Journal Pre-proof

Optimization of warehouse picking to maximize the picked orders

considering practical aspects

Kimiya Rahmani Mokarrari , Taraneh Sowlati , Jeffrey English ,

Michael Starkey

PII: S0307-904X(24)00324-X
DOI: https://doi.org/10.1016/j.apm.2024.06.037
Reference: APM 15585

To appear in: Applied Mathematical Modelling

Received date: 24 July 2023

Revised date: 18 June 2024
Accepted date: 21 June 2024

Please cite this article as: Kimiya Rahmani Mokarrari , Taraneh Sowlati , Jeffrey English ,
Michael Starkey , Optimization of warehouse picking to maximize the picked orders considering prac-
tical aspects, Applied Mathematical Modelling (2024), doi: https://doi.org/10.1016/j.apm.2024.06.037

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2024 Published by Elsevier Inc.

Highlights

• Proposed two mathematical modelling approaches for orders assignment and pickers
routing in warehouses.
• Included practical picking aspects, varying pickers’ capacity and items’ weight, in models.
• Maximized handled orders per shift while balancing pickers’ workload.
• Evaluated and compared the approaches by applying them to a real case study.

1
Optimization of warehouse picking to maximize the picked orders
considering practical aspects
Kimiya Rahmani Mokarraria, Taraneh Sowlatia 1, Jeffrey Englishb, Michael Starkeyb
a
Industrial Engineering Research Group, Department of Wood Science, University of British Columbia, Vancouver,
British Columbia, Canada; b Fujitsu Intelligence Technology Limited, Vancouver, British Columbia, Canada.

Abstract
Majority of previous studies on optimization of warehouse order picking lack real-world
application and overlook complexities such as varying items’ weight, workload balance, and
heterogeneous pickers. To address these issues, in this study, mathematical programming models
are developed based on two approaches to assign orders to pickers and determine pickers’ routes
in a North American consumer goods warehouse. Both approaches maximize the number of picked
orders. However, in the simultaneous modelling approach, decisions are made in one model, while
in the sequential modelling approach, first routing decisions then order assignment decisions are
made. The sequential modelling approach compared with the simultaneous one yields better results
in a shorter time. For the considered warehouse, the results suggest that the warehouse may not be
able to handle more than 700 orders per shift without expanding its workforce. Furthermore,
adding one picker results in an approximate 2% increase in picked orders. Additionally, balancing
pickers' workload is more effective when there are fewer than 600 orders available for picking per
shift, because with a higher number of orders, pickers remain busy throughout the entire shift. The
proposed mathematical modelling approaches enable warehouse managers and staff to process
more orders with available resources (equipment, time, staff) and make better decisions using
available data.

Keywords: order picking, warehouse optimization, routing optimization, order assignment,

balanced workload, workload equity

1
Corresponding author: Taraneh Sowlati; [email protected]; Industrial Engineering Research Group,
Department of Wood Science, University of British Columbia, 2931-2424 Main Mall, Vancouver, BC V6T-1Z4.

2
1. Introduction
Warehouses and distribution centers play an important role in supply chains and logistics systems
[1]. The primary operations within these facilities encompass receiving products, managing
inventory, assigning storage location to products, picking, sorting, shipping and delivering orders
[2]. In most warehouses, order picking, which refers to retrieving items from storage locations to
fulfill customers’ orders, is a key operation [2]. Widely recognized as one of the most labor-
intensive, time-consuming, and costly processes in warehouses [3, 4], order picking can account
for more than 50% of total warehouse operating costs [5, 6]. Therefore, optimizing the order
picking process can potentially reduce costs, enhance efficiency in warehouses, and facilitate the
timely delivery of products, which are essential for the survival of companies in today’s
competitive market [7].

In order to improve the order picking process, which can significantly impact warehouse
efficiency [3, 8], different optimization models and heuristic methods have been developed in the
literature. These models and methods were primary designed to make decisions regarding the
assignment of orders to pickers, batching of orders, and routing of pickers to streamline the order
picking process and reduce associated costs [3]. While research on these decisions contributes to
the literature, findings of academic papers are often unacceptable to warehouse managers as
academic papers often overlook the practical aspects of order picking [9]. Pinto and Nagano [4]
also stated that the majority of the proposed models for order picking problems in the literature
have not been applied to real case studies nor have considered the complexities of warehouses in
real applications.

This paper is based on a real case study of a warehouse located in North America in which various
complexities and practical aspects of real-world warehouses have been considered. These practical
aspects include the presence of multiple pickers, variations in picking vehicle and product
properties, and workload equity considerations, which are often overlooked in picking planning
[4, 9, 10]. Workload balancing becomes a critical issue in warehouses where large quantities of
orders are picked daily within a specific time limit [10]. Planning order picking operations and
improving efficiency in warehouses inevitably raises concerns about workload imbalance [11].
While workload balancing has been a popular research topic within manufacturing or assembly
lines, it has been less explored in the context of warehouse picking [10]. Accounting for these

3
factors in picking models is crucial as overlooking them can lead to infeasible solutions in real-
life cases [4, 12].

In our case study, the goal of warehouse managers was to maximize the number of picked orders
(i.e., fulfilled orders) in each working shift, while having a balanced workload for the pickers. It
is worth noting that, in this study, the picker’s workload is measured by the total working time of
each picker in a shift.

To achieve the warehouse managers’ goal, this study provides order picking planning models that
determine the assignment of orders to pickers, the sequence of orders that should be picked, and
the routing of pickers within the warehouse. To the best of authors’ knowledge, previous models
proposed for order picking have primarily focused on objectives including minimizing the picking
time, tardiness, picking distance, or costs for a fixed number of orders. To minimize these
objectives, it is essential to assume a fixed number of orders that should be fulfilled and assigned
to pickers and to treat this quantity as an input to the models.

Therefore, existing studies fail to address the critical question of determining the number of orders
a warehouse can handle and fulfill using its available resources, which is an essential piece of
information for the managers. This gap highlights the absence of models capable of determining
the maximum number of orders that can be fulfilled from the available customer orders within a
specific shift, given a certain number of pickers. Maximizing the number of fulfilled orders has an
indirect effect on reducing the picking time for each order. This is because, in order to fulfill a
greater quantity of orders, it becomes essential to pick each order more quickly. Having a clear
understanding of the maximum number of orders that the warehouse can handle helps managers
effectively plan the picking process, appropriately allocate resources, and mitigate the risk of
unfulfilled orders.

To address the problem at hand, mathematical models using two approaches (simultaneous and
sequential) are developed and tested on the case study. The performances of these models are
evaluated and compared. These mathematical models will provide a novel decision-making tool
designed to align with real-world scenarios, offering valuable insights for warehouse managers.
This work makes significant contribution both to the industry and the existing order picking
literature. Specifically, this study contributes to the industry by providing a decision-making
support tool that helps managers make informed and improved decisions. Additionally, it

4
contributes to the literature by considering a different objective function for picking planning and
presenting two mathematical programming models that approach the same problem from different
perspectives, emphasizing the importance of the effective mathematical modeling. Furthermore,
the proposed models illustrate the relationship between pickers’ workload balance, the number of
pickers, and the number of orders that can be completely fulfilled within a working shift, which
has not been addressed in the literature.

2. Literature review
The focus of this literature review is on picker-to-parts systems, considering that the majority of
current picking systems still operate on a picker-to-parts basis, wherein pickers collect ordered
items [4, 9]. This prevalence is due to the high costs associated with transitioning to automated
picking systems [13].

In addition to the picking system, different policies can be employed for the picking process. Batch
picking is a policy where orders are first grouped into batches and then batches are assigned to
pickers, allowing them to pick multiple orders simultaneously. Wave picking refers to a policy
where orders with the same shipment due dates are picked together. In the zone picking, the
warehouse is divided into different zones, then each picker is assigned to a specific zone to pick
items from orders located in that zone. Lastly, strict order picking is a policy where pickers traverse
the warehouse to pick the items of one order in each tour [14, 15].

Majority of papers in the context of warehouse picking optimization focused on batch picking
policy. In a batch picking policy, decisions such as order batching, batch assignment, batch
sequencing, and picker’s routing can be addressed to meet a certain criterion, such as time
minimization [4]. Simultaneous consideration of these decisions leads to more complex
mathematical programming models (e.g., [16-18]). Wave picking is often linked with batch
picking, as the picking time of a wave can be significantly affected by how orders are batched [19].
Previous studies, such as [19] and [20] addressed order batching, batch assignment and picker’s
routing decisions, within a warehouse considering wave picking policy.

The zone picking policy can improve the travel time, congestions in the warehouse, and
consequently the throughput of a picking system [21]. Various warehouse picking decisions have
been investigated in the context of zone picking, including storage assignment [22], scheduling

5
[10], routing [21], and batching [23]. Similar to wave picking, zone picking has often been
addressed simultaneously with batch picking in each zone.

In a strict picking or picking-by-orders policy, the picking process is less prone to errors. However,
it can be costlier and more time-consuming compared to picking by batches [4]. Despite these
disadvantages, the strict picking is still often preferred by firms due to its simple implementation
and accuracy [24, 25]. Moreover, for large order sizes, there is a little difference between the total
picking time of strict picking policy and that of the batch picking policy [24]. This might be another
reason why some firms prefer strict order picking policy over batch picking [24]. Despite the
commonality and advantages of strict order picking, less attention has been given to this picking
policy in the literature. The reviewed papers in the strict picking category mostly focused on
picker’s routing decisions [15, 26-28]. There are also papers focusing on more complex problems
that include simultaneous storage assignment and picker routing decisions [29].

Regardless of the picking policy, incorporating multiple pickers increases the complexity of
models. However, it can provide a more accurate representation of real-world warehouse
scenarios. When multiple pickers are considered under order batching policy, the batch assignment
decision is usually addressed in order to specify the pick list of each picker as observed in [6, 16-
18, 30-32]. Majority of papers in batch picking, wave picking, and zone picking have considered
multiple pickers. The strict order policy is the only category with more papers considering one
picker.

As shown in Table 1, the majority of papers, regardless of the picking policy, focus on models
aimed at minimizing time-related aspects of the picking process. This includes minimization of
tardiness, travel time, make-span, picking completion time, and even travelled distance. A few
other papers aimed at minimizing costs as their objective function, and only one paper considered
energy usage as the objective function [26]. The common assumption in such models is that a
certain number of orders should be picked within a specific time window, while minimizing the
mentioned objectives (e.g., [6, 19, 20, 31, 33, 34]). If this assumption is not met, no orders will be
processed, as not picking any orders would result in the minimum travel time or distance (i.e.,
zero) and incur no picking costs.

The mentioned objective functions help design optimized picking plans that reduce the time and
cost of order fulfillment. However, they fail to determine the maximum number of orders from the

6
available customer orders (i.e., demand) that the warehouse can handle in each planning period
with its available resources. It is crucial for warehouse managers to know this information to: (1)
utilize the available resources (e.g., time, picking staff, etc.) appropriately, (2) avoid overloading
the warehouse with too many orders, (3) guarantee timely customer order fulfillment, and (4)
prevent costs associated with late delivery or unfulfilled orders. Moreover, maximizing the number
of picked orders (i.e., fulfilled orders) eliminates the need for additional steps or tools to first
indicate the number of orders that can be fulfilled and then minimize the picking time or cost for
them. This aspect of order picking has not been investigated in the reviewed papers.

Another important conclusion from the literature review is that only a few papers considered
multiple aspects of warehouse picking simultaneously (e.g., [20, 22, 23, 31]). Accounting for
multiple aspects of picking is important because satisfying one picking objective might be in
conflict with other equally important ones. Therefore, there is a need to develop decision-making
tools and multi-objective models that can cover different priorities and objectives of warehouse
managers at the same time.

While the models in warehouse picking literature can potentially enhance the efficiency in a
warehouse, the findings of academic papers are often not acceptable to warehouse managers as
academic papers usually overlook the practical aspects of order picking [9]. Similar to the literature
review conducted by Pinto and Nagano [4], our literature review concludes that majority of the
proposed models for order picking problems in the literature have not been applied to real case
studies nor have considered the complexities of warehouses in real applications.

The complexities of real warehouses can stem from varying picking equipment types, varying
weights of items, and human factor challenges. Workload balancing is one of these human factor
challenges in warehouses [10]. While workload imbalance can be an issue in warehouses with any
picking policy, only papers related to zone picking policy have considered workload equity in their
models. Another mentioned practical factor is considering pickers' properties, such as capacity of
pickers or picking vehicles that can differ from one picker to another based on the equipment they
use. This factor has been taken into account by a few studies, such as [29, 31, 34], but often
overlooked. Specific properties of items, such as weight or volume, is another important practical
factor covered by some studies (e.g., [6, 26]), but often overlooked in the literature. Some papers
considered the weights of orders or the capacity required to fulfill each order (e.g., [6, 30, 31, 35]),

7
rather than focusing on the properties of each individual product. Accurate knowledge on such
properties is essential, and incorporating them in planning models leads to the development of
more practical models and greater usefulness of optimization results for warehouse managers.
Varying characteristics such as capacity and travelling speed of picking equipment pieces have
also been considered by a few studies (e.g., [6, 29]).

In summary, this study addresses the concern of determining the maximum number of orders a
warehouse can fulfill within a specific shift while balancing the workload for pickers. By
considering picking aspects, such as multiple pickers with varying capacities, item weights, and
workload balance, this research provides valuable insights for optimizing the picking process and
resource allocation. Unlike existing studies that often overlook these picking aspects, the current
research employs mathematical models to develop a decision-making tool aligned with real-world
situations. The developed models in this study not only contribute to industry practices by offering
managers informed decision-making support but also enrich the literature by introducing new
perspectives on order picking optimization. Furthermore, the study sheds light on the relationship
between pickers' workload balance, the number of pickers, and the fulfillment capacity of the
warehouse, a dimension that has been largely unexplored in prior research.

Table 1 summarizes the reviewed papers based on picking policy, picking decisions, number of
pickers, objective function, practical aspects, and whether or not a case study was considered.

8
 Table 1. Papers related to warehouse order picking grouped based on different picking policies.

Practical factors
Picking Number Case
Picking decisions Objective function Workload Varying picking Product Reference
policy of pickers study
equity vehicle property properties

• Order batching
• Batch assignment Many Min. tardiness [32]
• Batch sequencing

One Min. travelled distance [33, 36]

• Order batching
• Picker routing
Many Min. costs Capacity [34]

• Order batching
• Batch assignment Many Min. tardiness Capacity Order weights ✓ [6]
• Picker routing
• Order batching
• Batch sequencing One Min. tardiness Order weights [35]
• Picker routing
Batch
Min. tardiness [17]
picking
Min. picking time ✓ [18]
• Min. travelled distance
• Min. travelled time Order capacities [30]
• Min. tardiness
• Order batching
• Batch assignment • Min. travelled distance
Many
• Batch sequencing • Min. tardiness
• Capacity Order capacities [31]
Picker routing • Min. travelled time,
earliness and tardiness
Product weights
Min. pickers' completion
Product search [16]
time
& pick time

9
 Practical factors
Picking Number Case
Picking decisions Objective function Workload Varying picking Product Reference
policy of pickers study
equity vehicle property properties

Min. makespan ✓ [19]

• Order batching
• Min. makespan
Wave • Batch assignment
• Min. active pickers [20]
picking • Picker routing
Many (workload)
/Batch
picking • Load assignment
Min. costs Capacity [37]
• Picker routing

Routing Min. max. travelled time ✓ [21]

Many • Min. replenishment

Storage assignment • Min. total workload ✓ [22]
imbalance

Min. workload
Zone Scheduling - ✓ [10]
imbalance
picking

• Min. picking time

• Min. standard
Order batching - ✓ [23]
deviation of picking
time

10
 Practical factors
Picking Number Case
Picking decisions Objective function Workload Varying picking Product Reference
policy of pickers study
equity vehicle property properties
[27]
Min. travelled distance [28]
One [15]
Min. energy
Picker routing Product weights [26]
consumption
Strict
Product weights
picking Min. travelled distance [38]
Many & volume
Min. order service time [39]

• Storage One Min. costs [40]

assignment
Many Min. picking time Travelling speed Product weights [29]
• Picker routing

11
3. Material and methods
3.1. Case study
This paper was inspired by a real case study of the warehouse of a consumer goods distribution
company located in North America. The warehouse of this company is a multi-block warehouse
with a rectangular shape and a total space of 9,488 pallets. The warehouse has twelve main storage
areas (i.e., blocks), each of which containing specific types of products. A schematic layout of the
warehouse is provided in Figure 1.

Figure 1. Schematic layout of the warehouse.

Orders received by the warehouse are picked by 12 pickers per shift (eight-hour). Pickers are
divided into two groups based on the equipment they use. Eight pickers are pallet jack operators,
and four pickers are forklift operators. A pallet jack has a capacity of 2500 kg and a forklift’s
capacity is double than that of a pallet jack. According to the data received from the warehouse, in
the majority of picking cases, each order is picked by a single picker. Therefore, strict order picking
policy is assumed as the picking policy for this warehouse.

Based on the daily order records, covering a period of eight months from January to August 2022,
that the warehouse provided to us, the highest volume of orders received by the warehouse was
764 per day, and the lowest number of received daily orders was two orders, which were received
during a weekend. The average daily number of orders that the warehouse receives is

12
approximately 300 orders. Figure 2 shows the daily orders received by the warehouse in eight
months.

800

700

600
Number of orders

500

400

300

200

100

0
21-Mar

1-Jul
7-Jul

13-Aug
19-Aug
25-Aug
31-Aug
1-Aug
7-Aug
14-Jan

13-Feb
19-Feb
25-Feb

14-Apr
20-Apr
26-Apr

1-Jun
7-Jun
15-Mar

27-Mar
2-Apr
8-Apr

14-May
20-May
26-May
1-Feb
7-Feb

3-Mar
9-Mar

2-May
8-May
20-Jan
26-Jan
2-Jan
8-Jan

13-Jun
19-Jun
25-Jun

13-Jul
19-Jul
25-Jul
Jan Feb Mar Apr May Jun Jul Aug
Dates

Figure 2. Number of daily orders received by the warehouse from the beginning of January to the end of August
2022.
The warehouse management team were interested in a picking plan that maximized the volume of
orders handled (i.e., fulfilled) on each shift with the same number of pickers. Moreover, they
wanted to have a balanced workload for their pickers as much as possible and placed great
emphasis on using all the pickers.

3.2. Problem definition

Based on the case study and communications with the warehouse managers, the optimization
problem could be defined as a warehouse picking problem with a picker-to-part system and strict
picking policy. The purpose is to maximize the volume of orders fulfilled (picked) on each working
shift and to balance the workload of a fixed number of pickers while determining the assignment
of orders to pickers and specifying the tours that pickers should take in the warehouse to collect
the orders that were assigned to them. Determining the order assignment and pickers’ routes also
yields the working time of each picker in total.

13
The following conditions and assumptions are considered in the problem definition and
mathematical modeling based on the current situation at the warehouse and the comments provided
by warehouse managers:

• The warehouse has a dedicated storage strategy, where each item has a known and fixed
storage location.
• Each picker has a limited load-carrying capacity, which varies based on their picking
equipment.
• Pickers start their tours from a depot and return to it when picking an order is done.
• On each tour taken by a picker, one order is picked (i.e., strict order picking policy).
• Partial order picking is not allowed.
• There are enough items available in the warehouse to fill all orders on each shift.
• The searching time and retrieval time are considered in the travel time between two storage
locations and are the same for all types of items. For example, the travel time between
storage locations i and j includes the time that an average picker takes to go from i to j (x
minutes) plus an additional y minutes for searching and retrieval at storage location j.
• Items in a specific storage location are the same.

3.3. Mathematical models

In this section, two approaches are taken to provide mathematical models for the introduced
problem. The simultaneous modelling approach makes order assignment and picker routing
decisions at the same time to maximize the number of picked orders and minimize the imbalance
workload among pickers. In the sequential modelling approach, first orders’ routes are constructed
with the objective of minimizing the routing time, and then orders are assigned to the pickers. The
details of the two approaches and the rationale behind them are explained in following sub-
sections.

3.3.1. Simultaneous modelling approach

The elements of the developed simultaneous mathematical model, including the objective function,
constraints and decisions are shown in Table 2.

14
 Table 2. Elements of the simultaneous modelling approach.

Simultaneous order assignment and routing

Objective function:
• Maximize the assigned orders (i.e., fulfilled orders) to pickers and minimize workload
imbalance
Constraints:
• Start and end the tours at depot
• Visit storage location of all items of an order and visit each location once for each order
• Tour continuity
• Sub-tour elimination
• Picker’s capacity
• Available picking time
• Difference between pickers working time and the average workload
Outputs (decisions):
• Assignment of orders to pickers (𝑥𝑜𝑟 )
• Paths that each picker takes to collect items of an order (𝑦𝑜𝑖𝑗𝑟 )
• Completion time of picking an order by a picker (𝑝𝑜𝑟 )
• Total working time of each picker (∑𝑜∈𝑂 𝑝𝑜𝑟 )

The required sets, parameters, and variables to develop the model are listed in Table 3.

Table 3. Sets, parameters, and variables used to develop the simultaneous mathematical model.
Sets Description
𝐼 Set of storage locations (including the depot S) 𝑖, 𝑗, 𝑘 ∈ {1,2, … , 𝐺} ∪ {𝑆}
𝑅 Set of pickers 𝑟 ∈ {1,2, … , 𝑅′ }
𝑂 Set of orders 𝑜 ∈ {1,2 … , 𝑂′ }
Parameters Description
𝐶𝑟 Load carrying capacity of picker 𝑟
𝐷𝑖𝑗 Travel time between storage locations 𝑖 and 𝑗
𝑄𝑜𝑖 Quantity of item from storage location 𝑖 in order 𝑜
𝐾𝑜𝑖 1 if an item from storage location 𝑖 is in order 𝑜; 0 otherwise
𝐿𝑖 Load of an item from storage location 𝑖
𝑀 Big M – a large number used in sub-tour elimination constraints
𝑇𝑀 Maximum available time for picking
𝐴𝑊 Average workload

15
 Binary variables Description
𝑥𝑜𝑟 1 if order 𝑜 is assigned to picker 𝑟; 0 otherwise
𝑦𝑜𝑖𝑗𝑟 1 if storage location 𝑖 is visited right before storage location 𝑗 in order 𝑜 by
picker 𝑟; 0 otherwise
Positive variables Description
𝑡𝑜𝑖𝑟 Time at which picker 𝑟 enters storage location 𝑖 when picking order 𝑜
𝑝𝑜𝑟 Picking completion time of order 𝑜 by picker 𝑟
𝑢𝑟 Absolute difference between the workload of picker 𝑟 and the average
workload

The first term in the objective function, which is shown in equation (1), maximizes the number of
orders assigned to all pickers – in other words, fulfilled orders. The second term of the objective
function balances the workload by minimizing the summation of differences between each picker’s
workload and the average workload, over all pickers. The same measure for workload balance was
used in a log transportation optimization study [41] to balance the number of assigned truckloads
to contractors. In order to estimate the average workload in our study, the following optimization
model was run without considering the workload balance (i.e., excluding the second term of the
objective function and constraint (20)), and then the obtained total picking time was divided by
∑𝑜∈𝑂 ∑𝑟∈𝑅 𝑝𝑜𝑟
the number of pickers (i.e., ).
|𝑅|

𝑊1 and 𝑊2 are parameters and show the relative importance of the two terms in the objective
function. The summation of 𝑊1 and 𝑊2 equals one.

𝑀𝑎𝑥. (𝑊1 ∑ ∑ 𝑥𝑜𝑟 − 𝑊2 ∑ 𝑢𝑟 ) (1)

𝑜∈𝑂 𝑟∈𝑅 𝑟∈𝑅

Constraint set (2) shows that each order can be assigned to at most one picker.

∑ 𝑥𝑜𝑟 ≤ 1 ∀𝑜 ∈ {1, … , 𝑂′ } (2)

𝑟∈𝑅

Constraint set (3) accounts for the maximum load of items that each picker can carry when picking
an order. In other words, when an order is assigned to a picker, the summation of the weights of
all the items on that order needs to be less than or equal to the load-carrying capacity of the picker
that has been assigned to that order.

16
 ∑ 𝑄𝑜𝑖 𝐿𝑖 𝑥𝑜𝑟 ≤ 𝐶𝑟 ∀𝑟 ∈ {1, … , 𝑅 ′ }, 𝑜 ∈ {1, … , 𝑂′ } (3)
𝑖∈𝐺∪𝑆

Constraint set (4) limits the number of times that each picker can visit a storage location to at most
one for each order on each picking tour.

∑ 𝑦𝑜𝑖𝑗𝑟 ≤ 1 ∀𝑜 ∈ {1, … , 𝑂′ }, 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (4)

𝑖∈𝐺∪𝑆,𝑖≠𝑗

Constraint sets (5) and (6) show that if an order is assigned to a picker, the picker should start and
end their picking tour at the depot.

∑ 𝑦𝑜𝑖𝑗𝑟 = 𝑥𝑜𝑟 ∀𝑖 ∈ {𝑆}, 𝑟 ∈ {1, … , 𝑅′}, 𝑜 ∈ {1, … , 𝑂′} (5)

𝑗∈𝐺

∑ 𝑦𝑜𝑖𝑗𝑟 = 𝑥𝑜𝑟 ∀𝑗 ∈ {𝑆}, 𝑟 ∈ {1, … , 𝑅′}, 𝑜 ∈ {1, … , 𝑂′} (6)

𝑖∈𝐺

Constraint set (7) indicates that if an order is assigned to a picker, meaning that order assignment
variable (𝑥𝑜𝑟 ) gets the value of one, then the picker should visit all the storage locations of the
items that are on that order.

𝑄𝑜𝑗 𝑥𝑜𝑟 ≤ 𝑀 ( ∑ 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (7)
𝑖∈𝐺∪𝑆

While constraint set (7) ensures pickers visit the storage location of items on the assigned orders
(i.e., fulfilled orders), it does not necessarily avoid visiting storage location of items that are not
on the assigned order’s list. The reason is that when 𝑥𝑜𝑟 is equal to zero, 𝑦𝑜𝑖𝑗𝑟 on the right-hand
side can still get values. To avoid this situation, constraint set (8) needs to be added to ensure that
pickers would only visit a storage location if that item is on the order’s list assigned to them.

∑ 𝑦𝑜𝑖𝑗𝑟 ≤ 𝐾𝑜𝑗 ∀𝑜 ∈ {1, … , 𝑂′}, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (8)

𝑖∈𝐺∪𝑆

Constraint set (9) ensures the continuity of pickers’ tours, which means that when a picker visits a
storage location on a picking tour, they should leave that storage location to another one.

17
 ∑ 𝑦𝑜𝑖𝑗𝑟 = ∑ 𝑦𝑜𝑗𝑘𝑟 ∀𝑜 ∈ {1, … , 𝑂′}, 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (9)
𝑖∈𝐺∪𝑆,𝑖≠𝑗 𝑘∈𝐺∪𝑆,𝑘≠𝑗

Constraint set (10) prohibits tour creation for a single storage location (node).

𝑦𝑜𝑖𝑗𝑟 = 0 ∀𝑜 ∈ {1, … , 𝑂′}, 𝑟 ∈ {1, … , 𝑅′}, 𝑖, 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆}; 𝑖 = 𝑗 (10)

Constraint sets (11) and (12) calculate the time at which a picker enters the first storage location
after leaving the depot. When a picker goes from the depot to a storage location, 𝑡𝑜𝑗𝑟 gets assigned
a value after the first stop based on the traveling time between the first stop and the depot. It should
be noted that constraint set (11) alone is not sufficient to indicate the time value (𝑡𝑜𝑗𝑟 ) as it does
not limit the value of 𝑡𝑜𝑗𝑟 . Therefore, constraint set (12) is needed to create a limit for 𝑡𝑜𝑗𝑟 , so that
the two constraint sets make 𝑡𝑜𝑗𝑟 equal to time at which a picker enters a storage location. M is a
big number used in these constraints to neutralize the effect of these constraint in case 𝑦𝑜𝑖𝑗𝑟 = 0.

𝑡𝑜𝑗𝑟 ≥ 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {𝑆}, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (11)

𝑡𝑜𝑗𝑟 ≤ 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {𝑆}, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (12)

Constraint sets (13) and (14) eliminate the subtours and calculate the time at which a picker enters
a storage location while picking an order. The subtour elimination formulation used in this study
is known as MTZ subtour elimination [42].

According to constraint set (13), between each consecutive pair of 𝑖 − 𝑗 (i.e., storage locations,
except for the depot), the time at which the picker enters storage location 𝑗 should be higher than
the time at which the picker visited storage location 𝑖. In this way, the unconnected subtours are
prevented. Again, in here, constraint set (14) needs to be added to limit the value of 𝑡𝑜𝑗𝑟 when 𝑖 −
𝑗 is within the picking tour. These two constraint sets make the value of 𝑡𝑜𝑗𝑟 equal to the time at
which the picker visited previous storage location (𝑡𝑜𝑗𝑟 ) plus the travel time between storage
locations 𝑖 and 𝑗.

18
 𝑡𝑜𝑗𝑟 ≥ 𝑡𝑜𝑖𝑟 + 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (13)

𝑡𝑜𝑗𝑟 ≤ 𝑡𝑜𝑖𝑟 + 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (14)

Constraint sets (15) and (16) are similar to constraint sets (13) and (14). The only difference is that
these constraint sets calculate the time at which the picker returns back to the depot.

𝑡𝑜𝑗𝑟 ≥ 𝑡𝑜𝑖𝑟 + 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {1, … , 𝐺}, 𝑗 ∈ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (15)

𝑡𝑜𝑗𝑟 ≤ 𝑡𝑜𝑖𝑟 + 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗𝑟 ) ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {1, … , 𝐺}, 𝑗 ∈ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (16)

While constraint sets (11) – (16) determine the times at which a picker visits a storage location
(i.e., 𝑡𝑜𝑗𝑟 ) if that location is within pickers’ tour (i.e., 𝑦𝑜𝑖𝑗𝑟 = 1), they cannot limit the values of
𝑡𝑜𝑗𝑟 to zero when a storage location is not visited in a picking tour (i.e., 𝑦𝑜𝑖𝑗𝑟 = 0). Therefore,
constraint set (17) is added to ensure that if a storage location is not visited by a picker, then the
value of the picker’s visiting time at that location is zero.

𝑡𝑜𝑗𝑟 ≤ 𝑇𝑀 ∑ 𝑦𝑜𝑖𝑗𝑟 ∀ 𝑜 ∈ {1, … , 𝑂′}, 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (17)

𝑖∈𝐺∪𝑆

Constraint set (18) indicates the picking completion time of an order by a picker, which is equal
to the time that a picker returns to the depot.

𝑝𝑜𝑟 = 𝑡𝑜𝑖𝑟 ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {𝑆}, 𝑟 ∈ {1, … , 𝑅′} (18)

Constraint set (19) ensures that the completion time of all orders by a picker would not exceed the
total available time for picking.

∑ 𝑝𝑜𝑟 ≤ 𝑇𝑀 ∀𝑟 ∈ {1, … , 𝑅′} (19)

𝑜∈𝑂

19
Constraint set (20) calculates the difference between a picker’s workload (i.e., total picking time)
and the average workload. With 𝑢𝑟 in the objective function, this variable 𝑢𝑟 will get the minimum
possible value.

|∑ 𝑝𝑜𝑟 − 𝐴𝑊| ≤ 𝑢𝑟 ∀𝑟 ∈ {1, … , 𝑅′} (20)

𝑜∈𝑂

As constraint set (20) makes the model non-linear, it is divided into the following two constraints
shown in (20-a) and (20-b) to avoid non-linearity.

∑ 𝑝𝑜𝑟 − 𝐴𝑊 ≤ 𝑢𝑟 ∀𝑟 ∈ {1, … , 𝑅′} (20-a)

𝑜∈𝑂

∑ 𝑝𝑜𝑟 − 𝐴𝑊 ≥ −𝑢𝑟 ∀𝑟 ∈ {1, … , 𝑅′} (20-b)

𝑜∈𝑂

The last two constraints, (21) and (22), show the domains and ranges of variables.

𝑦𝑜𝑖𝑗𝑟 , 𝑥𝑜𝑟 ∈ {0,1} ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖, 𝑗 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (21)

𝑡𝑜𝑖𝑟 , 𝑝𝑜𝑟 , 𝑢𝑟 ≥ 0 ∀𝑜 ∈ {1, … , 𝑂′}, 𝑖 ∈ {1, … , 𝐺}, 𝑟 ∈ {1, … , 𝑅′} (22)

3.3.2. Sequential modelling approach

This approach is another way of looking at the problem at hand. As the picking policy of the
warehouse of our case study is strict picking, the picking process of each order is independent of
other orders. Therefore, instead of a single model that maximizes the number of picked orders, two
separate models can be developed to serve the same purpose. That is, first, the picking route for
picking each order is determined in order to minimize the picking time of each order individually.
This model is executed for each order to determine the required time and tour for picking each of
them. Then, in the second model, the orders (considering their required picking time) are assigned
to pickers to maximize the number of assigned orders (i.e., fulfilled orders). In this way, instead
of minimizing the picking time indirectly by maximizing the number of picked orders, the time

20
aspect of picking is directly taken into account and minimized. Figure 3 shows the steps of the
sequential modelling approach.

Phase 2: Assignment of orders to pickers and

Phase 1: Routing of orders
balancing workload
Objective function:
Objective function: • Maximize the assigned orders (i.e.,
• Minimize the picking time of each order fulfilled orders) to pickers and minimize
Constraints: imbalance workload

Inputs
• Start and end the tours at depot Constraints:
• Visit storage location of all items on an • Picker’s capacity
order • Available picking time
• Tour continuity • Difference between pickers working time
• Sub-tour elimination and the average workload
Outputs (decisions): Outputs (decisions):
• Paths that should be taken to collect items • List of orders assigned to pickers (𝑥𝑜𝑟 )
of an order (𝑦𝑜𝑖𝑗 ) • Total working time of each picker
• Completion time of picking an order (𝑝𝑜 ) (∑𝑜∈𝑂 𝑝𝑜 𝑥𝑜𝑟 )

Figure 3. Steps and elements of the sequential modelling approach.

The decision variables for the sequential model are shown in Table 4. The sets and parameters
used in this model are the same as those defined in Table 3.

Table 4. Variables used to develop the sequential mathematical model

Binary variables Description
𝑥𝑜𝑟 1 if order 𝑜 is assigned to picker 𝑟; 0 otherwise
𝑦𝑜𝑖𝑗 1 if storage location 𝑖 in order 𝑜 is visited right before storage location 𝑗 in
order 𝑜
Positive variables Description
𝑡𝑜𝑖 Time at which storage location 𝑖 is visited when picking order 𝑜
𝑝𝑜 Picking completion time of order 𝑜
𝑢𝑟 Maximum difference between the workload of picker 𝑟 and the average
workload

The mathematical model for each phase of the sequential modelling approaching approach is as
follows:

21
Phase 1

The objective function (23) of this model is to minimize the travel time for picking. Constraint sets
(24) – (36) serve the same purpose as constraint sets (5), (6), and (8) – (18) in Section 3.3.1.

Note that the following model should be executed for each order (i.e., o) in the set of customer
orders (i.e., {1, … , O′ }):

𝑀𝑖𝑛. ∑ ∑ 𝐷𝑖𝑗 𝑦𝑜𝑖𝑗 (23)

𝑖∈𝑆∪𝐺 𝑗∈𝑆∪𝐺

Subject to:

∑ 𝑦𝑜𝑖𝑗 = 1 ∀𝑖 ∈ {𝑆} (24)

𝑗∈𝐺

∑ 𝑦𝑜𝑖𝑗 = 1 ∀𝑗 ∈ {𝑆} (25)

𝑖∈𝐺

∑ 𝑦𝑜𝑖𝑗 = 𝐾𝑜𝑗 ∀𝑗 ∈ {1, … , 𝐺} (26)

𝑖∈𝐺∪𝑆

∑ 𝑦𝑜𝑖𝑗 = ∑ 𝑦𝑜𝑗𝑘 ∀𝑗 ∈ {1, … , 𝐺} (27)

𝑖∈𝐺∪𝑆 𝑘∈𝐺∪𝑆

𝑦𝑜𝑖𝑗 = 0 ∀𝑖, 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆}; 𝑖 = 𝑗 (28)

𝑡𝑜𝑗 ≥ 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀ 𝑖 ∈ {𝑆}, 𝑗 ∈ {1, … , 𝐺} (29)

𝑡𝑜𝑗 ≤ 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀ 𝑖 ∈ {𝑆}, 𝑗 ∈ {1, … , 𝐺} (30)

𝑡𝑜𝑗 ≥ 𝑡𝑜𝑖 + 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀𝑖, 𝑗 ∈ {1, … , 𝐺} (31)

𝑡𝑜𝑗 ≤ 𝑡𝑜𝑖 + 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀𝑖, 𝑗 ∈ {1, … , 𝐺} (32)

𝑡𝑜𝑗 ≥ 𝑡𝑜𝑖 + 𝐷𝑖𝑗 − 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀𝑖 ∈ {1, … , 𝐺}, 𝑗 ∈ {𝑆} (33)

22
 𝑡𝑜𝑗 ≤ 𝑡𝑜𝑖 + 𝐷𝑖𝑗 + 𝑀(1 − 𝑦𝑜𝑖𝑗 ) ∀𝑖 ∈ {1, … , 𝐺}, 𝑗 ∈ {𝑆} (34)

𝑡𝑜𝑗 ≤ 𝑇𝑀 ∑ 𝑦𝑜𝑖𝑗 ∀ 𝑗 ∈ {1, … , 𝐺} ∪ {𝑆} (35)

𝑖∈𝐺∪𝑆

𝑝𝑜 = 𝑡𝑜𝑖 ∀𝑖 ∈ {𝑆} (36)

Phase 2

The objective function (37) is the same as the model in Section 3.3.1. Constraint sets (38) – (41)
serve the same purpose as constraint sets (2), (3), (19) and (20) in Section 3.3.1.

𝑀𝑎𝑥. (𝑊1 ∑ ∑ 𝑥𝑜𝑟 − 𝑊2 ∑ 𝑢𝑟 ) (37)

𝑜∈𝑂 𝑟∈𝑅 𝑟∈𝑅

Subjected to:

∑ 𝑥𝑜𝑟 ≤ 1 ∀𝑜 ∈ {1, … , 𝑂′ } (38)

𝑟∈𝑅

∑ 𝑄𝑜𝑖 𝐿𝑖 𝑥𝑜𝑟 ≤ 𝐶𝑟 ∀𝑟 ∈ {1, … , 𝑅 ′ }, 𝑜 ∈ {1, … , 𝑂′ } (39)

𝑖∈𝐺∪𝑆

∑ 𝑝𝑜 𝑥𝑜𝑟 ≤ 𝑇𝑀 ∀𝑟 ∈ {1, … , 𝑅 ′ } (40)

𝑜∈𝑂

|∑ 𝑝𝑜 𝑥𝑜𝑟 − 𝐴𝑊| ≤ 𝑢𝑟 ∀𝑟 ∈ {1, … , 𝑅 ′ } (41)

𝑜∈𝑂

Constraints (21) and (22) in Section 3.1.1 should also be added to show the domains and ranges
of variables.

4. Results
In this section, the results of the proposed models are reported and compared. The models for both
approaches are solved using a commercial optimization solver, Gurobi 9.5.2 (Copyright © 2022,
Gurobi Optimization, LLC). The models were executed on a desktop computer with Intel ® Core
(TM) i7-11700 @ 2.50 GHz processor and 16.0 GB RAM.

23
4.1. Results of the simultaneous and sequential modelling approaches
Table 5 presents the results from both the simultaneous and sequential modelling approaches,
comparing them across various scenarios, while each scenario is based on a specific number of
orders received at the warehouse. In all scenarios, 12 pickers are considered, and the available
picking time is an eight-hour shift, as detailed in the case study section.

It should be noted that in the execution and coding of the simultaneous modelling approach,
decision variables 𝑦𝑜𝑖𝑗𝑟 , which determines the routing of pickers for orders, are only defined for
pairs of items 𝑖 and 𝑗 that exist in order 𝑜. This way, the number of required binary variables is
reduced and the execution of the model is faster.

To show the importance of workload equity and to compare the order picking plans with and
without workload equity considerations, each scenario is solved two times. SM1, SM2, and SM3
refer to the simulations modeling approach where only maximization of the picked orders is
considered, which means that the objective function of the model is 𝑀𝑎𝑥. ∑𝑜∈𝑂 ∑𝑟∈𝑅 𝑥𝑜 𝑟 , and
SM1*, SM2*, and SM3* refer to scenarios where workload equity is also considered, which means
that the objective function of the model is 𝑀𝑎𝑥. (𝑊1 ∑𝑜∈𝑂 ∑𝑟∈𝑅 𝑥𝑜 𝑟 − 𝑊2 ∑𝑟∈𝑅 𝑢𝑟 ). Similarly,
SQ1, SQ2, and SQ3 refer to the sequential modelling approaching approach where only
maximization of the picked orders is considered, and SQ1*, SQ2*, and SQ3* refer to scenarios
where workload equity is also considered.

It should be noted that 𝑊1 and 𝑊2 are considered 0.9 and 0.1, respectively, because maximizing
the number of picked (i.e., fulfilled) orders was the priority for the warehouse managers. Moreover,
the average workload (𝐴𝑊) used in constraint set (20) of the model is calculated by dividing the
total working time of all pickers obtained from solving the model without workload equity
considerations by the number of pickers. The two terms of the objective function are also
normalized to ensure they have the same range and effect in the objective function.

The simultaneous modelling approach requires a huge memory space, which led to an out-of-
memory error when 700 orders were considered as input. This shows the limitation of the
simultaneous modelling approach in terms of the number of received orders that could be
considered.

24
 Table 5. Comparison between the results of the simultaneous and sequential modelling approaches.
Model results
Total
Number of Number of
Number of workload
Case unique items Solution picked
Case description received deviations
scenarios in all the time (s) orders
orders (min)
orders
(∑ ∑ 𝑥𝑜𝑟 )
𝑜∈𝑂 𝑟∈𝑅 (∑ 𝑢𝑟 )
𝑟∈𝑅

Without workload equity considerations

SM1 470 298 799.47 a
(simultaneous modelling approach)
With workload equity considerations
SM1* 1,637 298 1.64
(simultaneous modelling approach)
300 orders 183
Without workload equity considerations
SQ2 480 300 2589.23 a
(sequential modelling approach)
With workload equity considerations
SQ2* 480 299 21.27
(sequential modelling approach)
Without workload equity considerations
SM2 1,506 496 327.47 a
(simultaneous modelling approach)
With workload equity considerations
SM2* 1,947 495 0.62
(simultaneous modelling approach)
500 orders 190
Without workload equity considerations
SQ2 720 500 2423.11 a
(sequential modelling approach)
With workload equity considerations
SQ2* 720 499 11.06
(sequential modelling approach)
Without workload equity considerations
SM3 4,004 668 14.80 a
(simultaneous modelling approach)
With workload equity considerations
SM3* Out of memory
(simultaneous modelling approach)
700 orders 193
Without workload equity considerations
SQ3 1020 700 530.56 a
(sequential modelling approach)
With workload equity considerations
SQ3* 1,020 699 21.10
(sequential modelling approach)
Table 5 footnote: SM1, SM2, and SM3 are case scenarios without workload balancing consideration under the
simultaneous modelling approach. SM1*, SM2*, and SM3* are case scenarios with workload balancing consideration
under the simultaneous modelling approach. SQ1, SQ2, and SQ3 are case scenarios without workload balancing
consideration under the sequential modelling approach. SQ1*, SQ2*, and SQ3* are case scenarios with workload
balancing consideration under the sequential modelling approach.
a
This variable was not defined in the related mathematical model, and its value is calculated manually by determining
the average workload in the respective case and calculating the total workload deviation from the average workload
across all pickers.

As it is seen for all three cases, the sequential modelling approach could result in a higher volume
of picked orders within a shift, and has a much shorter solution time. It should be noted that the
difference between the number of picked orders in the two approaches mainly stemmed from the
optimization gap, not the available working time for pickers. Additionally, it is evident that the

25
total workload deviation from the average workload for all the pickers is much larger when
workload equity is not considered. However, this difference becomes less noticeable as the number
of orders increases. This is because with larger order volumes, pickers are working almost full-
time, causing the difference between their workloads to decrease.

Figure 4 shows the working time of each picker for each case scenario listed in Table 5. Solid red
bars, which represent the results obtained from the sequential modelling approach considering
workload equity, show less working time for pickers in all three cases while more orders are
picked. The issue of workload inequality becomes most apparent in scenarios where sequential
modeling approach is employed without workload equity considerations.

(a) SM1, SM1*, SQ1, SQ1* (300 orders) (b) SM2, SM2*, SQ2, SQ2* (500 orders)
8 8
Working time of pickers (hrs)
Working time of pickers (hrs)

6 6

4 4

2 2

0 0

without equity-simultaneous modelling approach (SM1) without equity-simultaneous modelling approach (SM2)
with equity-simultaneous modelling approach (SM1*) with equity-simultaneous modelling approach (SM2*)
without equity-sequential modelling approach (SQ1) without equity-sequential modelling approach (SQ2)
with equity-sequential modelling approach (SQ1*) with equity-sequential modelling approach (SQ2*)

(c) SM3, SM3*, SQ3, SQ3* (700 orders)

8
Working time of pickers (hrs)

without equity-simultaneous modelling approach (SM3)

with equity-simultaneous modelling approach (SM3*)
without equity-sequential modelling approach (SQ3)
with equity-sequential modelling approach (SQ3*)

Figure 4 (a-c). Comparison between the total working hours of pickers obtained from the simultaneous and
sequential approaches.
Figure 5 illustrates the utilized man-hours as a proportion of the total available man-hours
(calculated by multiplying the number of pickers by the hours in a working shift) with the aim of

26
maximizing the fulfilled orders based on the scenarios outlined in Table 5. It is shown that in both
modelling approaches, the utilization of man-hours remains relatively consistent with and without
workload balancing considerations. This consistency was anticipated, given that the number of
fulfilled orders remains approximately the same with or without workload balancing
considerations.

(a) SM1, SM1*, SQ1, SQ1* (300 orders) (b) SM2, SM2*, SQ2, SQ2* (500 orders)
100 100
man-hour (hours)

man-hour (hours)
80 80
60 60
40 40
20 20
0 0
modelling approach

modelling approach

with equity-sequential
sequential modelling

modelling approach

modelling approach

with equity-sequential
sequential modelling
modelling approach

modelling approach
without equity-

without equity-
approach (SQ1)
without equity-

approach (SQ2)
without equity-
simultaneous

simultaneous

simultaneous

simultaneous
with equity-

with equity-
(SM1*)

(SM2*)
(SM1)

(SM2)
(SQ1*)

(SQ2*)
(c) SM3, SM3*, SQ3, SQ3* (700 orders)
100
Out of memory
man-hour (hours)

80
60
40
20
0
modelling approach

modelling approach

with equity-sequential
sequential modelling

modelling approach
without equity-

approach (SQ3)
without equity-
simultaneous

simultaneous
with equity-

(SM3*)
(SM3)

(SQ3*)

Figure 5 (a-c). Comparison between the utilized man-hour to fulfill orders and the available man-hour at the
warehouse.
Moreover, both Figure 4 and Figure 5 demonstrate that under the sequential modelling approach,
fewer man-hours are utilized compared to that in the simultaneous modelling approach. The reason
is that in instances where the number of received orders is considerably below the warehouse
capacity, the simultaneous modelling approach, which focused on maximizing the fulfilled orders,
may not necessarily prioritize picking routes with minimum picking time. Conversely, the
sequential modelling approach minimizes picking times for each order before maximizing the
number of fulfilled orders, resulting in a reduced need for man-hours. However, this distinction
becomes less noticeable when a higher number of orders are received (specifically, 700 orders and

27
above). In such cases, to fulfill more orders, the simultaneous modelling approach must also seek
picking routes with minimized picking time, diminishing the disparity between the results of the
two models.

4.2. Number of picked orders for different case scenarios

The model for the sequential modelling approach is executed for cases where 300, 500, 700, 900,
1100, and 2000 orders are available to be picked. This analysis was carried out using the sequential
modelling approach as this approach proved to be faster and more efficient than the simultaneous
modelling approach, and could yield results for 700 orders and above.

Figure 6 shows that with the current resources at the warehouse (i.e., pickers and picking
equipment pieces), 99.7% of 300 orders, 99.8% of 500 orders, and 99.9% of 700 orders that are
received at the warehouse can be picked. However, the percentage of fulfilled orders decreases to
96.9% 92.5%, and 75.3% when the number of orders placed by customers increases to 900, 1100,
and 2000 orders, respectively. In the case of 2000 orders, four pickers are working for 7.9 hours,
while the remaining eight pickers are working full time. The sequential modelling approach took
approximately 40 minutes to achieve this result.

1100 100 %

1000 95 %
900 90 %
Percentage of picked orders

800
85 %
700
Picked orders

80 %
600
75 %
500
70 %
400
65 %
300

200 60 %

100 55 %

0 50 %
300 500 700 900 1100 2000
Received orders
Picked orders % of picke orders

Figure 6. Quantity and the percentage of picked orders considering different number of received orders at the
warehouse.

28
6.3 Impact of number of pickers on fulfilled orders
As expected, to pick more orders, more pickers are required. Figure 7 shows the increase in the
number of picked orders when one picker is added to the current 12 pickers for a case where 1100
orders are received at the warehouse. The simultaneous modelling approach could not yield any
results due to memory shortage for this volume of orders. Therefore, this analysis was performed
using only the sequential modelling approach. According to the results, on average, 2%
improvement in the number of picked orders can be obtained by adding one picker.

9% 1100

Number of picked orders

Percentage of increase in
number of picked orders

8%
7% 1050
6%
5%
1000
4%
3%
2% 950

1%
0% 900
0 1 2 3 4 5
Added pickers
Orders % of increased picked orders by increasing a picker

Figure 7. Number of picked orders and percentage of improvement in the number of picked orders by adding
pickers to the current number of pickers (12) working at the warehouse.

5. Discussion and managerial insights

In the majority of warehouse picking problems in the literature, the picking plan is determined by
minimizing the picking time, distance, or costs, considering a certain number of orders to be picked
[6, 19, 20]. In this way, the model would not be able to provide any insights on how to decide on
the volume of orders that the warehouse can handle (i.e., fulfill). However, in this study, which is
inspired by a real warehouse picking problem, instead of considering a fixed number of orders to
be picked, we considered this factor as a decision variable and provided a model that maximizes
the number of picked orders based on the available customer orders (i.e., demand). Maximizing
the number of picked orders would indirectly affect the picking time as well. The reason is that to
pick more orders, picking routes and plans with shorter picking times should be determined.
However, when the number of received orders at the warehouse is considerably lower than the
actual order handling capacity of the warehouse, the model would not choose the picking routes
with the shortest picking time (i.e., optimized tours) necessarily by maximizing the number of

29
picked orders (refer to Figure 4 and 5). The reason is that all the received orders can be easily
handled during the available time by the available pickers even without choosing the shortest
picking tours. Therefore, only when the number of received orders is high, maximizing the number
of picked orders can result in forming picking tours that are shorter and optimized in this sense.
Because of this, we also provided another approach (i.e., the sequential modelling approach) for
the introduced problem that maximizes the number of picked orders while determining the
optimized picking tours (in terms of picking time). In this approach, the shortest picking tours for
orders are determined first, and then, orders are assigned to pickers to maximize the number of
picked orders. By utilizing this approach, even during periods of low order volume (i.e., customer
demand), the sequence of items that pickers need to pick (i.e., pick tours) is determined in a way
that pickers’ working time is minimized. This way, the pickers' time and energy can be saved,
allowing them to focus on other tasks in the warehouse. In general, the comparison between the
two approaches shows the superiority of the sequential modelling approach. By breaking down the
main problem into two stages, the complexity of the model is reduced, resulting in better outcomes
in a shorter solution time.

From a practical standpoint, the proposed mathematical models for both approaches can assist
warehouse managers in setting realistic expectations for the maximum number of orders that the
warehouse can process within a working shift. This ultimately leads to increased customer
satisfaction by ensuring timely deliveries and preventing over-promising. Additionally, by having
information on the maximum order-handling capacity of the warehouse, warehouse managers can
generate more revenue by avoiding the underutilization of resources.

To fulfill orders above the warehouse’s maximum order-handling capacity, it is necessary to have
additional picking staff. By increasing the number of pickers in the proposed model, information
regarding the potential improvement in the fulfillment of orders can be provided to warehouse
managers. This information enables the warehouse managers to conduct further analyses to
determine if the benefits from a higher number of fulfilled orders would offset the costs of hiring
new staff and acquiring additional picking equipment.

Another derived insight is that workload equity considerations make more sense for situations
where a lower number of orders are received at the warehouse. As reported in Figures 4 and 5, a
higher volume of orders leads to a less noticeable difference between the working time of pickers

30
with and without workload equity considerations. The reason is that all the pickers are working for
almost the entire available time. However, when the number of available orders is lower, workload
equity consideration is important, and it helps to keep pickers’ schedules similar and fair.
Therefore, the proposed model can help managers determine the maximum number of orders the
warehouse can handle and, if the actual orders received from customers are close to this number,
they do not need to worry about balancing the workload.

6. Conclusions
This research was inspired by the real case study of a warehouse of a consumer goods distribution
company located in North America, which operates on a picker-to-part picking system and a strict
picking policy. Two mathematical programming approaches were developed to maximize the
number of picked orders during a shift in a warehouse, while taking workload equity among the
pickers into account. The models were designed to determine the order picking plans within a
working shift by making decisions on the assignment of orders to the available pickers, the routes
or paths that should be taken to pick each order, and the working time of each picker.

The difference between the two approaches is that the simultaneous modelling approach
maximizes the number of picked orders through which the picking tours are determined, while the
sequential modelling approach first forms the optimal picking tours in terms of picking time and
then maximizes the number of orders with specific picking tours that can be assigned to pickers.

Both mathematical models were tested on different scenarios from the case study. The sequential
modelling approach outperformed the simultaneous modelling approach in terms of the solution
time, the volume of picked orders, and the working time of pickers. The performed analyses
demonstrated that the order handling capacity of the warehouse decreased when more than 700
orders were received. Therefore, the warehouse should increase the number of pickers if it wishes
to accept and fulfill more than 700 orders per shift. An approximate 2% improvement in the
number of fulfilled orders can be achieved by adding a picker.

In conclusion, the main benefit of the proposed approaches to optimize the warehouse picking is
to increase the number of orders that can be picked with the same warehouse staff and assign
similar workloads to pickers. This will allow the company to grow its order volume more quickly
without expanding its workforce while keeping pickers’ workloads fair and similar.

31
The warehouse optimization also contributes to reducing the tribal knowledge needed to generate
picklists and pick orders, as in the current situation in the considered case study, the warehouse
staff must manually generate a picklist based on their own knowledge of the warehouse and decide
how to best go about picking. The warehouse picking optimization makes this process automated
and based on the latest available data, which eventually will provide a circumstance under which
the warehouse staff can be more productive.

This study took three practical factors into account in the order picking planning, resulting in a
more practical model. However, there were other practical factors that could not be included due
to a lack of proper data, which limits the effectiveness of the proposed model. For example, this
study treated the searching time and retrieval time as equal for all pickers and items, while
searching time of pickers is affected by pickers’ experience and spatial memory. The retrieval time
for different items can also be affected by the size and weight of items. Accounting for these and
other practical factors [9], such as pickers skills, picker fatigue, safety constraints, or pickers’
learning effect (e.g., [43, 44]), can be interesting directions for future research.

The proposed models can efficiently address the warehouse's needs within a reasonable timeframe.
For instance, in a scenario involving 2000 received orders, 1506 orders were fulfilled with nearly
all pickers operating full time. This solution was achieved within approximately 40 minutes using
the sequential modelling approach. While the proposed models proved suitable for even larger
problem sizes than those in this study, future research could explore heuristic or metaheuristic
algorithms to expedite solutions for large size problems.

Acknowledgment
The authors would like to thank the Natural Sciences and Engineering Research Council of Canada
(NSERC RGPIN-2019-04563); and Mitacs, Fujitsu Intelligence Technology Limited and
Quantum Algorithms Institute (MITACS IT33313) for their financial support in conducting this
research. The authors also express their gratitude to Fujitsu Intelligence Technology Limited and
their client for providing the required data. For confidentiality purposes, the identity of the client
(warehouse) remains undisclosed.

32
Disclosure Statement

No potential conflict of interest was reported by the authors.

Declaration of generative AI
During the preparation of this work, generative artificial intelligence (AI) and AI-assisted
technologies were not used.

Data Availability Statement

The authors confirm that the majority of data supporting the findings of this study, including
records of number of daily orders received at the warehouse, type and weight of the picking
equipment, and number of pickers are available within the article. However, data regarding the
name, specific location, and layout of the warehouse, as well as their product names and
characteristics cannot be shared as it is against the internal policies of the warehouse company.

References

[1] M. R. Islam, S. M. Ali, A. M. Fathollahi-Fard, G. Kabir, A novel particle swarm

optimization-based grey model for the prediction of warehouse performance, J. Comput. Des.
Eng. 8(2) (2021), 705-727. https://doi.org/10.1093/jcde/qwab009.

[2] R. de Koster, T. Le-Duc, K. J. Roodbergen, Design and control of warehouse order picking:
A literature review, Eur. J. Oper. Res. 182(2) (2007) 481-501.
https://doi.org/10.1016/j.ejor.2006.07.009.

[3] M. Masae, C. H. Glock, E. H. Grosse, Order picker routing in warehouses: A systematic

literature review, Int. J. Prod. Econ. 224 (2020) 107564.
https://doi.org/10.1016/j.ijpe.2019.107564.

[4] A. R. F. Pinto, M. S. Nagano, A comprehensive review of batching problems in low-level

picker-to-parts systems with order due dates: Main gaps, trade-offs, and prospects for future
research, J. Manuf. Syst. 65 (2022) 1-18. https://doi.org/10.1016/j.jmsy.2022.08.006.

[5] Ç. Cergibozan, A. S. Tasan, Order batching operations: an overview of classification,

solution techniques, and future research, J. Intell. Manuf. 30(1) (2019) 335-349.
https://doi.org/10.1007/s10845-016-1248-4.

[6] M. Farhadi Sartangi, A. Husseinzadeh Kashan, H. Haleh, A. Kazemi, A Mixed Integer Linear
Formulation and a Grouping League Championship Algorithm for a Multiperiod-Multitrip Order
33
Picking System with Product Replenishment to Minimize Total Tardiness, Complexity (2022) 1-
24. https://doi.org/10.1155/2022/1382558.

[7] M. Yu, Enhancing warehouse performance by efficient order picking (2008).

https://repub.eur.nl/pub/13691/EPS2008139LIS9058921673YU.pdf (accessed 11 July 2023)

[8] F. Yener, H. R. Yazgan, Optimal warehouse design: Literature review and case study
application, Comput. Ind. Eng. 129 (2019) 1-13. https://doi.org/10.1016/j.cie.2019.01.006.

[9] S. Vanheusden, T. van Gils, K. Ramaekers, T. Cornelissens, A. Caris, Practical factors in

order picking planning: state-of-the-art classification and review, Int. J. Prod. Res. (2022) 1-25.
https://doi.org/10.1080/00207543.2022.2053223.

[10] S. Vanheusden, T. van Gils, A. Caris, K. Ramaekers, K. Braekers, Operational workload

balancing in manual order picking, Comput. Ind. Eng. 141 (2020) 106269.
https://doi.org/10.1016/j.cie.2020.106269.

[11] S. Vanheusden, T. van Gils, K. Braekers, K. Ramaekers, A. Caris, Analysing the

effectiveness of workload balancing measures in order picking operations, Int. J. Prod. Res.
60(7) (2022) 2126-2150. https://doi.org/10.1080/00207543.2021.1884307.

[12] T. van Gils, A. Caris, K. Ramaekers, K. Braekers, R. B. M. de Koster, Designing efficient

order picking systems: The effect of real-life features on the relationship among planning
problems, Transp. Res. Part E: Logist. Transp. Rev. 125 (2019) 47-73.
https://doi.org/10.1016/j.tre.2019.02.010.

[13] Ö. Öztürkoğlu, D. Hoşer, An Evaluation of Order-picking Tour Efficiency in Two-block

Warehouses, OSCM (2019) 74-87. https://doi.org/10.31387/oscm0370225.

[14] J. J. Bartholdi, S. T. Hackman, Warehouse & Distribution Science, The Supply Chain and
Logistics Institute (2019). https://www.warehouse-science.com/book/editions/wh-sci-0.98.1.pdf
(accessed 17 June 2024)

[15] N. Shetty, B. Sah, S. H. Chung, Route optimization for warehouse order picking operations
via vehicle routing and simulation, SN Appl. Sci. 2(2) (2020) 311.
https://doi.org/10.1007/s42452-020-2076-x.

[16] M. Haouassi, Y. Kergosien, J. E. Mendoza, L.-M. Rousseau, The integrated orderline

batching, batch scheduling, and picker routing problem with multiple pickers: the benefits of
splitting customer orders, Flex Serv Manuf J. 34(3) (2022) 614-645.
https://doi.org/10.1007/s10696-021-09425-8.

[17] A. Scholz, D. Schubert, G. Wäscher, Order picking with multiple pickers and due dates –
Simultaneous solution of Order Batching, Batch Assignment and Sequencing, and Picker
Routing Problems, Eur. J. Oper. Res. 263(2) (2017) 461-478.
https://doi.org/10.1016/j.ejor.2017.04.038.

34
[18] T. van Gils, A. Caris, K. Ramaekers, K. Braekers, Formulating and solving the integrated
batching, routing, and picker scheduling problem in a real-life spare parts warehouse, Eur. J.
Oper. Res. 277(3) (2019) 814-830. https://doi.org/10.1016/j.ejor.2019.03.012.

[19] E. Ardjmand, H. Shakeri, M. Singh, O. Sanei Bajgiran, Minimizing order picking makespan
with multiple pickers in a wave picking warehouse, Int. J. Prod. Econ. 206 (2018) 169-183.
https://doi.org/10.1016/j.ijpe.2018.10.001.

[20] S. A. B. Rasmi, Y. Wang, H. Charkhgard, Wave order picking under the mixed-shelves
storage strategy: A solution method and advantages, Comput. Oper. Res. 137 (2022) 105556.
https://doi.org/10.1016/j.cor.2021.105556.

[21] S. Saylam, M. Çelik, H. Süral, The min–max order picking problem in synchronised
dynamic zone-picking systems, Int. J. Prod. Res. (2022) 1-19.
https://doi.org/10.1080/00207543.2022.2058433.

[22] J. C.-H. Pan, P.-H. Shih, M.-H. Wu, J.-H. Lin, A storage assignment heuristic method based
on genetic algorithm for a pick-and-pass warehousing system, Comput. Ind. Eng. 81 (2015) 1-13.
https://doi.org/10.1016/j.cie.2014.12.010.

[23] X. Wang, J. Zhang, H. Shang, A real-time Synchronized Zone order picking system for
balancing picker’s workload, ICIC express letters. 4(5) (2013) 8.

[24] C. G. Petersen, G. Aase, A comparison of picking, storage, and routing policies in manual
order picking, Int. J. Prod. Econ. 92(1) (2004) 11-19. https://doi.org/10.1016/j.ijpe.2003.09.006.

[25] R. Davidson, What is Order Picking? And What Are The Best Methods?, SoftwareConnect.
https://softwareconnect.com/warehouse-management/order-picking-methods/ (accessed 06 July
2023)

[26] A. Atashi Khoei, H. Süral, and M. K. Tural, Energy minimizing order picker forklift routing
problem, Eur. J. Oper. Res. 307(2) (2022) 604-626. https://doi.org/10.1016/j.ejor.2022.08.038.

[27] F. Weidinger, Picker routing in rectangular mixed shelves warehouses, Comput. Oper. Res.
95 (2018) 139-150. https://doi.org/10.1016/j.cor.2018.03.012.

[28] F. Weidinger, N. Boysen, and M. Schneider, Picker routing in the mixed-shelves

warehouses of e-commerce retailers, Eur. J. Oper. Res. 274(2) (2019) 501-515.
https://doi.org/10.1016/j.ejor.2018.10.021.

[29] X. Xu and C. Ren, A novel storage location assignment in multi-pickers picker-to-parts

systems integrating scattered storage, demand correlation, and routing adjustment, Comput. Ind.
Eng. 172(2022) 108618. https://doi.org/10.1016/j.cie.2022.108618.

[30] J. A. Cano, Formulations for joint order picking problems in low-level picker-to-part
systems, Bull. EEI. 9(2) (2020) 834-842. https://doi.org/10.11591/eei.v9i2.2110.

35
[31] J. A. Cano, A. A. Correa-Espinal, and R. A. Gómez-Montoya, Mathematical programming
modeling for joint order batching, sequencing and picker routing problems in manual order
picking systems, J. King Saud Univ. Eng. Sci. 32(3) (2020) 219-228.
https://doi.org/10.1016/j.jksues.2019.02.004.

[32] S. Henn, Order batching and sequencing for the minimization of the total tardiness in
picker-to-part warehouses, Flex Serv Manuf J. 27(1) (2015) 86-114.
https://doi.org/10.1007/s10696-012-9164-1.

[33] B. Aerts, T. Cornelissens, and K. Sörensen, The joint order batching and picker routing
problem: Modelled and solved as a clustered vehicle routing problem, Comput. Oper. Res.
129(2021) 105168. https://doi.org/10.1016/j.cor.2020.105168.

[34] M. Yousefi Nejad Attari, A. Ebadi Torkayesh, B. Malmir, and E. Neyshabouri Jami, Robust
possibilistic programming for joint order batching and picker routing problem in warehouse
management, Int. J. Prod. Res. 59(14) (2021) 4434-4452.
https://doi.org/10.1080/00207543.2020.1766712.

[35] T.-L. Chen, C.-Y. Cheng, Y.-Y. Chen, and L.-K. Chan, An efficient hybrid algorithm for
integrated order batching, sequencing and routing problem, Int. J. Prod. Econ. 159(2015) 158-
167. https://doi.org/10.1016/j.ijpe.2014.09.029.

[36] P. Yang, Z. Zhao, and H. Guo, Order batch picking optimization under different storage
scenarios for e-commerce warehouses, Transp. Res. Part E: Logist. Transp. Rev. 136 (2020)
101897. https://doi.org/10.1016/j.tre.2020.101897.

[37] J. Liang, Z. Wu, C. Zhu, and Z.-H. Zhang, An estimation distribution algorithm for wave-
picking warehouse management, J. Intell. Manuf. 33(4) (2022) 929-942.
https://doi.org/10.1007/s10845-020-01688-6.

[38] T. Chabot, L. C. Coelho, J. Renaud, and J.-F. Côté, Mathematical model, heuristics and
exact method for order picking in narrow aisles, J. Oper. Res. Soc. 69(8) (2018) 1242-1253.
https://doi.org/10.1080/01605682.2017.1390532.

[39] F. Chen, H. Wang, Y. Xie, and C. Qi, An ACO-based online routing method for multiple
order pickers with congestion consideration in warehouse, J. Intell. Manuf. 27(2) (2016) 389-
408. https://doi.org/10.1007/s10845-014-0871-1.

[40] A. Silva, L. C. Coelho, M. Darvish, and J. Renaud, Integrating storage location and order
picking problems in warehouse planning, Transp. Res. Part E: Logist. Transp. Rev. 140 (2020)
102003. https://doi.org/10.1016/j.tre.2020.102003.

[41] S. Ghotb, T. Sowlati, J. Mortyn, D. Roeser, and V. C. Griess, A goal programming model
for the optimization of log logistics considering sorting decisions and social objective, Can. J.
For. Res. 52(5) (2022) 716-726. https://doi.org/10.1139/cjfr-2021-0203.

36
[42] C. E. Miller, A. W. Tucker, and R. A. Zemlin, Integer Programming Formulation of
Traveling Salesman Problems, J. ACM. 7(4) (1960) 326-329.
https://doi.org/10.1145/321043.321046.

[43] J. Zhang, X. Zhang, and N. Zhang, Considering pickers’ learning effects in selecting
between batch picking and batch-synchronized zone picking for online-to-offline groceries,
Appl. Math. Model. 113 (2023) 358-375. https://doi.org/10.1016/j.apm.2022.09.009.

[44] E. H. Grosse and C. H. Glock, An experimental investigation of learning effects in order

picking systems, J. Manuf. Technol. Manage. 24(6) (2013) 850-872.
https://doi.org/10.1108/JMTM-03-2012-0036.

Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.

☐ The author is an Editorial Board Member/Editor-in-Chief/Associate Editor/Guest Editor for

[Journal name] and was not involved in the editorial review or the decision to publish this article.

☐ The authors declare the following financial interests/personal relationships which may be considered
as potential competing interests:

37