Online Customers in Control: Reducing Warehouse Costs Through Differentiated Lead Times

(1)

Online Customers in Control: Reducing Warehouse Costs

Through Differentiated Lead Times

Master Thesis, MSc Supply Chain Management

University of Groningen, Faculty of Operations

(2)

2

ABSTRACT

The objective of is this thesis is to determine what labor cost reduction can be obtained in an e-fulfillment warehouse through the introduction of differentiated lead times. First, a literature review is done to determine how labor costs should be estimated and which factors may influence these measures when lead times are differentiated. The effect of two of these factors on labor costs is determined by quantitative modeling in which different demand patterns are analyzed. With this model, the costs with a fixed lead time are calculated and then compared to the costs with differentiated lead times to determined the theoretical cost reduction. The factors taken into account are reduced demand uncertainty which reduces overtime and the application of workload control which enhances productivity. The factors respectively showed a theoretical cost reduction of at most 31% and 9%.

(3)

3

MANAGEMENT SUMMARY

E-fulfillment magazijnen hebben te maken met sterke schommelingen in werklast en korte doorlooptijden. Het aantal online bestellingen fluctueert sterk gedurende de week, waardoor e-fulfillment magazijnen gemiddeld zo’n 30% van de wekelijkse orders op maandag verwerken. Ook gedurende de dag fluctueert de werklast sterk, omdat veel consumenten ’s avonds een bestelling plaatsen. Cut-off tijden worden steeds later en na de cut-off tijd is er een sterke stijging in werklast te zien. In 2013 garandeerde 50% van de e-tailers in Nederland 24-uurslevering, een belangrijke reden voor de druk die e-fulfillment magazijnen op maandag en aan het eind van de dag hebben om orders tijdig te kunnen verwerken. Echter hebben consumenten juist steeds meer behoefte hebben aan flexibiliteit en bepalen liever zelf wanneer zij hun bestelling thuis ontvangen. Veel van de pakketten wordt namelijk niet bij de eerste bezorgpoging afgeleverd omdat er niemand thuis is.

Het doel van dit onderzoek was om te bepalen welke kostenbesparing op arbeid behaald kan worden in een e-fulfillment magazijn door de klant de afleverdag te laten bepalen. Eerst werden experts op het gebied van magazijnen geconsulteerd wat het effect van gedifferentieerde doorlooptijden op kosten in een magazijn zou kunnen zijn. Naast arbeidskosten, zouden ook kosten van materieel en ruimte verlaagd kunnen worden door het verlagen van pieken. De verwachting is dat ook op voorraadkosten besparingen mogelijk zijn. Echter is gekozen om alleen arbeidskosten te onderzoeken, gezien het relatief grote deel dat arbeidskosten innemen in een magazijn. Met literatuuronderzoek werd bepaald welke factoren arbeidskosten zouden kunnen beïnvloeden bij gedifferentieerde doorlooptijden. De gemiddeld langere doorlooptijd zou kunnen zorgen voor meer zekerheid over toekomstige werklast. Daarnaast is de verwachting dat een hogere productiviteit en vermindering van leegloop kan worden behaald door het afvlakken van pieken.

Vervolgens werd een model ontwikkeld om arbeidskosten te berekenen. Hierin werden twee factoren meegenomen, namelijk het aanpassen van het moment van orderverwerking waardoor productiviteit verbetert en de hogere zekerheid over toekomstige werklast waardoor minder overuren gemaakt worden. Het model werd toegepast met data van een retail magazijn waar weborders en traditionele orders verwerkt worden. Eerst werden arbeidskosten bepaald bij een vraagklasse met doorlooptijd van een dag waarbij alle orders direct worden verwerkt. Vervolgens werd een tweede vraagklasse met doorlooptijd van twee dagen toegevoegd. Kosten werden berekend voor verschillende scenario’s waarin de verhouding van de twee vraagklassen varieerden.

(4)

(5)

5

1. INTRODUCTION

Warehouses in an e-commerce environment have a growing need for efficiency (Agatz, Fleischmann, & van Nunen, 2008). Transaction sizes are typically very small (Tarn, Razi, Wen, & Perez Jr, 2003) and this makes order picking labor intensive. Additionally, orders must be processed in a relatively short timeframe and there is much demand uncertainty (Tarn et al., 2003). About half of Dutch e-tailers promises next day delivery and nearly one-third uses a cut-off time after 9.00 PM as the latest time an order can be placed to be handled the same day (Global Webshop Logistics 2013). Demand fluctuations directly affect workloads and capacity utilization in e-fulfillment warehouses and pose a challenge on capacity management: “In addition to annual demand patterns, demand differences during the day

(morning–evening) and during the week (mid-week–week-end) are particularly important in e-fulfillment. Staffing levels need to be adjusted to these demand fluctuations.” (Agatz et al.,

2008: 350). According to the paper Sectormanagement Transport & Logistiek 2011, published by ING, online customers often find selecting delivery day more important than next day delivery. In 2013, about 15% of Dutch e-tailers let their customers choose a delivery day, according to the Global Webshop Logistics 2013 Report. When offered this possibility, about 86% of customers opt for delivery in three days instead of the next day (PostNL, 2013). A result of letting the customer choose the delivery day is that lead times are differentiated because the amount of days between order arrival and delivery due date varies. In comparison to next day delivery, orders do not need to be picked and packed at the moment of arrival so a peak in demand in the evening does not lead to a workload peak of the same level. Studies have confirmed that there is a relationship between workload level and warehouse performance (Gue, Meller, & Skufca, 2006; Heath, Ciarallo, & Hill, 2012; Parikh & Meller, 2009). A workload peak is expected to result in a lower performance, as is a relatively low workload level. Bertrand & Van Ooijen (2002) showed that when the relationship between workload and productivity is taken into account, workload control can improve performance. Workload control is about decoupling order entry from order release by regulating the release of orders onto the shop-floor according to workload norms (Thürer, Stevenson, & Silva, 2011). In addition, differentiated lead times reduce demand uncertainty because some demand information is available at an earlier moment. Studies have shown that reduced demand uncertainty can lower labor costs by the use of flexible workers and reduced overtime (Pinker & Larson, 2003; Buyukkaramikli, Bertrand & van Ooijen, 2011).

(7)

7 differentiated lead times. Figure 1 shows the research model representing the objective and scope of this research.

FIGURE 1. Research Model

The model developed in this thesis shows the effect of differentiated lead times on costs in an e-fulfillment warehouse through both directions shown in the research model. Managers can use the model to decide which delivery day possibilities to offer and to make order release decisions.

This thesis is an addition to literature on differentiated lead times because it takes an operations perspective and focuses on labor costs. The topic of lead time differentiation in e-commerce until now was only studied in relation to yield management and delivery costs (Agatz, 2009) and inventory management (Chen, 2001; Huang, Liao & Robb, 2011; Kocaga & Sen, 2007; Sarkar & Shewchuk, 2013; Tempelmeier, 2006; Wang, Cohen, & Zheng, 2002). Additionally, the model developed in this thesis complements workload control and order release literature by theoretically applying the workload control concept to an e-fulfillment warehouse. Workload control and order release have mainly been studied in a production (Breithaupt, Land, & Nyhuis, 2002; Thürer et al., 2011) or supply chain (Chan, Humphreys & Lu, 2001) context. Furthermore, the model answers to Thürer et al.’s (2011) call for analytical research into workload control developing heuristics and models to help managers make faster decisions.

Next, the research questions are given, after which the methodology used in this research is explained. Part two provides the theoretical background on the topic of this study. In part three, the model is explained and in part four the results from applying the model are given. Part five contains a discussion of the results compared to what is known in literature and discussed in part two. In part six the research is summarized and conclusions, recommendations and possibilities for further research are provided.

1.1 Problem Definition and Research Questions The research question addressed in this research is as follows:

What labor cost reduction can be achieved in an e-fulfillment warehouse through differentiated lead times?

To answer the research question, the following sub questions will be answered:

(8)

8 2. What are the labor costs when lead times are fixed and orders are processed

immediately on the day of arrival?

3. What are the labor costs when lead times are differentiated and the workload control concept is applied?

4. What is the theoretical labor cost reduction from differentiated lead times compared to fixed lead times, when demand is known?

1.2 Methodology

First, a literature review will be done to determine how operational warehouse costs should be measured and how these measures may be influenced by differentiated lead times. Specific attention will be paid to workload control and the relationship between workload and productivity. Additionally, experts in the field of warehousing will be asked for their input on which costs and which factors are most important. An answer to question one is the result of this part of the thesis.

The main methodology used in this research is quantitative modeling, which main aim is to “build objective models that explain the behavior of operational processes” (Karlsson, 2009). A key aspect of a quantitative modeling approach is that variables are causally related and the future state of the model can be predicted (Karlsson, 2009). This is important here, as the cost reduction that can be obtained through differentiated lead times needs to be calculated.

Figure 2 shows the steps taken to answer research questions two, three and four. Soepenberg, Land & Gaalman (2012) followed the same steps in their research on the actual implementation of workload control to control logistical performance in a single e-fulfillment warehouse. Contrary to their study, in this thesis the application of workload control is purely theoretical, but the steps are appropriate to compare the before and after situation here as well. In the first step the costs related to the current situation with fixed lead times are measured. The current situation contains that all orders need to be delivered the next day and are released immediately after arrival. The second step involves the implementation of the workload based order release mechanism. Multiple differentiated lead time scenarios with varying distributions of demand lead times are analyzed. In the third and final step, the results of the workload control mechanism are compared to the results of the baseline measurement in the first step to answer research question four.

FIGURE 2 Research steps (Soepenberg, Land & Gaalman, 2012)

(9)

9 extended workload control job shop models by taking the relationship between workload and productivity into account. Although the context of this thesis is an e-fulfillment warehouse and not a job shop, both contexts have similar key management issues which is why the application of the model in an e-fulfillment warehouse is deemed appropriate. For “job shop

like production situations”, achieving “short lead times, reliable due dates” and “a reasonable level of resource utilization” are key management issues (Bertrand & Van Ooijen,

2002: 665) which can be solved by controlled order release to the shop floor. As described above, e-fulfillment warehouses need to deal with short lead times. On time delivery is very important in e-commerce and there is a need for efficiency in labor intensive e-fulfillment warehouses (Agatz et al., 2008).

The next part describes the context of warehouses, e-fulfillment warehouses, workload control, differentiated lead times and warehouse costs and the studies of Pinker & Larson (2003) and Bertrand & Van Ooijen (2002) are explained further.

2. THEORETICAL BACKGROUND

2.1 Warehouses

Warehouses are an essential part of a supply chain (Gu, Goetschalckx, & McGinnis, 2007) because of their function to enable material flow buffering, assembly of materials from multiple suppliers and/or value-added-processing activities. There is a substantial body of literature addressing the different aspects of warehouses such as design, operations and specific activities such as order picking. An overview of research addressing warehouses can be found in one of the literature reviews (De Koster, Le-Duc & Roodbergen, 2007; Gu et al., 2007; Gu, Goetschalckx, & McGinnis, 2010; Rouwenhorst, Reuter, Stockrahm, van Houtum, Mantel, & Zijm, 2000; Van den Berg & Zijm, 1999). This research focuses on distribution warehouses in an e-fulfillment context, where products are stored and orders of external customers are fulfilled (Rouwenhorst et al., 2000).

2.2 E-fulfillment warehouses

According to Agatz et al. (2008) e-fulfillment warehouses have an increasing need for efficiency due to demand fluctuations and the long tail of e-commerce. Because of the long tail, the number of different products stored is very large. Additionally, orders in the B2C e-commerce environment involve small quantities (Tarn et al., 2003). The combination of a large assortment of products and small single orders create a very costly order picking process in distribution warehouses (Rouwenhorst et al., 2000). Especially in e-fulfillment warehouses, where relatively labor intensive split-case or piece-picking methods are often applied, order picking is costly (Agatz et al., 2008).

(10)

10 are often impulse purchases increasing demand uncertainty, according to Buck Consultants International, 2006. Quick delivery is an important part of fulfillment quality in e-commerce (Chen & Chang, 2003; Xing, Grant, McKinnon, & Fernie, 2010), so many e-tailers offer fast delivery and have to deal with short lead times. Additionally, cut-off times are extended to a later hour to attract customers (Global Webshop Logistics, 2013), creating the need in e-fulfillment warehouses to operate under even shorter lead times.

2.3 Workload control

The workload control concept was developed in the 1980s for a job shop environment (Breithaupt et al., 2002). Bertrand & Van Ooijen (2002) stated that workload control is necessary for systems with high capacity utilization, varying order arrival patterns and where the workload influences processing times. Henrich, Land, & Gaalman (2004) found that workload control becomes more applicable when variability increases, for example when there are differences in due date. The core aspect of workload control is decoupling order entry from order release by regulating the release of orders onto the shop-floor according to workload norms (Thürer et al., 2011). By regulating order release, workloads can be adjusted to be near the ideal workload allowing for a higher utilization of the work station (Bertrand & Van Ooijen, 2002). Figure 3 shows where order release fits within the concept of workload control as explained by Henrich, Land, & Gaalman (2004).

FIGURE 3 Order release within the workload control concept (Henrich, Land & Gaalman, 2004), at a warehouse

After arrival, orders wait in a pre-shop pool for release. The release depends on the workload of capacity groups compared with workload norms and the urgency of the order (Henrich, Land, & Gaalman, 2004; Fernandes & Carmo-Silva, 2011). Thürer et al. (2009) found that rush orders should be prioritized at the release stage. After release, orders enter the shop floor, meaning that orders are added to pick lists at an e-fulfillment warehouse.

2.4 Warehouse costs

(11)

11 customer classes on the degree of advance demand information provided, used early fulfillment to improve performance and showed that their policy reduced inventory required. Inventory costs are not considered in this research, although it should be acknowledged that substantial cost reductions may be obtained here as well as is shown by the authors mentioned above. In this thesis, only labor costs are considered. As stated above, labor intensity in e-fulfillment warehouses is high. Especially order picking is labor intensive and order picking costs are estimated to determine more than half of total warehouse costs (De Koster et al., 2007). In the next section, a description of differentiated lead times is given and the way labor costs may be affected by differentiated lead times is discussed.

2.5 Differentiated lead times and labor costs

Wang, Cohen, & Zheng’s (2002) study of a distribution system with differentiated lead times serves as an example of what is considered differentiated lead times in this research. The study looked at a system with orders in two service classes, one class with an ‘emergency’ lead time that is fulfilled immediately and the other class with a ‘non-emergency’ lead time that is fulfilled a certain number of days later. Demand lead time is a term often used to describe the time between order arrival and due date when “customers do not require

immediate delivery of orders, but allow for a fixed delay” (Hariharan & Zipkin, 1995).

Batching. Order batching is an operational decision problem determining which orders are released and picked together and can improve performance and reduce costs in an order picking system (Hsieh & Huang, 2011). Henn (2012) showed that maximal completion time of online orders can be reduced by order batching making it possible to decrease working hours and overtime. Bukchin, Khmelnitsky & Yakuel (2012) developed a decision making policy for a dynamic order batching problem to minimize tardiness and overtime costs. The policy determines when an order picker should go on an order picking tour or when the picker should wait until more orders arrive to add to the tour. The results of the study showed that costs are reduced when demand lead times are longer, which is what happens when lead times are differentiated.

(12)

12 reduction is achieved because more contingent workers are contracted and less work is done during overtime when more demand information is available. The authors stressed that only increasing labor flexibility does not improve performance unless more demand information is available. Buyukkaramikli, Bertrand & van Ooijen (2011) also found that the use of workload information reduces costs. They studied the increase and decrease of staffing levels in a make to order system with short and reliable (fixed) lead times where capacity flexibility plays an important role.

Influence workload size. Studies have shown that workload level affects system performance in warehouses. High workload levels may lead to congestion which reduces order picking speed. The main cause for this speed reduction is that order pickers are blocked by other pickers and their carts in the same area (Heath, Ciarallo, & Hill, 2012). Blocking can result in picker idle time and lower productivity and increases with the amount of pickers (Parikh & Meller, 2009). However, a high workload does not necessarily lead to more congestion. Gue, Meller, & Skufca (2006) discovered that when orders are picked in a very busy area with a narrow aisle and high pick density, there tends to be less congestion among workers. The reason is that more time is spent picking and less time is spent travelling so less blocking can occur.

Bertrand & Van Ooijen (2002) took the relationship between workload and productivity into account in their simulation study into the effect of workload control on total throughput time. The results show that workload control affects performance and that a production system can become unstable without workload control, when the relationship between workload and productivity is considered. The system was studied under immediate release and workload control. For workload control, Bertrand & Van Ooijen (2002) used a symmetrical release policy in which orders are held back when there is a peak in demand and orders are pulled forward and released earlier when demand is relatively low. In their study, Bertrand & Van Ooijen (2002) assumed an ideal workload exists under which individual productivity is highest. The effective processing time is based on a linear function in which the difference between actual and ideal workload determines a variable part of the processing time.

As can be taken from the previous paragraphs, lead time differentiation can decrease labor costs in several ways. Taking all of these relationships into account in one study is too much. The relationship between workload and productivity and the fraction of contingent workers and overtime will be used to determine the labor cost reduction in the analytical model, which is explained in the next part.

3. MODEL

(13)

13 relationship between workload and productivity and the application of workload control are explained.

3.1 Notation

In Table 1 the notation of all parameters, variables and functions of the model is given. Parameters

Crw cost of a regular worker per hour

λi allocation of demand to class (i=1, 2)

Demand uncertainty

Ccw cost of a contingent worker per hour

Cot cost of overwork per hour

Workload control

α sensitivity of productivity to workload level

IW ideal workload

m minimal processing time

Variables

Dt demand on day t

R1,t number of class 1 orders released on day t

Wt workload in number of orders on day t

PTt effective processing time per order in hours on day t

tpt t total throughput time in hours on day t

DC daily costs

TC total costs

RW regular workers in hours

Demand uncertainty

CW contingent workers in hours

OT overtime in hours Decision variables

Workload control

R2,t fraction of class 2 orders released on day t 0, 0.2, 0.4, 0.6, 0.8, 1

Q2,t fraction of class 2 orders not released on day t 1 – R2,t

TABLE 1. Model notation

3.2 Differentiated lead times

(14)

14 of one day and class 2, λ2, has a lead time of two days. The allocation of demand to classes varies in the so called ‘demand class scenarios’ and is as follows:

Scenario 0: λ1 = Dt λ2 = 0 Scenario 1: λ1 = .7 * Dt λ2 = .3 * Dt Scenario 2: λ1 = .5 * Dt λ2 = .5 * Dt Scenario 3: λ1 = .3 * Dt λ2 = .7 * Dt

To determine demand level Dt, a dataset of Districon of daily demand levels in an e-fulfillment retail warehouse is used. This way, the cost reduction possible through differentiated lead times can be determined by comparing the actual labor costs of scenario 0 to costs in the other scenarios.

3.3 Demand uncertainty

Pinker & Larson (2003) assume the daily amount of work is a Poisson random variable which is divided in two batches with a different demand uncertainty. When there is more work in the batch with lower demand uncertainty (Batch 1), the manager has more information available when notifying the employment agency. A longer lead time means more certainty of future work, so Batch 1 is considered as the equivalent of Class 2 of Wang, Cohen & Zheng (2003) with a lead time of two days. Pinker & Larson (2003) also considered flexibility of workers, but this is not taken into account here, as it does not help to solve the objective of this research. Pinker & Larson (2003) analyzed five scenarios similar to Wang, Cohen & Zheng’s (2003) scenario’s described above:

(Batch 1, Batch 2) (20, 0) (14, 6) similar to scenario 3 (10, 10) similar to scenario 2 (6, 14) similar to scenario 1 (0, 20) similar to scenario 0

The last scenario with only Batch 2 demand is the current situation when there is a fixed lead time of one day. The first scenario of Pinker & Larson with only Batch 1 demand will not be used here, since the goal is to study the cost reduction possible through differentiated lead times and this scenario has no differentiated lead time.

Pinker & Larson (2003) assumed the following cost parameters, which are used as factors on the actual hourly wage of regular workers in the warehouse.

Crw = 1, Ccw = 1.2, Cot = 2

(15)

15 guarantee levels were calculated and used as input into the model. Figure 4 shows that the fraction of the workforce that is contingent increases with the availability of demand information. So when there is less demand uncertainty, more flexible capacity is reserved (Pinker & Larson, 2003). This is interesting, as the cost of contingent workers is larger than of regular workers, but can be explained by Figure 5. When demand uncertainty decreases, the number of overtime shifts also decreases and costs of overwork are larger than contingent worker costs.

FIGURE 4. Contingent fraction of workforce for different information scenarios and flexibility levels (Pinker & Larson, 2003)

FIGURE 5. Number of overtime shifts used for different information scenarios and flexibility levels (Pinker & Larson, 2003)

Pinker & Larson (2003) only reported the number of overtime shifts and not the fraction of work done during overtime, which is needed for the model. So the number of total shifts was determined by using the fraction of contingent workforce and the number of contingent shifts that was given in the results, as is shown in Table 2.

Scenario Average fraction of workforce that is contingent Number of contingent worker shifts Number of total shifts 0 0.21 12 57.31 1 0.22 33 148.73 2 0.29 59 202.58 3 0.35 80 227.76

(16)

16 Then, the number overtime shifts was divided by the number of total shifts to get the fraction of overtime shifts, shown in Table 3.

Scenario Number of overtime shifts Fraction of overtime shifts

0 30 0.52

1 30 0.20

2 27 0.13

3 24.6 0.11

TABLE 3. Fraction of work done during overtime per scenario

To determine costs considering Crw, Ccw and Cot, tpt,t is first divided in two parts, planned working hours and overtime. The division is based on the fraction of overtime shifts. Then, the planned working hours are divided in contingent worker and regular worker hours, based on the fraction of contingent workforce.

3.4 Workload and productivity

In the final part of the theoretical background the relationship between workload level and productivity was described. Different from Bertrand & Van Ooijen (2002), this relationship for the total system’s productivity is addressed and not individual productivity. The same dataset as described above is used to determine the relationship between workload and productivity at this warehouse. The trend line is used to determine the processing time per order as a function of the daily workload in orders, based on the following function of Bertrand & Van Ooijen (2002):

Pt = (1 + α ( | W – IW | / IW ) ) * m

The function will be changed to represent the actual relationship between workload level and productivity from the dataset in the pick and pack process more accurately. The new functions are described in the Results part.

3.5 Workload control

Similar to Wang et al. (2002), demand Dt is divided in three groups according to the way workload level Wt is affected. The groups are defined below:

R1,t = number of class 1 orders that are released on day t R2,t = number of class 2 orders that are released on day t

Q2,t = number of class 2 orders that are held back and released on day t+1 Workload at day t, Wt, is thus defined as follows:

Wt = R1,t + R2,t

(17)

17 determined once and cannot be changed anymore. This is the best representation of a real situation, since managers cannot change workloads on a day looking back. The results obtained from the model are described in the next part.

4. RESULTS

4.1 Description of warehouse

The dataset contained three months of daily workload data from a warehouse of a multi-channel retailer. Figure 6 shows the process at the warehouse of which data was used. Online orders that arrive before 6.00 P.M. are released into the order picking process the same day. Traditional store orders are picked per store and online orders are picked in batches as a single store. After picking, online orders are sorted and packed in a separate e-commerce area. At the end of the day, the package delivery company picks up the online orders and delivers them to the customer the following day.

FIGURE 6 Process of an online order at multi-channel warehouse

The data shows that online demand and workload fluctuate heavily over the different weeks, but the distribution of workload over days follows a similar pattern. On Mondays, workload is highest after which it drops substantially on Tuesday and increases slightly on the following days. Figure 7 shows the average distribution of workload over days at the warehouse of which data is used. The standard deviation of the daily workload at the warehouse under study is less than a third of the average each day, showing only a small variation in the data.

(18)

18 4.2 Demand uncertainty and overtime

Table 4 and Table 5 below show the costs of overtime, contingent workers and regular workers in the picking and packing process respectively as percentage of total costs. Also, per scenario the cost reduction compared to Scenario 0 is given. When the fraction of class 2 demand increases, less work is done during overtime and more contingent workers are hired. Overtime costs make up a smaller part of total costs when more demand is assigned to class 2. The largest amount of total costs is still made by regular workers, which is a more constant percentage than the other cost factors. The cost reduction that is a result of having more demand information available varies from 25% in the scenario with most class 1 demand to 31% with most class 2 demand.

Scenario Overtime costs as % of total costs Contingent worker costs as % of total costs Regular worker costs % of total costs Total labor costs Cost reduction compared to Scenario 0 0 72% 7% 21% € 887.311 1 37% 16% 47% € 667.047 25% 2 26% 24% 50% € 626.188 29% 3 21% 31% 48% € 614.437 31%

TABLE 4. Costs per scenario in picking process

Scenario Overtime costs as % of total costs Contingent worker costs as % of total costs Regular worker costs % of total costs Total labor costs Cost reduction compared to Scenario 0 0 72% 7% 21% € 444.449 1 37% 16% 47% € 334.120 25% 2 26% 24% 50% € 313.654 29% 3 21% 31% 48% € 307.768 31%

TABLE 5. Costs per scenario in packing process

The next part contains the cost analysis made to determine the effects of workload control when productivity is related to workload.

4.3 Workload control and productivity

The nature of the relationship between workload and productivity differs between the picking and packing process. For both processes, several linear functions were created with different values of α for different workload levels to represent the functions from the real data.

(19)

19 FIGURE 8. Relationship between workload and productivity in picking process

The α values were calculated by filling in the Pt function explained in part three with points from the trend line. Table 6 shows the α values used to determine productivity at different workload levels in the picking process.

Lower workload limit Upper workload limit α

< 250 -10,5195 250 350 -9,3103 350 450 -8,0731 450 550 -6,8067 550 650 -5,8163 650 750 -5,1119 750 850 -4,7038 850 950 -4,2857 950 1050 -3,9536 1050 1300 -3,3956 1300 1500 -3,0451 1500 1700 -2,7452 1700 1900 -2,5000 1900 2100 -2,3143 2100 2750 -1,9385 2750 3250 -1,5857 3250 3750 -1,4026 3750 4250 -1,2343 4250 4750 -1,1429 4750 5250 -1,0286 5250 6500 -0,8571 6500 7500 -0,6429 7500 8500 -0,5143 > 9500 0,0001

TABLE 6. α value per workload level in picking process

(20)

20 FIGURE 9. Workload distribution picking process after workload control in scenario 0

FIGURE 10. Workload distribution picking process after workload control in scenario 1

In each scenario, the variation in workload over days increases. In the picking process, a low workload leads to a large productivity loss. Workloads peaks lead to the highest productivity, and the lowest costs, as can be seen in Table 7.

Scenario Costs

Cost reduction compared to Scenario 0

0 € 190.399

1 € 185.750 2%

2 € 183.703 4%

3 € 180.100 5%

TABLE 7. Cost reduction from workload control and productivity in picking process

(21)

21 nature than in the picking process. Processing time per order is lowest at a lower workload level and increases to a maximum around 5000 orders per day after which it decreases again. The trend line is not linear and becomes steeper at smaller and larger workload levels.

FIGURE 13. Relationship between workload and productivity in packing process Table 8 shows the values of α in the packing process.

Lower workload limit Upper workload limit α

< 950 0,5015 950 1450 0,4415 1450 1950 0,3925 1950 2450 0,3270 2450 2950 0,2747 2950 3450 0,2208 3450 3950 0,1145 3950 4450 0,0981 4450 4950 0,0491 4950 5450 0,0000 5450 5950 0,0491 5950 6450 0,0981 6450 6950 0,1145 6950 7450 0,2208 7450 7950 0,2747 7950 8450 0,3270 8450 8950 0,3925 8950 9450 0,4415 > 9450 0,5015

TABLE 8. α value per workload level in packing process

(22)

22 Figure 14. Workload distribution packing process after workload control in scenario 0

Figure 15. Workload distribution packing process after workload control in scenario 1

In the packing process, a larger cost reduction was obtained than in the picking process. Table 9 shows the costs and cost reductions found in each scenario in the packing process.

Scenario Costs

Cost reduction compared to Scenario 0

0 € 310.778

1 € 292.634 6%

2 € 291.446 6%

3 € 283.142 9%

TABLE 9. Cost reduction from workload control and productivity in packing process

(23)

23

5. DISCUSSION

In the previous part, the cost reductions that can be obtained through differentiated lead times that resulted from the quantitative model were given. In this part, the results are discussed and compared to the literature from part two. First, the cost reduction through overtime and contingent workers, based on Pinker & Larson (2003) is addressed, followed by the cost reduction from workload control. Finally, the generalizability of results is discussed.

The results from the previous part show that when the introduction of differentiated lead times reduces overtime and increases number of contingent workers, labor costs can be reduced with 31%. An important thing to take into account with this cost reduction is that it is completely based on the results of Pinker & Larson (2003). The applicability of their study was not tested in an e-fulfillment warehouse. However, Pinker & Larson’s (2003) study can at most points be applied to an e-fulfillment warehouse given the workplace they considered. Their model takes into account labor demand uncertainty and staff absenteeism and processed work inventory is not allowed, which is in line with e-fulfillment warehouses. The possibility of backlogging work is the only aspect that may reduce the applicability, because in e-commerce on time delivery is very important (Agatz et al., 2008). Therefore, a larger backlog penalty than used by Pinker & Larson (2003) may be appropriate in e-fulfillment warehouses. The cost reduction resulting from enhanced productivity through workload control is lower than from reduced demand uncertainty. When workload control was applied and productivity was related to workload, a maximum cost reduction of 9% was found in the packing process. Apparently, the gain in productivity seems to be smaller than the gain from reduced overtime. The cost reductions found in the analyses of both factors do follow a similar pattern along the demand scenarios. Pinker & Larson (2003) found that when demand uncertainty decreases, costs decrease as well, because less work is done during overtime. Their finding was confirmed in the analysis of workload control and workload related productivity. In both processes, the cost reduction was largest in the scenario where the fraction of demand with the longest lead time was largest.

Online orders make up only a small part of total workload at the analyzed warehouse. The cost reduction presented comes from only labor costs related to online orders. The cost reduction on the total labor costs would therefore be a lot smaller. However, with the expected further growth of online sales, labor costs will be determined more by online demand so in the future the effect described here will become a lot more realistic.

(24)

24 was only applicable when a second demand class was introduced. Before, orders needed to be processed on their arrival date so it was not possible to move orders to the next day. However, since there was only one extra demand class introduced, more evidence is needed to confirm that the applicability of workload control increases with the growth of due date differences. The relationship between workload and productivity was different in the picking and packing process. In the picking process, a higher workload level leads to higher productivity, which is the opposite of Heath et al.’s (2012) finding. Apparently, in this warehouse high workload does not lead to congestion, or congestion does not reduce order picking speed. It may be that Gue et al.’s (2006) argument can be applied in this warehouse, that less blocking occurs because more time is spent picking than travelling when workloads are higher. In the packing process, at medium workload levels, productivity was lowest. This contradicts the assumption of Bertrand & Van Ooijen (2002) of an ideal workload under which productivity is highest. An explanation for this difference may be that Bertrand & Van Ooijen (2002) assumed a relationship between individual productivity and workload, while here productivity and workload of the total system were analyzed. There is no evidence of the relationship between individual productivity and workload in this warehouse. It would be interesting to look into the individual relationship, because if Bertrand & Van Ooijen’s (2002) assumption is correct, the relationship would disappear because of the system’s productivity function. Because there are different productivity curves in the picking and packing process, results show that both processes have a different workload distribution that lowers costs. This is an interesting topic to take a further look into. The pack process namely follows the pick process, and if the workload distribution over days differs between the processes, it would lead to a lot of orders waiting between the process increasing lead times and reducing an efficient flow of orders through the process.

(25)

25 al., 2008), workload levels can be adapted to a more productive level by giving customers a price incentive for a certain days. The model developed in this research can be used to determine what the effect of a certain number of orders on a certain delivery day is on labor costs, which can be used to determine the price incentive.

It is expected that the cost reduction resulting from reduced demand uncertainty and less overtime can be generalized to other e-fulfillment warehouses. Except for the possibility of backlogging work, the workplace context of the model based on Pinker & Larson (2003) is characterized by key e-fulfillment warehouse issues. If the cost reduction found from increased productivity through workload control can be generalized to other e-fulfillment warehouses depends on the demand pattern and the relationship between workload and productivity. According to Agatz et al. (2008), there are specific demand patterns during the week in e-fulfillment. If demand follows a similar distribution over the week as in the warehouse where the data was collected, workload control can have the same effect on costs as long as there is a similar relationship between workload and productivity. It is very important to note that the nature of the relationship between workload control and productivity is essential in the possible cost reduction.

6. CONCLUSION

In this part, the research is summarized and conclusions are given to answer the research question. Also, recommendations following from the conclusions and possible further research issues are discussed.

In this thesis, a quantitative model was developed that analyzed the cost reduction possible through the introduction of differentiated lead times. The aim of the model was to answer the following research question:

Which labor cost reduction can be achieved in an e-fulfillment warehouse through differentiated lead times?

(26)

26 reduction of at most 9%. The results showed that when the fraction of demand with a longer lead time is larger, a larger cost reduction can be obtained.

Because of the cost reductions found, e-tailers and e-fulfillment warehouses are recommended to consider letting their customers determine the delivery day. However, the impact on other warehouse costs should be taken into account as well. The extra handling and space costs that may result from processing orders a day before they need to be delivered were not analyzed in this research. However, the same result can be obtained without processing orders earlier, by applying demand management. Although with demand management, costs will be incurred as well since customers need a price incentive to choose a certain delivery day. But, the additional handling and space costs may also break even with other cost reductions that were not taken into account. For instance, since peaks are reduced, the handling material can also be reduced. In future research, these costs may be taken into account as well to give a more complete picture of the influence of differentiated lead times on costs.

This research has only studied the theoretical introduction of differentiated lead times and workload control. However, the application of workload control involves organizational issues and technology investments that were not considered. Therefore, future empirical research needs to point out if the cost reduction found here also applies in practice.

The two cost reductions possible through differentiated lead times, from workload control and reduced demand uncertainty, were analyzed separately. Future research may combine these cost reducing factors into a single model to determine a more complete effect of differentiated lead times. The model will become more difficult because demand uncertainty will change when the release of orders with a longer lead time is changed. When orders are still released on the day of arrival, it is not clear what the effect on demand uncertainty will be.

(27)

27

BIBLIOGRAPHY

Agatz, N. A. H., Fleischmann, M., & van Nunen, J. A. E. E. 2008. E-fulfillment and multi-channel distribution – A review. European Journal of Operational Research, 187(2): 339–356.

Agatz, N. A. H. 2009. Demand Management in E-fulfillment. Erasmus Research Institute of Management (ERIM).

Baker, P. 2008. The design and operation of distribution centres within agile supply chains. International Journal of Production Economics, 111(1): 27–41.

Bertrand, J. W. M. & Van Ooijen, H. P. G. 2002. Workload based order release and productivity: A missing link. Production Planning & Control: The Management of Operations, 13(7): 665-678.

Breithaupt, J., Land, M., & Nyhuis, P. 2002. The workload control concept: Theory and practical extensions of Load Oriented Order Release. Production Planning & Control, 13(7): 625–638.

Bukchin, Y., Khmelnitsky, E., & Yakuel, P. 2012. Optimizing a dynamic order-picking process. European Journal of Operational Research, 219(2), 335–346.

Buyukkaramikli, N. C., Bertrand, J. W. M., & Ooijen, H. P. G. 2011. Periodic capacity management under a lead-time performance constraint. OR Spectrum, 35(1): 221–249. Chan, F. T. S., Humphreys, P., & Lu, T. H. 2001. Order release mechanisms in supply chain

management: A simulation approach. International Journal of Physical Distribution & Logistics Management, 31(2): 124–139.

Chen, F. 2001. Market segmentation, advanced demand information, and supply chain performance. Manufacturing & Service Operations Management, 3(1): 53–67.

Chen, S., & Chang, T. 2003. A descriptive model of online shopping process: Some empirical results. International Journal of Service Industry Management, 14(5): 556–569. De Koster, 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.

Fernandes, N. O., & Carmo-Silva, S. 2011. Workload control under continuous order release. International Journal of Production Economics, 131(1): 257–262.

Global Webshop Logistics. 2013. Global Webshop Logistics 2013 Report.

(28)

28 Gu, J., Goetschalckx, M., & McGinnis, L. F. 2010. Research on warehouse design and performance evaluation: A comprehensive review. European Journal of Operational Research, 203(3): 539–549.

Gue, K. R., Meller, R. D., & Skufca, J. D. 2006. The effects of pick density on order picking areas with narrow aisles. IIE Transactions, 38(10): 859–868.

Hariharan, R. & Zipkin, P. 1995. Customer-order information, lead-times, and inventory. Management Science, 41: 1599-1607

Heath, B. L., Ciarallo, F. W., & Hill, R. R. 2012. An agent-based modeling approach to analyze the impact of warehouse congestion on cost and performance. The International Journal of Advanced Manufacturing Technology, 67(1-4): 563–574.

Henn, S. 2012. Algorithms for on-line order batching in an order picking warehouse. Computers & Operations Research, 39(11): 2549–2563.

Henrich, P., Land, M., & Gaalman, G. 2004. Exploring applicability of the workload control concept. International Journal of Production Economics, 90: 187–198.

Hsieh, L. & Huang, Y. 2011. New batch construction heuristics to optimise the performance of order picking systems. International Journal of Production Economics, 131(2): 618–630.

Huang, S., Liao, B., & Robb, D. J. 2011. Inventory management in a (Q,r) inventory model with two demand classes and flexible delivery. International Journal of Innovative Computing Information and Control, 7(3): 1243-1254.

Karlsson, C. 2009. Researching Operations Management. Routledge: New York.

Kocaga, Y. L., & Sen, A. 2007. Spare parts inventory management with demand lead times and rationing. IIE Transactions, 39(9): 879-898.

Land, M. J. & Gaalman, G. J. C., 1998. The performance of workload control concepts in job shops: Improving the release method. International Journal of Production Economics, 56–57: 347–364.

Parikh, P. J., & Meller, R. D. 2009. Estimating picker blocking in wide-aisle order picking systems. IIE Transactions, 41(3): 232–246.

Pinker, E. J., & Larson, R. C. 2003. Optimizing the use of contingent labor when demand is uncertain. European Journal of Operational Research, 144(1): 39–55.

Polivka, A.E. 1989. On the definition of ‘contingent work’. Monthly Labor Review, 112(12): 9–16.

(29)

29 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 Journal of Operational Research, 122(3): 515–533.

Soepenberg, G. D., Land, M. J., & Gaalman, G. J. C. 2012. Adapting workload control for job shops with high routing complexity. International Journal of Production Economics, 140(2): 681–690.

Sarkar, S., & Shewchuk, J. P. 2013. Use of advance demand information in multi-stage production-inventory systems with multiple demand classes. International Journal of Production Research, 51(1): 57–68.

Tarn, J. M., Razi, M. A., Wen, H. J., & Perez Jr, A. A. 2003. E-fulfillment: The strategy and operational requirements. Logistics Information Management, 16(5): 350–362.

Tempelmeier, H. 2006. Supply chain inventory optimization with two customer classes in discrete time. European Journal of Operational Research, 174(1): 600-621.

Thürer, M., Stevenson, M., & Silva, C. 2011. Three decades of workload control research: A systematic review of the literature. International Journal of Production Research, 49(23): 6905–6935.

Van den Berg, J. P., & Zijm, W. H. M. 1999. Models for warehouse management: Classification and examples. International Journal of Production Economics, 59(1-3): 519–528.

Wang, Y., Cohen, M. A., & Zheng, Y. 2002. Differentiating customer service on the basis of delivery lead-times. IIE Transactions, 34(11): 979–989. .

Online Customers in Control: Reducing Warehouse Costs Through Differentiated Lead Times