly be seen in the means plots. The differences between can-order levels 1-7 are indeed not statistically significant. While most of the must-order groups are statistically significantly different from each other, as indicated by the means plots. This shows that applying the can-order policy has a large effect on all variables. But that the height of the can-order level only has a large effect on balancing the workload, and less on OOS and Backroom.

These results show that by applying the can-order policy, the number of OOS’s are reduced. However, the number of products that have to be stored in the backroom also drastically increases. This is an unwanted effect, as this will result in more in-store inventory and more ‘double handling of goods’. The next section discusses the results of experiment set 2.

5.2 Results set 2: Increasing shelf capacities

(35)

35

(36)

(37)

37

(38)

38

(39)

39

For the StDevs and the Backroom, the Tukey post hoc test showed a statistically significant difference between each level of the shelf capacity, as suggested by the means plots. No post hoc test is performed for the can-order policy applied or not, since this consists of only 2 levels. For OOS almost all groups were statistically significant from each other as well.

To summarize, the can-order policy does seem to be able to influence the workload better with increasing shelf capacity. The OOS’s do decrease with increasing shelf capacity and by applying the order policy, but the rate of decrease is not changed much by applying the can-order policy. And lastly, the number of products in the backroom decreases with increasing shelf capacity, and the rate of decrease is also not changed much by applying the can-order policy. These results show how the height of the can-order level, the height of the must-order level, and the shelf capacity influence the performance of the can-order policy. The results however show that the policy is overcorrecting, throwing the workload off-balance towards the Monday. Therefore experiment set 3 was created to determine an appropriate policy for the case company.

5.3 Results set 3: Practical application

(40)

40

(41)

41

These results show a possible configuration of can- and must-order levels and the combination of stores and days on which to apply the policy. For the case company a can-order level of 7 and a must-order level of 20% of the shelf capacity seem to be appropriate settings. Figure 5.25 shows the average daily workload without the can-order policy applied (level 0) and with the above mentioned heights of both levels. With this configuration, better balanced workload can be achieved for the case company, it furthermore lowers the number of OOS’s. The figure shows that the workload is not perfectly balanced over the days, but it shows a large improvement over the case without balancing. The negative effect that does remain is the increase in products that need to be stored in the backroom, however the Backroom does show an improvement over the percentages achieved at experiment set 1 and 2.

(42)

42

6 Discussion

The results show that the workload in a DC for non-perishable groceries can be better balanced by applying a variant of the can-order replenishment policy. The results show that the height of the can-order level influences the performance, the higher the can-order level is set, or actually the larger the difference between the can- and must-order level, the more the workload in the DC is influenced. This is because the larger the difference, the more room the policy has to influence the workload. Furthermore, The higher the can-order level is set, the less OOS’s occur in the stores. However, this also results in more products that need to be stored in the backroom of the stores, which is also known as the backroom effect (BRE) (Eroglu et al., 2013). This is an undesirable effect because backroom stock is not on display, and therefore not productive (La Londe & Masters, 1994). Backroom stock furthermore has a larger tendency to become lost, damaged, or stolen, and it leads to the inefficient double or even triple handling of goods (La Londe & Masters, 1994). This answers the first research question, higher can-order levels ensure more workload influencing, lower number of OOS’s, but also higher number of products stored in the backroom.

The height of the must-order level also influences the performance of the can-order policy, which answers the second research question. In the case study, higher must-order levels led to better balanced workloads, to lower number of OOS’s, and to larger number of products in the backroom. The balancing of the workload is however dependent on the height of the can-order level as well, the larger the difference between the two levels, the more the policy is able to influence the workload. Increasing the shelf capacity furthermore influences the performance, which answers the third research question. With increased shelf capacities, the policy is better able to influence the workload in the DC because the can-order policy can be applied to more products, and the number of OOS’s and products in the backroom decreases. This occurs because demand which is larger than the current shelf capacity can now be fulfilled from the shelves instead of from the backroom. As Eroglu et al. (2013) indicate, the backroom effect is a function of case pack size, shelf space and reorder point. Therefore, decreasing the case pack sizes might create similar effects as increasing the shelf capacities. With smaller case packs, the DC can replenish a product sooner, without the need to store part of the case pack in the backroom. Furthermore, the can-order policy can then be applied to more products, because the difference between the shelf capacity and the case pack size of some products is at the moment marginal. Both are however not incorporated easily, increasing the shelf capacities means that the stores either need to grow in size, or that products need to be taken out of the assortment to create space, which might not be liked by customers. Case pack sizes are furthermore determined by the suppliers of the products, which are not likely to change their case packs just for one retail chain.

(43)

43

also on the stores. When the can-order policy would be applied to all stores the same, the workload might develop a tendency towards a particular day. This entails that the workload would still be unbalanced, but that the days with the highest workload would simply shift to other days. When the can-order policy is applied differently between groups of stores, the workload becomes much better balanced. This answers the main research question; ‘How can the workload

in a distribution centre for non-perishable groceries be better balanced through applying a variant of the can-order replenishment policy throughout the supply chain? While taking the downstream effects into account.’ Supply chain partners could, thanks to CPFR tools like the

(44)

44

7 Conclusion and future research

Large variations in daily workload in a DC can have a detrimental effect on the entire supply chain (Sainathuni et al., 2014). This study considered the design of a variant of the can-order replenishment policy to achieve better supply chain performance in a grocery retail supply chain, with as main goal the balancing of workload in a dry grocery DC. This is facilitated by close cooperation throughout the supply chain thanks to CPFR tools like the SAP F&R system of the case company. This study adopted the periodic review perspective of the can-order policy of Johansen and Melchiors (2003), instead of the traditional continuous review perspective. Through discrete simulation, the effects of different heights of the parameters of the can-order policy were studied. This research was of an exploratory nature, and therefore the precise figures and configurations are applicable to just a selection of all stores and products of the case company. Nonetheless, since the case used in this study represents typical retail chain operations, the insights and mechanisms may be generalizable to certain extend. The managerial implications of this study are mainly that supply chain partners could (and should), as mentioned in the discussion, apply the can-order policy through CPFR tools on certain days and only to a selection of stores. This ensures that the workload in the DC will be better balanced, while downstream of the DC, these stores will face less OOS’s, and only these stores will have to store additional products in the backroom. Thereby increasing the supply chain performance and limiting the negative effects of the can-order policy. This study has furthermore shown the importance of taking a holistic view, as improving the performance of one node in a supply chain might create undesirable effects upstream or downstream in the supply chain.

This study contributes to the literature in several ways. First, it has shown that current distribution optimization models do not take warehouse workload into consideration (e.g. the PIDRP by Bard & Nananukul, 2009; Lei et al., 2006), or do it without considering downstream consequences, for instance by setting capacity limitations on the DC (e.g. the PIRP by Gaur & Fisher, 2004; and the PVRP by Ronen & Goodhart, 2008). Secondly, it has shown the disruptive effects promotions can have on a DC, and thereby indirectly on the entire supply chain. Promotions create peaks in workload because each store need to have the promotional products in stock at the start of the promotions week, but preferably not much sooner. Thirdly, this study has shown that the can-order policy can be applied to achieve a different goal than the traditional minimization of transportation and distribution costs. It can ensure benefits throughout the supply chain by better balancing the workload in a DC and reducing the number of OOS’s in the stores. It does however produce a negative side effect by increasing the number of products in the backroom of the stores. Lastly, it has shown the interplay of the different parameters of the can-order policy, and the effects they have on the performance, in terms of workload balancing, number of OOS’s and number of products in the backroom.

(45)

45

this is just a subset of 12 products of 17 stores, while in reality there are 254 stores to which DC Gieten delivers 7600 products. The factors which makes concrete implementation recommendations difficult, is that the planograms of each store are different. Not every store has every product included in its assortment. Moreover, a product which is a fast mover in one region of the country might be a slow mover in another, or might not even be included in the assortment because there is no demand for it.

Future research should focus on how the negative effects of the increase in the number of products that need to be stored in the backroom can be mitigated. Furthermore, since applying the can-order policy through CPFR with the current goal has not been examined yet in the literature, future research should focus on the organizational side. How the policy can actually be implemented in practise and what the effects will be of this throughout the supply chain. Moreover, multiple case studies, preferably between different non-perishable goods industries, e.g. clothing retail/consumer durables industry, should be performed in future research to determine the generalizability of the findings of this research, and to explore possible enablers and barriers of implementing the policy.

(46)

46

8 References

Abdelmaguid, T.F., & Dessouky, M.M. (2006). “A genetic algorithm approach to the integrated inventory-distribution problem”. International Journal of Production Research, 44(21): 4445-4464.

Atkins, D., & Iyogun, P.O. (1988). Periodic versus ‘can-order’ policies for coordinated multi item inventory systems. Management Science, 34(1): 791-796.

Balintfy, J.L. (1964). “On a basic class of multi-item inventory problems”. Management

Science, 10(2): 287-297.

Bangsow, S. (2010). “Manufacturing simulation with plant simulation and simtalk: Usage and

programming with examples and solutions”. Munich: Springer Sciences & Business

Media.

Banks, J., Carson, J.S., Nelson, B.L., & Nicol, D.M. (2009). Discrete-event system simulation, 5th Edition Upper Saddle River, NJ: Prentice Hall.

Bard, J.F., & Nananukul, N. (2009). “The integrated production-inventory-distribution- routing problem”. Journal of Scheduling, 12(3): 257–280..

van den Berg, J. P. (2007). Integral Warehouse Management: the next generation in

transparency, collaboration and warehouse management systems. Utrecht: Management

Outlook publications.

Blattberg, R. C., Briesch, R., & Fox, E. J. (1995). “How promotions work”. Marketing

science, 14(3): G122-G132.

Branska, L., & Lostakova, H. (2011). “CPFR method application in supply chain involving continuous productions”. 20th International Conference on Metallurgy and Materials, 18(22): 5.

van Donselaar, K.H., Gaur, V., van Woensel, T., Broekmeulen, R.A.M., & Fransoo, J.C. (2010). “Ordering behavior in retail stores and implications for automated replenishment”.

Management Science, 56(5): 766-784.

van Eijs, M.J.G. (1994). “On the determination of the control parameters of the optimal can-order policy”. Mathematical Methods of Operations Research, 39(3): 289-304.

Eroglu, C., Williams, B.D., & Waller, M.A. (2013). “The backroom effect in retail operations”.

Production and Operations Management, 22(4): 915-923.

Fearne, A. (1996). “Editorial note''. Supply Chain Management, 1(1): 3-4.

Fisher, M.L. (1997). “What is the right supply chain for your product?” Harvard Business

Review, 75(2): 105-117.

Gaur, V., & Fisher, M.L. (2004). “A periodic inventory routing problem at a supermarket chain”.

Operations Research, 52(6): 813-822.

Gruen, T.W., & Corsten, D.S. (2007). A comprehensive guide to retail out-of-stock reduction in

the fast-moving consumer goods industry. Washington, DC: Grocery Manufacturers

Association, Food Marketing Institute, and National Association of Chain Drug Stores. Harmon, R.L. (1993). Reinventing the warehouse: World class distribution logistics. New York:

(47)

47

Howell, D. C. (2002). Statistical Methods for Psychology, 5th Edition Pacific Grove, CA: Duxbury.

Hübner, A.H., Kuhn, H., & Sternbeck, M.G. (2013). “Demand and supply chain planning in grocery retail: An operations planning framework”. International Journal of Retail and

Distribution Management, 41(7): 512-530.

Information Resources Management Association (IRMA). (2013). Supply chain management:

Concepts, methodologies, tools, and applications (3 Volumes). Hershey, PA: IGI Global.

Jane, C.C., & Laih, Y.W. (2005). “A clustering algorithm for item assignment in a synchronized zone order picking system.” European Journal of Operational Research, 166(2): 489 -496.

Johansen, S.G., & Melchiors, P. (2003). “Can-order policy for the periodic-review joint

replenishment problem”. The Journal of the Operational Research Society, 54(3): 283 -290.

Karlsson, C. (Ed.) (2009). Researching operations management. New York: Routledge. Khouja, M., & Goyal, S. (2008) “A review of the joint replenishment problem literature: 1989

-2005”. European Journal of Operational Research, 186(1): 1-16.

Kim, B. I., Heragu, S. S., Graves, R. J., & Onge, A. S. (2003). “A hybrid scheduling and control system architecture for warehouse management”. Robotics and Automation, IEEE

Transactions on, 19(6): 991-1001.

Köchel, P., & Nieländer, U. (2005). “Simulation-based optimisation of multi-echelon inventory systems”. International journal of production economics, 93(1): 505-513.

deKoster, R., Le-Duc, T., & Roodbergen, K. J. (2007). “Design and control of warehouse order picking: A literature review”. European Journal of Operational Research, 182(2): 481 -501.

Kumar, K. (2001). “Technology for supporting supply chain management: Introduction”.

Communications of the ACM, 44(6): 58-61

La Londe, B. J., & Masters, J. M. (1994). ‘Emerging logistics strategies: blueprints for the next century’. International journal of physical distribution & logistics management, 24(7): 35-47.

Lei, L., Liu, S., Ruszczynski, A., & Park, S. (2006). “On the integrated production, inventory and distribution routing problem”. IIE Transaction, 38(11): 955–970.

Melchiors, P. (2002). “Calculating can-order policies for the joint replenishment problem by the compensation approach”. European Journal of Operational Research, 141(3): 587 -595.

Robinson, S. (2014). Simulation the practice of model development and use. London: Palgrave Macmillan.

Ronen, D., & Goodhart, C.A. (2008). “Tactical store delivery planning”. The Journal of the

Operational Research Society, 59(8): 1047-1054.

Rouwenhorst, B., Reuter, B., Stockrahm, V., van Houtum, G.J., Mantel, R.J., & Zijm, W.H.M. (2000). “Warehouse design and control: Framework and literature review”. European

(48)

48

Sachs, K., Kounev, S., Bacon, J., Buchmann, A. (2009). “Performance evaluation of message -oriented middleware using the SPECjms2007 benchmark”. Performance Evaluation, 66(1): 410-434.

Saghir, M., & Jönson, G. (2001). “Packaging handling evaluation methods in the grocery retail industry”. Packaging Technology and Science, 14(1): 21-29.

Sainathuni, B., Parikh, P.J., Zhang, X., & Kong, N. (2014). “The

warehouse-inventory-transportation problem for supply chains”. European Journal of Operational Research, 237(2): 690-700.

Shannon, R.E. (1975). Systems simulation: The art and science. Englewood Cliffs, NJ: Prentice Hall.

Skjoett-Larsen, T., Thernøe, C., & Andresen, C. (2003). “Supply chain collaboration: Theoretical perspectives and empirical evidence”. International Journal of Physical Distribution &

Logistics Management, 33(6): 531-549.

van der Sluis, E. (1993). “Reducingsystem nervousness in multi-product inventory systems”.

International Journal of Production Economics, 30(1): 551-562.

VanVoorhis, C.R.W., & Morgan, B.L. (2007). “Understanding power and rules of thumb for determining sample sizes”. Tutorials in Quantitative Methods for Psychology, 32(2): 43-50.

Strack, G., Fortz, B., Riane, F., & van Vyve, M. (2011). “Comparison of heuristic procedures for an integrated model for production and distribution planning in an environment of shared resources”. Université catholique de Louvain, Center of Operations Research and

Econometrics (CORE).

Terzi, S., & Cavalieri, S. (2004). “Simulation in the supply chain context: a survey”. Computers

in Industry, 53(1): 3-16.

Tompkins, J.A., White, J.A., Bozer, Y.A., & Tonchoco, M.A. (2003). Facilities Planning. New York : John Wiley & Sons.

Trauzettel, V. (2014). “Optimal stocking of retail outlets: The case of weekly demand pattern. Business Logistics in Modern Management, 14(1): 3-11.

Voss, C. (2009). In Karlsson, C. (Ed.) (2009). Researching operations management. New York: Routledge.

Waller, M. A., Heintz Tangari, A., & Williams, B. D. (2008). “Case pack quantity's effect on retail market share: An examination of the backroom logistics effect and the store-level fill rate effect”. International Journal of Physical Distribution & Logistics

Management, 38(6): 436-451.

(49)

49

Appendix 1: Popular routing models

Significant attention has been paid to try to determine optimal distribution and replenishment strategies in a supply chain. The main objective is to balance inventory and transportation costs (Sainathuni et al., 2014). Many different approaches which take the perspective of the supplier are proposed, like the production-inventory-distribution-routing problem (PIDRP) (e.g.; Bard & Nananukul, 2009; Lei et al., 2006), which combines decisions of production and distribution lot sizing level with vehicle routing decisions (Strack et al., 2011). Good results based on reduced inventory and transportation costs can be achieved through these PIDRP models, however, they assume deterministic demand. In environments with dynamic demands, these models will not work. Another approach is the integrated inventory-distribution problem (IIDP) introduced by Abdelmaguid and Dessouky (2006), which considers inventory and distribution costs, while allowing for backlogging. Good results based on reduced inventory and transportation costs can be achieved through these IIDP models as well, however, since backlogging is allowed, it might result in a less desired performance later on in the supply chain (retailers). Furthermore, the above mentioned models do not take warehouse workload into consideration. Which is supported by observations by Sainathuni et al. (2014) that warehousing decisions are not included when developing a distribution plan for the supply chain.

One model that does take warehouse workload into consideration is the warehouse-inventory-transportation problem (WITP) introduced by Sainathuni et al. (2014). In the WITP, the objective is to determine an optimal distribution plan to minimize total distribution costs, by also including worker congestion in the warehouse. Considerable savings in the total distribution cost can be achieved, together with significant reduction in workload variance at the warehouse. The model is however sensitive to other warehousing decisions like aisle configuration, technology at the warehouse and allowable level and productivity rate of temporary workers.

The most well-known model is however the periodic (vehicle) routing problem (PRP/PVRP). In the basic PVRP, the goal is to determine deliveries (or pickups) to customer locations in a way that each location is delivered to (or picked up at) at the required frequency during the planning horizon (Ronen & Goodhart, 2008). Many variations and additions have been proposed and analysed over the years, like heterogeneity of truck fleet, static versus dynamic demand, and many more. Literature on warehouse workload balancing (or reduction of variation) is scarce, mostly the literature is focussed on balancing the workload in a warehouse between zones of the warehouse (e.g.; Jane & Laih, 2005) or between gantry robots in an automated warehouse (e.g.; Kim et al., 2003). A few routing studies however do incorporate the workload at the DC in the model. The only two articles found in the literature that do incorporate these workload constraints, are the one by Gaur and Fisher (2004), and the one by Ronen and Goodhart (2008).