A case study on human VRP solving practices in a parcel delivery setting

(1)

A case study on human VRP solving

practices in a parcel delivery setting

MSc thesis Supply Chain Management

University of Groningen

Faculty of Economics and Business

Author: Rudy Niemeijer (S2751453) Supervisor: dr. ir. P. Buijs

Co-assessor: dr. ir. S. Fazi

Wordcount: 8027

(2)

Abstract

Research on Vehicle Routing Problem (VRP) algorithms is widely covered in current literature. Although VRP algorithms are increasingly useful in finding good solutions for VRPs, they are not widely used in practice. There are two perspectives in the literature on the underlying cause of why this is not the case. The first is that VRP algorithms do not capture the complexity of reality well enough. The second is that humans are inherently averse to algorithms. This thesis proposes a third perspective, namely that there are substantial differences between algorithms and human problem solving practices. The aim of this thesis is to explore what can be learned from human VRP solving practices to advance VRP algorithm research and implementation.

(3)

1 Introduction

Parcel delivery companies are always in search of more efficient ways to let their couriers deliver parcels. These companies recognise the importance of route optimi-sation to provide shorter and less time consuming routes for their couriers. A topic widely covered by literature on route optimisation is the Vehicle Routing Problem (VRP). The VRP is concerned with the optimum routing of a fleet of delivery vehi-cles between a depot and a large number of customers (Dantzig and Ramser, 1959). A human route planner is able to provide a solution for a simple VRP with a small fleet and a limited number of customers. When the VRP becomes more complicated, a VRP algorithm should be of help to find a good solution.

It is widely assumed that these algorithms are increasingly useful in finding an efficient solution. This is reflected by the fact that numerous papers in the field of Operations Research use algorithms as a basis to improve route optimisation for parcel couriers (Abu Al Hla et al., 2019; He and Yang, 2018; Kim, 2019; Masson et al., 2017; Simoni et al., 2018; Xu et al., 2011). However, these developments are not translated into the practical application of algorithms. In parcel delivery companies, algorithms are not widely used (Gayialis et al., 2019; McCrea, 2017; Salter, 2014). Currently, there are two perspectives on precisely why VRP algorithms are seldom used in practice.

(6)

A second perspective is that that humans are limited in their level of technology acceptance. Research involving algorithm solutions have demonstrated that humans tend to favour human generated solutions over algorithm based solutions, even when the algorithm outperforms the human (Dietvorst et al., 2015). This phenomenon is termed algorithm aversion and has been demonstrated with data forecasting (Di-etvorst et al., 2015) and medical diagnosis (Porat et al., 2017). It is argued that this phenomenon prevents algorithms from being extensively used. The underlying factors that influence this phenomenon is not fully researched in a VRP context.

Both perspectives bring a plausible explanation for the limited use of VRP algo-rithms in practice. However, the focus lies primarily on the algorithm itself. Imple-menting algorithms in practice is bound to involve interaction with humans and their respective approach towards solving routing problems. In both perspectives, human problem solving practices are not involved. Understanding these practices may help expose differences between algorithms and human processes. Hence, in this thesis, a third perspective is proposed, namely that there are substantial differences between algorithms and the human VRP solution strategies. It is argued that these differences prevent the extensive use of algorithms in practice.

Although there are several studies about human solution strategies on the travel-ling salesman problem (TSP) (MacGregor and Chu, 2011), which is a simpler form of the VRP, there are a limited number of papers about human performance on VRPs (Fontaine et al., 2020). Moreover, thus far, these studies are only conducted in an experimental setting. Virtually no research has been published on understanding how humans actually go about real-life VRPs, i.e. in an uncontrolled complex environment (Tan and Staats, 2020). An abstract experimental VRP setting may not resemble the complexity of reality well enough in order to fully understand the human approach for real-life VRPs. Hence, the following research question is raised.

What can be learned from human VRP solving practices in order to advance VRP algorithm research and implementation?

(7)

(8)

2 Theoretical framework

The theoretical framework is subdivided into three parts. First, algorithms and rent developments in the field of algorithm improvement are discussed. Second, cur-rent knowledge on human VRP solution strategies is considered. Thereafter, external factors are discussed that might influence human VRP solving practices.

2.1 VRP algorithms

The origins of the VRP can be found in the well-known Travelling Salesman Problem (TSP). This problem describes a set of locations that should be visited once and only once by the salesman and return to his starting position (Flood, 1956). The objective of solving the TSP is to minimise the total distance travelled. The VRP is a generalisation of the TSP (Dantzig and Ramser, 1959). In essence, the VRP extends the TSP by adding several salesmen routes and constraints. Usually, the minimisation of the objective is based on the distances between customers, which are called the costs. Alternatively, the costs can also be defined in other terms such as travel time, pollution, fuel consumption or a combination of these elements (Pahlavani and Delavar, 2014; Srivatsa Srinivas and Gajanand, 2017).

2.1.1 Current developments

There are many articles to be found in the literature on algorithms to find solutions for VRPs. These algorithms involve many different VRP variants and the variety is ever increasing (Gayialis et al., 2019). In the review of Braekers et al. (2016) and a follow up by Elshaer and Awad (2020), trends are identified in the type of characteristics prevalent in VRP algorithm research.

(9)

used (Braekers et al., 2016). Only a few ’general’ algorithms are proposed that would be applicable in multiple variants of scenarios (Braekers et al., 2016; Gayialis et al., 2019).

Figure 2.1: Relative presence of VRP variants in literature. Adapted from Elshaer and Awad (2020).

(10)

2.1.2 Limitations of rich VRPs

Since Dantzig and Ramser (1959) have proposed the VRP, many algorithms have been constructed and improved to solve it (Ando and Taniguchi, 2006; Hofmann et al., 2017; Masson et al., 2017; Xu et al., 2011). As mentioned before, there is a huge variety of specific VRPs and algorithms to solve them. Some examples of current algorithm improvements are involving more complex elements such as variable delivery windows (Kim, 2019), multi-tour optimisation (Martins-Turner and Nagel, 2019), alternative Urban Consolidation Center configuration (Simoni et al., 2018), bus-road mixed networks (He and Yang, 2018) and driver behaviour (Abu Al Hla et al., 2019).

(11)

2.2 Human VRP solution strategies

There are a multitude of experimental studies on human performance on TSP solving (MacGregor and Chu, 2011). In general, it is known that humans perform well in a simple TSP setting, i.e. with a small number of nodes (customers) (MacGregor and Chu, 2011). Furthermore, a human is able to find better routes than simple construction algorithms, even when the participant had no particular knowledge of the TSP and were simply asked to sketch a route covering 10 or 20 nodes (MacGregor and Chu, 2011). As opposed to TSP literature, less experimental studies exist on human performance on VRP solving. The study of Fontaine et al. (2020) is one of few to do this. In general, they found that humans rarely find the optimal solution. More specifically, humans perform better than the worst algorithms and worse than the best algorithms. In other words, humans are generally average performing compared to algorithms. Furthermore, individuals tend to switch strategies regularly. They found that humans show a rather unpredictable sequence of strategies. Examples of these strategies are nearest neighbour heuristic, savings heuristic or a sweep strategy (Fontaine et al. (2020)).

Furthermore, Fontaine et al. (2020) provides a more tangible advice in terms of the VRP solving process: planners (or algorithms) should make clusters. Then, given the clusters, couriers can make their own routing decisions. Nevertheless, humans do not perform well in making clusters, but are just as good as algorithms in making local solutions. Sometimes ’quick and dirty’ visual solving approaches can even outperform more cognitively strenuous approaches such as clustering (Kefalidou and Ormerod, 2014). Kefalidou and Ormerod (2014) speculate that this might have to do with the environment in which the task is performed.

(12)

Considering the ecological validity of the aforementioned experiments, there is a vast difference in problem size, number of variables and constraints between the ex-periments and real-life VRPs. For instance, the exex-periments Fontaine et al. (2020) contained at most 25 nodes and a capacitated fleet of at most 8 trucks. The experi-ments of Kefalidou and Ormerod (2014) contained at most 45 nodes and a capacitated fleet of at most 7 trucks. An interview with a courier of one of the largest parcel de-livery companies in the Netherlands revealed that he delivers between 120 and 180 parcels daily (ter Haar, 2018). To emphasize: this is the number of addresses (i.e. nodes) only for one courier. One can imagine how much more complex real-life VRPs might be. Moreover, the real-life constraints that are present in parcel delivery prac-tices add another level of complexity. Assuming that humans are more adaptable to the miscellaneous constraints of reality, questions might be raised to what extent humans would compare to algorithms in practice. Furthermore, the environment in which the VRPs are solved is very different in practice. It is not known how external factors may influence human VRP solving practices.

2.3 External factors

As opposed to core human VRP solution strategies, several external factors can be identified that may affect the practical approach to VRPs. It is plausible, however, that also other factors affect the human VRP solving process and its relation the VRP algorithms. Factors that are not directly emerging from the literature on VRP characteristics and algorithms.

2.3.1 Autonomy, complexity and job satisfaction

(13)

following way.

In job dimensions, autonomy is defined as “the degree to which the job provides substantial freedom, independence, and discretion to the employee in scheduling the work and in determining the procedures to be used in carrying it out.” (Hackman and Oldham, 1975). In parcel delivery practices, this would translate to the level of freedom couriers receive to make routing decisions. When couriers have little auton-omy, most aspects of the VRP solution would be in the hands of upper management. Alternatively, when couriers have much autonomy, they have a great influence on the outcome of the VRP. In other words, autonomy determines the distribution of responsibility of the VRP solving job, as follows from the theory of Hackman and Oldham (1975).

In VRPs, autonomy is occasionally taken into account in the form of a cost in the objective function (Abu Al Hla et al., 2019). Moreover, in the literature, it is consistently assumed that the planner has more autonomy (and consequently more responsibility) than the courier (Srivatsa Srinivas and Gajanand, 2017). However, in a real-life setting, another factor might play a role in this respect: job satisfaction. It is known that autonomy moderates the relationship between job complexity and job satisfaction (Chung-Yan, 2010). A complex job is characterised as mentally chal-lenging, requiring the courier to use a number of complex skills. In contrast, a simple job can be performed by adhering to standard procedures and does not require the courier to alter his work methods (Chung-Yan, 2010). This concept is comparable to skill variety in the core job dimensions (Hackman and Oldham, 1975). Chung-Yan (2010) argues that increasing levels of job complexity should be matched by a rise in job autonomy to mitigate negative effects on job satisfaction.

(14)

2.3.2 Route-choice behaviour

Route-choice behaviour is a topic discussed in various fields of science such as In-formation Engineering (Ciscal-Terry et al., 2016), Transportation Science (Lin et al., 2012), Civil Engineering (Samimi et al., 2010; Zhu and Levinson, 2015), Environmen-tal Sustainability (Muslim et al., 2018) and Operations Management(Srivatsa Srinivas and Gajanand, 2017). As opposed to the research on human VRP solution strategies, route-choice behaviour covers the behaviours of humans while being en route.

Muslim et al. (2018) argue that route-choice behaviour is rather complex and is influenced by psychological, psychosocial and cognitive attributes. Lin et al. (2012) agree that the practice of mapping route-choice behaviour is especially difficult. For instance, the studies from Zhu and Levinson (2015) and Ciscal-Terry et al. (2016) show that humans do not take the shortest path when given full autonomy in route-choice decision. This contradicts the widely applied assumption that humans would rationally choose for the shortest path. Regarding VRPs, the solution is commonly a set of nodes in a specific order and not a precise route description. Therefore, the assumed distance (cost) between two nodes may not reflect the actual distance a human would travel. Many behavioural aspects, such as age, gender, task urgency, physiological status and weather, can influence route-choice behaviour (Lin et al. (2012)). It is expected that this will also influence the parcel delivery processes.

2.3.3 Real-time traffic congestion

(15)

(16)

3 Methodology

Following the theoretical framework, there exists a clear need for theory building focusing on human VRP solving in practice and the interconnection with VRP al-gorithm research and implementation. Fundamental theory on human VRP solving in a practical non-experimental setting is fairly immature. As mentioned before, academics have pointed out that algorithms have not evolved to fully capture the complexity of reality. In this research, the aim is to capture the human VRP solving practices in all its complexity. An exploratory stance in this process will help to gain a better understanding of the subject to properly build theory around it. A case study is well suited for theory building in a fairly immature and complex environment is a case research (Benbasat et al., 1987; Fynes et al., 2015; Voss et al., 2016). To match the proposed aim of this research, the inductive research process, as described by Kovács and Spens (2005), is adopted.

3.1 Case selection

This research concerns a single case: a parcel delivery company. This company is a market leader in the Netherlands. This case satisfies several important criteria for this research. The case is a well-established company in the parcel delivery industry. The depot, from which the participants in this case operate, typically handles over 40,000 parcels on a daily basis. Moreover, the range of experience of the participants is high. The enhances the opportunity to examine the process from multiple perspectives. Some have less than one year experience, while others have witnessed the parcel delivery industry bloom decades ago. Furthermore, the case does not involve VRP algorithms, but relies solely on human performance. Therefore, this case captures pure human processes which helps to maximise the exposure of the differences between pure human problem solving and algorithms.

(17)

is chosen to allow for greater in-depth observations. This is justified regarding the complexity of the subject and immaturity of the theory. A disadvantage of the single case study is the limited generalisability, compared to multiple case studies. This will be taken into account into interpreting the implications of this research.

3.2 Coding and analysis

The research consists of semi-structured interviews with the couriers and process managers. The interviews are partly structured through a prepared set of questions (Appendix A). However, during the interviews the participants are encouraged to elaborate on topics valuable to them in solving their particular routing problem, and thus valuable for this research too. To ensure reliability, key questions are asked to every participant in similar fashion in order to verify certain statements and/or infor-mation. Furthermore, question repeating, repeating a question in a slightly altered fashion to the same participant, further enhances the level of verification of statements and/or information. With the key informants, the PMs, the conveyed information by the couriers is verified. The happened during the interview with the PMs.

(18)

(19)

4 Results

With help of the aforementioned coding and analysis method (Auerbach and Silver-stein, 2003) a coding tree is constructed. The coding tree can been found in appendix B. The analysis below is subdivided into operational methods and external factors. In appendix C all identified distinct constraints as well as the objectives are listed and described. The operational methods of the PMs and couriers use is conceptualised in figure 4.1. Concepts corresponding to operational processes displayed in the figure are highlighted in bold.

4.1 Operational methods

The foundation of the VRP solving method is the use of fixed routes. The couriers are structurally connected to one specific fixed route that covers a portion of the total map. Fixed routes are developed and improved from two input sources. The first input source is the clustering of delivery addresses based on their common zip code. For each zip code area the average parcel delivery rate is monitored. Typically, a fixed route comprises several zip code areas, being the building blocks of the fixed route. The second input source is the route knowledge of couriers. Couriers are regularly consulted in the process of fixed route development and improvement. The couriers suggest what changes to be made to the fixed routes according to their practical experiences. By gut feeling they evaluate the route and suggest what route changes may structurally decrease travel distance.

(20)

Figure 4.1: Conceptual framework of the operational methods. The operational meth-ods can be defined as a feedback loop of operational processes (

), between which information (→) is flowing. The quadrants depict who is responsible for the process and whether the process belongs to daily operations or long-term development. The arrow captions depict the circumstances in which the process should be performed or not. The information flow between the manager and the courier is connected in the following way: the courier passes his/her route knowledge to the PM and the manager provides the actual route composition on a daily basis.

route is termed open route. During transferring, the couriers’ well-being, experience and preferences are taken into account. Moreover the PM will avoid interfering with outsourced routes and vehicle separation.

(21)

process of zip code transferring can become complex when many neighbouring fixed routes are close to being overburdened. This sometimes causes a back-and-forth be-tween different route compositions and zip code area transfers. When the problem is unsolvable by the PM, an extra (off duty) courier and truck is added to the fleet to cover for the open route(s). If no extra couriers are available then the parcels from the open route(s) are transferred to the routes for the following delivery day, but this is highly exceptional.

The PM transfers the (altered) route to the courier. Prior to leaving the depot, the courier manually checks the given route on two criteria. First, the courier checks whether the addresses seem to be in the correct order. This ’correct order’ is usually defined as the fixed route the courier is used to. Due to software or manual errors, some addresses might not be in place according to the fixed route of the courier. Furthermore, the courier might take into account opening hours of companies as well as the visual appearance of the route. Secondly, when the courier has received extra zip code areas due to redistribution, he/she assesses the geographical location of the extra zip code areas relative to zip code areas of the fixed route. Usually, the extra zip code areas are added at the end of the route. However, when the extra zip code area is tightly neighbouring the fixed route, it sometimes, but not often, is decided to position it inside the fixed route. The underlying motivation of pre-shift route checking is to increase the percentage of time window adherence for the day. In other words, the better the courier has predicted the order in which he will deliver the parcel, the higher the percentage of time window adherence.

(22)

For every encounter, the courier assesses whether it was incidental or struc-tural. When the same problem reoccurs over several shifts, the courier will deem it a structural problem and adds this knowledge to his general route knowledge. In turn, this information is shared with the PM, who will take this information into account at structurally developing the fixed routes.

4.2 External factors

4.2.1 Courier autonomy, job satisfaction and ’walking the extra mile’

Autonomy is a valuable asset for the couriers. Within the delivery company, the courier is granted a sufficient amount of autonomy. This is reflected by the fact that couriers are completely free to alter their routes if they deem it necessary. Some argue that the level of autonomy influences the level of job satisfaction of the courier. One courier associates having less autonomy with "feeling like a robot", which ultimately reduces the level of job satisfaction. Furthermore, it seems that having more auton-omy and job satisfaction results in couriers being motivated to ’walk the extra mile’. An example of this is when a courier goes to a delivery address a second time at the end of his/her shift, because the customer was not home at the time of initial delivery attempt.

4.2.2 Vehicle routing technology

During the interviews, the topic of vehicle routing technology was often touched upon. Upon asking to what extent the courier believes technology might improve parcel delivery practices, there was a clearly observable consensus among the couriers. In essence, couriers are doubtful whether technology can further help improve parcel delivery practice in terms of efficiency. Couriers argue that “a computer doesn’t know which streets are accessible and which not”, “the computer doesn’t always know what reality looks like” and that couriers generally know more detailed information about reality than any computer would.

(23)

sequence of streets to drive through.” This seems to be particularly true for densely populated areas, where the complexity of traffic congestion usually plays a greater role than in more rural areas.

Furthermore, some argue that involving more technology in the practice of parcel delivery may impair the level of autonomy of the couriers.

4.2.3 Courier experience and route stability

The majority of the couriers report that they know their fixed delivery area by heart. This includes knowing street names, specific addresses, and sometimes even knowing the customers personally. The courier explain that this is due to the couriers having a fixed route, which enables to quickly learn and memorise the streets and addresses. This would not be possible when the route assignment was variable among the couri-ers. The couriers suggest that having the knowledge of the street configurations contributes greatly to the efficiency of the couriers travel and delivery times.

4.2.4 Traffic congestion

(24)

5 Discussion

The results are analysed and put into perspective with respect to current literature. This leads to the development of five propositions, presented in this section. The first three propositions focus on human VRP solution strategies, whereas the last two propositions focus on external factors.

5.1 Human VRP solution strategies

From the framework in figure 4.1 the operational process is conceptualised as a contin-uous feedback loop. In this feedback loop, the solving strategy is alternating between a long-term and a daily focus. The long-term focus is about global route develop-ment. This is propelled by courier route knowledge development, but mainly applied by the PM. A fundamental element of the global route development is clustering of zip code areas. The daily focus is local oriented, where the global route distribution remains essentially fixed, except for alterations between neighbouring routes or within a route. Once the routes are set for the day, couriers start their shift and encounter a tremendous number of different problems while being en route. Some examples of this are toilet breaks, a certain spatial layout of streets, public events, road obstructions, walking versus driving, not being able to find a parking spot and the list goes on. Consequently, the courier has to make local routing decisions frequently.

In VRP algorithm research, the challenge is often to involve more complex, realistic characteristics. Regarding the observations on the sheer complexity of local route decision making from the couriers, this could also be regarded as a new challenge to develop more rich algorithms to handle the complexity on a local level. However, in light of this research, this is not recommend. Following from the debate on whether the “proper richness” of algorithms can ever be practically achieved, it is proposed to actually avoid this level of realistic complexity. After all, there is enough suspicion that algorithms struggle to handle this type of characteristics, while there is no reason to doubt that couriers actually can.

(25)

It is observed that, in practice, many human characteristics, such as courier well-being, experience and preferences, are taken into account. These characteristics de-termine the delivery rate of the courier, which, as mentioned by some couriers, can differ greatly. For instance, the delivery rate of inexperienced and experienced couri-ers is 80 and 140 parcel per day, respectively. Therefore, it is assumed that these characteristics greatly affect the outcome of the VRP solution.

As mentioned in the theoretical framework, there is limited focus on human char-acteristics in algorithm research. From this, it is proposed that algorithm research may benefit from taking into account more human characteristics to match real-life VRP practices. However, this does not apply for any type of human characteristics.

A distinction should be made in computational ability of algorithms to include these human characteristics. In other words, some VRP characteristics might be harder to translate in computable numbers than others. For instance, involving char-acteristics such as visually appealing routes might be harder to quantify than, for example, courier experience. It can be argued that more subjective-type characteris-tics are better accounted for by humans than algorithms, because, after all: opinions can vary among people to what extent a route is visually appealing or not. In con-trast, courier experience can be well-expressed in terms of number of parcels delivered, or total number of shifts completed. Hence, the recommendation is that algorithm research should focus on human characteristics for which strong evidence exists that algorithms would in practice outperform humans in processing these characteristic. Proposition 2 In order to advance VRP algorithm research, a greater focus should lay on involvement of well-quantifiable human characteristics.

(26)

nodes (addresses), 10 vehicles and possibly a few constraints, real-life VRPs are in-herently disparate from experimental settings. Hence, similar to suspicion raised in the theoretical framework, the ecological validity of current experiments is question-able. Hence, it is proposed that the experimental setting should be complicated by involving more nodes, vehicles and constraints.

Proposition 3 To enhance the ecological validity of experimental studies on human VRP solution strategies, experiments should involve more complex VRPs, with many nodes, vehicles and constraints, to resemble real-life environments.

5.2 External factors

Regarding the five core job dimensions of Hackman and Oldham (1975), it is observed that the courier is granted ample autonomy. Couriers are entirely free to decide on and deviate from route decisions. In other words, the courier has a high responsibility in route decision making. On a global level, the PM has the overarching responsibility of the zip code area distribution. As Srivatsa Srinivas and Gajanand (2017) state, algorithms consistently assume that the planner has more autonomy than the courier. Following from the observations, this is not confirmed in practice. In fact, the courier has much more influence on the actual execution of the route than the PM.

(27)

Proposition 4 The ensure proper algorithm implementation, courier autonomy should match the relative job complexity to preserve job satisfaction.

In this case study, the couriers are assigned to fixed routes. Furthermore, there exists a tendency for PMs as well as the courier to follow this fixed route. Therefore, on a global level, the variability of the route is considered to be low for the couriers. This allows for an environment where the courier knows his/her global delivery area by heart and increases the benefits of courier experience. The participants suggest that this highly influences their job performances in terms of travel and delivery time, still in a complex real-life environment.

However, on a local level, it has been observed that couriers experience a tremen-dous amount of variability as a result of encountering the complexities of the real-life environment. Some quite trivial occurrences can have a great influence on the route. Elements such as bathroom breaks, road obstructions and specific spatial street lay-outs vary on a daily basis. One can decide to model for it, or one can decide to let the courier account for it. There seems to be a trade-off between variability, caused by complexity emerging from reality, and efficiency gain. When variability is low, the potential for improvement is high. When variability is high, one can improve some-thing, but the gain might not hold for all circumstances in general. In other words, the variability of real-life might impede the algorithm gains that can be achieved in this setting. This theory is conceptualised in figure 5.1.

(28)

(29)

6 Conclusion

This research is an attempt to identify differences between human problem solving and VRP algorithms and develop propositions to advance VRP algorithm research and implementation. In this research, an inductive single case study is performed in the form of qualitative interviews. After transcribing and coding, the results are analysed and compared to literature on algorithm research and human VRP solution strategies. From this, a conceptual framework of the operational methods within the company is developed. Thereafter, the findings are put into perspective with current literature, from which five propositions are developed. These propositions impose theoretical as well as managerial implications, as discussed in this section. Furthermore, limitations and recommendations for future research are discussed.

6.1 Theoretical implications

(30)

Furthermore, in current literature, a limited number of experimental studies on hu-man VRP solution strategies have been conducted. However, the complexity and ver-satility of real-life VRPs is very different from current experimental settings. Hence, the ecological validity of these experiments becomes questionable. It is proposed that this can be mitigated by making the VRPs more complex in the experimental setting, in order to better resemble real-life VRPs. This will enhance the ecological validity, and will therefore lead to a better understanding of human VRP solution strategies in practice.

6.2 Managerial implications

In order to successfully implement an algorithm in practice, complexity, autonomy, job satisfaction and variability should be taken into account. Two propositions have been developed that explain how these factors are interrelated. The level of autonomy should match the level of complexity involved in the courier’s job in order to retain the level of job satisfaction among employees. Furthermore, it is suggested that the potential of algorithms lies within less variable and complex environments, whereas the courier can account for the variability emerging from real-life complexity. This has implications on how algorithms should be implemented in practice. It is rec-ommended that the algorithm should not be applied for the entire solving process. Preferably, algorithms are used to develop good initial solutions to the real-life VRP by conceptualising the VRP setting. Then, humans can account for complex real-life characteristics that are bound to come into play in practice.

6.3 Limitations and future research

(31)

In order to maximise the exposure of the differences between human VRP solving practices and VRP algorithms a case was selected in which algorithms are not used entirely. However, this did not settle assumptions made on actual algorithm imple-mentation in practice. It would be valuable to research the actual effects of VRP algorithm implementation in practice. Specifically, three interesting questions can be raised with respect to the impairment of autonomy, caused by implementation. The notion that job complexity would influence job satisfaction is predicated on the level of autonomy the courier perceives. It would be interesting to test how implementation affects courier autonomy. Suppose algorithms could be perfectly dynamic to real-life VRPs, would the autonomy of the courier then decrease? And if so, would this be accompanied by a loss of job complexity? And from a psychological perspective, how would this influence the level of job satisfaction of the courier?

(32)

A

Interview questions

The questions are design primarily designed for the couriers. Similar questions are asked to the PMs, however, altered where needed in order to be applicable.

1. What does your typical working day look like? 2. How many parcels do you deliver on average? 3. When do you consider a certain route to be good ?

4. What factors are influencing the efficiency of your route? 5. Is your current route as efficient as possible?

6. How much freedom (autonomy) do you perceive to make decisions? 7. Do factors such as experience, age, etc. play a role?

8. How does traffic congestion influence a route?

(33)

B

Coding tree

Quotes beginning with (PM) are from the process manager. Other quotes are from the couriers. Grey cells represent an overlap between two bordering themes.

Example quote (translated) Code Second order Theme (PM) “I think a courier delivers parcels between 6 and 7 hours daily.”

Time capacity

Capacity

Human VRP “You basically plan routes mostly according to number of stops.”

“If routes are too full, then some addresses have to be taken out”

Volume capacity “We work with cubic meters. Around 6 cubic meters fits in a bus. If a route

has more than 6 cubic meters, some stops need to be taken out. We look at couriers that drive in the neighbourhood, who have less parcels.”

“You plan routes mostly on the basis of number of stops. But, you also have to check how much parcels fit in the bus.”

Combination of time and volume capacity “A courier has one hour before and one hour to deliver the parcel” Customer time

window

Time window indication “You have a time window indication, TWI in short. It shows a certain begin

and endtime. Between those times, the parcel needs to be delivered.” “I think 97% of parcels need to be delivered within the TWI. If they don’t then the company gets in trouble with their clients.”

Time window percentage score “The goal is to deliver 95% of the parcels within the TWI to do well”

“You have a very wide time window” Time window adherence “It shouldn’t be the case that when you encounter a street which is due at

the end of your route, that you would also deliver at that street, because this street might be out of your current time window.”

“But it can happen that people order an extremely large number of parcels. For example, I would normally have 150 stops on the entire route, but if people order more I might have 300, so I get a shorter route than that. Then I drive half of my route to 60%, and then another driver is deployed to deliver the rest.”

Stop transfer: capacity

Stop transfer “Sometimes I simply have too many addresses in [village] that does not fit in

my bus, which costs too much time. Then, what they do is then just take out part of my route. They then throw it in someone else’s route, who has less parcels.”

(34)

“For example, if I were ill, route 51 would be divided into routes neighbouring to it.”

(PM) “If we take something out, we’re actually trying to take out the zip code

areas of the routes.” Order of routes “Suppose I have relatively few addresses, then they choose to put some extra

addresses to my route. If I have a little more, in principle they do not choose to add anything to my route.”

“Actually you always drive the same route. Actually a bit stupid of course,

but that’s how we do it.” Constant routes

Constant routes “Yes, you actually have a standard route.”

“And that’s also because there are so many orders nowadays that I always

have at least one parcel in every street.” Demand “You actually already have a fixed route with regular customers.”

“Depending on how many regular customers you have in your fixed route,

determines how that route will turn out during the day.” Efficiency “In principle, you keep the same route, because that is simply the most

ef-ficient. At some point you get to know it more and more and then it goes faster and faster.”

(PM) “Yes right, there are few people who want a lot of change.” Tendency to adhere to constant route

“Yes, it is just the route that I just have in my head. I have to say that I am somewhat obsessive if it doesn’t work out that way.”

(PM) “My colleague does route optimisation. In any case, he keeps track of

all routes in a file. It also checks whether changes are visible.” Improvement

Route development “At some point you do not change anything because you know that it is simply

the best route.”

“At some point I was asked by the guy of the TWI to write my route it on

paper. Because the route order had to be passed on to him.” Communication “I sometimes deviate, but I do not communicate that to the TWI guy, because

I am not sure that I deviate every day.”

Question: “If you have a day with few regular customers?” Answer “Then I

tend to make almost experimental routes. It has been thought up in advance.” Experimenting “I deviate that more out of common sense because I just prefer that way”

(35)

“Then they search for the zip codes which are next to each other, which they

then make up a route.” Clustering process (PM) “Someone from the management at the depot will just start somewhere

to click routes, so to speak. He/she collects all zip code areas until it has a whole route, and you try to do that as beautifully as possible.”

(PM) “We have also outsourced many routes to entrepreneurs. They have a

specific delivery routes.” Outsourcing Outsourcing (PM) “If it is the case that we want to transfer a zip code area, then you also

have to go into discussion with them.” (PM expresses annoyance)

“At some point we get the scanner. And the route is broadly stated in there. Then we can put that in order ourselves.”

Courier route checking

Pre-shift route checking “Every morning, if I have time, before I start loading, I go through the scanner

and actually put it in order as I have it in my head.”

“But in the end I just know where I have to drive and what is the most efficient for me to drive.”

Route order assessment “If you scroll through that route, you can see the order in which they are

listed. Some streets is always put in wrong, and I have to move it.”

“They are very keen on those TWI scores, it is also in their interest. So, they have to adjust the route to get a certain TWI score.”

Time window adherence through route checking “Since we work with a time indication, which is very important for [company],

we must also adjust it ourselves and fill in how we drive the route.”

(PM) “If they are over 60, you just leave them on their own route. You are

not going to mess with that.” Courier age

Employee well-being “I just want to work relaxed. If you work 5 days, it is quite energy-consuming

work.” Stress

“One day it is stress, the other day it is fun.”

(PM) “For example, for someone who has back problems, we can also sort by weight, we can also see how heavy a parcel is.”

Physical impairments (PM) “For example, we can choose to take out heavy parcels.”

“But if I can see, for example, during loading that I have large parcels, I often choose to do that address at the beginning of my route. In that way I get rid of those large parcels earlier so that I have more space in my bus, so I have a somewhat more relaxing working day.”

Heavy parcels

Idiosyncratic preference

(36)

“There is also a difference between parcel deliverers who love to walk and deliverers who love to stay on that bus right up to the front door.”

Walking versus driving “Yes, what I also do now: On the [square in city] you now have a larger market

area. It is stretched across the sidewalk. That part I often do by foot.” “And that means that you sometimes go the extra mile for some people. Sometimes you are left with a parcel of someone who was not there, while you know that they are there, later in the day. Then you will visit it again.”

’To walk the extra mile’

“But people also know, of course, what time you are there. Not everyone keeps track of that on the computer, it just arriving at a certain time. If you drive the route the same every time, they will know about it, it is about that time. That is more pleasant for the customer.”

“There may be a pole on the road somewhere. For example, I also have a

street, then the second house is after the pole, while it is the same street.” road obstructions

Complexity of reality “At the [street] you also have the [supermarket]. And on that road to get to

the [street] is also the loading point of the [supermarket], where there is often a truck.”

“Sometimes I don’t have food with me, so I have to go to the supermarket:

’Can I plan the route differently?’ ” Breaks (PM) “Yes, and things you don’t know is when someone wants to have lunch,

and what would be a tactical place for that, where they can also go to the toilet.”

“It is possible that a house lies just in the middle of a street, so it is easier to get from one side because that street is at the end, for example. It depends a bit on where the house is in the street.”

Spatial layout of streets

“For example, I have a street, where I always have half in the middle of my route. And the other half of that street, while it is really next to each other, is much later in my route. But that’s just because it’s a different zip code.” “If I had to pass a house anyway. For example, you have a street and you have a block of houses. And I have to be at the furthest house. I also pass by anyway because the street next to it I have to go too. So then I skip it

(37)

“Every now and then I deviate. Because you have a number of streets, where you can not really stand at the door with the bus. But via another street, if you are already in a street, I just mention street A and street B is perpendic-ular to the other street, then you do not have to drive all the way to street B but then you walk from street A to street B, because that is a bit faster.” “Sometimes you have a residential area with a lot of cars, like, "Okay, then I better drive differently." For example.”

Circumstantial occurrences “Yes, what could also be a running race or other type of events, of course,

means that you have to drive your route differently.”

“Only it is of course very difficult to take current information such as traffic and events into account.”

Complexity of reality “There are often things that the computer does not know. And the computer

does not know where you can and cannot go by bus.”

“I have enough freedom, and I think it’s a wonderful job.” Prevalence of autonomy

Autonomy

External factors “I think [company] wants to preserve the freedom and autonomy for its

em-ployees.”

“Then you have a certain zip code and then indeed that freedom comes from

the drivers.” Use of autonomy “The better you know your route, the more you will play with it.”

“Yes, that you don’t feel like a robot, but you can also take that freedom if you think it will work out more efficiently for you.”

Trade-off technology versus autonomy (PM) “But I think if you really look at day routes where it’s really people’s

job, something they do every day. I always wonder, say, in theory how nice it is if your freedom is increasingly limited.”

“That is the most important thing. If you don’t enjoy it, I think the day will

be very long.” Job satisfaction

Job satisfaction “But it is damn nice if you sit on the couch at 3:00 PM or 4:00 PM when

you know that you were actually scheduled until 5 AM. That gives a kind of satisfaction.”

“The other part, of course, is the human. Someone also wants to be free in how he/she works. And that is also valuable. Employee satisfaction is also valuable. Autonomy, versatility. You can pin down everything very strictly, but perhaps that could also be counterproductive.”

Job satisfaction and autonomy

(PM) “But I think if you really look at day routes where it’s really people’s

(38)

“People are going to know you and you are going to know people, that is so much fun.”

Customer-courier relationships “If I haven’t been there for a week, people will say "hey where were you?"

And that’s nice, that’s the recognition.”

“And I have always been in the city center and I know exactly every alley” Fixed route knowledge

Experience “I have been working there for a year and a half now. And I already know it

better than my native village.”

“What is often the case with such a market, you notice that the first time. And then you deal with it, on the spot.”

Practical experience “The other day you just know, oh it’s Tuesday, I know that so and so are not

at home.”

(PM) “On a busy day, an experienced person could do 140 addresses within 8 hours.”

Delivery rate

inexperience/experienced (PM) “An inexperienced person who could do maybe 80.”

“Anyway, I have to give them credit for that, it is built up very nicely, until the moment they say: well you can do it and then they really let you go, so to speak.”

Training new employees “The previous deliverer of this route taught me a route.”

“No, I was never really bothered by that, traffic.” Prevalence of traffic congestion

Traffic congestion Question: “Do you ever encounter from traffic congestion?” Answer: “Actually

not in my route.”

“Well, I try to follow the route as much as possible yes.” Courier response to traffic congestion Question: “And when traffic congestion occurs?” Answer: “No, then I’ll just

wait. Then I usually have to go that way.”

(PM) “But I think, in more densely populated areas there is a certain traffic

complexity, I think that the driver is often smarter.” Area difference (PM) “Sometimes they have to learn something new. Many people are afraid

of that.”

Resistance to change

Innovation and change (PM) “Very often, if there is something new, something different, another

product or something is added, that is often difficult.”

(PM) “Those are all those very subtle things you might be able to solve with

algorithms, but I don’t think the technology is enough developed for that.” Technology

(39)

(PM) “What may seem technically smart turns out to be very awkward streets to go through all the time.”

“Our depot in [location] is divided into 9 shifts. Shift 1 starts at about 7 o’clock and shift 9 starts at about 12 o’clock. You arrive, you get a bus with you. You dock it at the traffic center. You load the parcels on the bus. You will receive a route list for this. And then you just drive your route. Along withith the scanner.”

Job description courier

Case information (PM) “I am a process manager myself, I plan the routes early in the morning

and I am a point of contact for the drivers. I have around 35 drivers under me. I do the scheduling for the drivers. Some days I lead the process on the floor, then you also manage external parties.”

Job description process manager

“In this period we have between 140 and 150 stops.”

Average order size “For example, yesterday it was 41000 I thought.”

Table B.1: Coding tree.

(40)

C

Objectives and constraints

C.1 Process Manager

Objective

The objective for PMs is to minimise travel distance and minimise the number of vehicles used per day. For PMs the objective is quite similar to that of traditional VRP algorithms. In the first place the objective is to minimise the total travel distance. Additionally, the PMs try to minimise the fleet size per day. This incentivises the volume capacity utilisation of each vehicle as much as possible. Constraints

(PM1) Parcel delivery guarantee

In principle, every parcel needs to be delivered on the specific day it is assigned to be delivered. This is the hardest constraint of the list and will only be violated as a last resort.

(PM2) Volume capacity

The number of parcels that can be delivered in one route is limited by the volume capacity of the van. The van contains around 6 cubic meters of parcel space.

(PM3) Time capacity

Every fixed route is constructed from multiple zip code areas. Per zip code area the average delivery rate is monitored, from which it can be estimated how much time the courier needs to deliver the number of parcels per shift. Every courier has roughly 6 to 7 hours per shift available between leaving the depot and returning to the depot. The remaining time is used for loading the van, checking the route and other routine preparations. Hence, the number of parcels the courier is able to deliver per shift is limited.

(PM4) Experience of courier

The level of experience of the courier affects the delivery rate of parcels. An inexperienced courier can deliver roughly 80 parcels per day whereas an experienced courier delivers up to 160 parcels. This affects the time capacity, because the average delivery rate of parcels per zip code area determines the total time needed to deliver the parcel.

(PM5) Constant courier to fixed route assignment

(41)

(PM6) Courier well-being

The PMs take the well-being of the couriers into account. They monitor the personal prefer-ences as well as the health conditions of the couriers and take this into account in the process. This might involve aspects such as age, physical impairments (for example back problems) and stress.

(PM7) Interference with outsourced routes

Some fixed routes are outsourced to a third party delivery service. The PMs are inclined to avoid the trouble of altering and interfering with the fixed routes that are covered by third party delivery services.

(PM8) Zip code areas

From the individual addresses, clusters are formed according to their common zip code. Route altering by PMs will mostly be limited to the zip code areas. Altering route positions of individual addresses rarely happens.

(PM9) Fixed routes

The PMs are strongly inclined to keep the routes as constant as possible for the couriers. When the routes are constant, variability is decreased for the couriers. When variability decreases the efficiency of delivery increases.

(PM10) Vehicle separation

A major assumption is that having multiple vehicles driving in the same zip code area is inefficient. Hence, this is avoided.

C.2 Courier

Objective

The objective for couriers is to minimise the total travel distance and/or minimise the level of stress they experience during delivery. Some couriers might want to focus on efficiency while enduring higher levels of stress, while others are focused on reducing their own experience of stress. The courier applies the following constraints.

Constraints

(C1) Time windows

(42)

(C2) Opening hours of companies

The opening hours of companies might not be aligned with the shift times of the courier. Hence, it might happen that companies are closed and consequently not able to receive a parcel during a courier’s shift. The courier takes this into account.

(C3) Follow the fixed route

The couriers are inclined to follow the fixed route they are assigned to. (C4) Location of extra addresses

Next to the fixed route couriers tend to follow, additional addresses from other routes might be added. These extra addresses need to be fit in the fixed route based on their geographical location. When an additional address is closely neighbouring the fixed route it might be fit between the fixed route. However, the additional addresses are usually added to the end of the fixed route.

(C5) Visually appealing route

Couriers associate a visually appealing route with a sense of efficiency. Two couriers have described their routes as like "one long snake" swaying through the zip code areas.

(C6) Position of delivery address relative to route and current position

It might occur that a certain geographical layout of streets, allows for a reduction in travel distance. An example of this is when delivery addresses are geographically neighbouring deliv-ery addresses from other streets. Reconfiguring the order of delivdeliv-ery might yield a reduction in travel distance. To properly explain this phenomenon, it is best described with an example from one of the couriers.

(43)

In this example, the courier has to deliver the last parcel of street A at the end of the street. Moreover, the courier also has to deliver parcels in street B. While being on street A, naturally the courier would drive to the address following street A and deliver the parcels on street B at some other moment. However, it would be quicker to skip the address in street A, and deliver its parcel while being parked on street B. From street B, the courier can simply walk to the address at street A, reducing the travel distance.

(C7) Idiosyncratic preferences

There are various idiosyncratic preferences that can affect the route configuration. Some examples of these preferences are the tendency to deliver heavy parcels first, or the distance couriers are prepared to walk as opposed to driving. Another example is the extent couriers are willing to ’walk the extra mile’ for the customers. For instance, returning to an address at which the customer was initially not present to receive the parcel.

(C8) Lunch/bathroom breaks

Couriers take into account where they can purchase their lunch (when they have not prepared lunch beforehand) and where they can go to the bathroom. This might affect the configuration of the route.

(C9) Road obstructions

(44)

References

Y. Abu Al Hla, M. Othman, and Y. Saleh. Optimising an eco-friendly vehicle routing problem model using regular and occasional drivers integrated with driver behaviour control. Journal of Cleaner Production, 234:984–1001, 2019. ISSN 09596526. doi: 10.1016/j.jclepro.2019.06.156. URL https://doi.org/10.1016/j.jclepro.2019.06.156.

N. Ando and E. Taniguchi. Travel time reliability in vehicle routing and scheduling with time windows. Networks and Spatial Economics, 6(3-4):293–311, 2006. ISSN 1566113X. doi: 10.1007/s11067-006-9285-8.

C. Archetti, D. Feillet, M. Gendreau, and M. G. Speranza. Complexity of the vrp and sdvrp. Transportation Research Part C: Emerging Technologies, 19(5):741–750, 2011.

C. Auerbach and L. B. Silverstein. Qualitative Data : An Introduction to Coding and Analysis, volume 21. New York University Press, 2003. ISBN 9789896540821. URL

http://ebookcentral.proquest.com/lib/rug/detail.action?docID=865323.

I. Benbasat, D. K. Goldstein, and M. Mead. The case research strategy in studies of information systems. MIS quarterly, pages 369–386, 1987.

K. Braekers, K. Ramaekers, and I. Van Nieuwenhuyse. The vehicle routing problem: State of the art classification and review. Computers & Industrial Engineering, 99:300–313, 2016.

D. Cattaruzza, N. Absi, D. Feillet, and J. González-Feliu. Vehicle routing problems for city logistics. EURO Journal on Transportation and Logistics, 6(1):51–79, 2017. ISSN 21924384. doi: 10.1007/s13676-014-0074-0.

G. A. Chung-Yan. The nonlinear effects of job complexity and autonomy on job satisfaction, turnover, and psychological well-being. Journal of occupational health psychology, 15(3):237, 2010.

W. Ciscal-Terry, M. Dell’Amico, N. S. Hadjidimitriou, and M. Iori. An analysis of drivers route choice behaviour using GPS data and optimal alternatives. Journal of Transport Geography, 51: 119–129, 2016. ISSN 09666923. doi: 10.1016/j.jtrangeo.2015.12.003. URL

http://dx.doi.org/10.1016/j.jtrangeo.2015.12.003.

C. Cotta, M. Sevaux, and K. Sörensen. Adaptive and multilevel metaheuristics, volume 136. Springer, 2008.

(45)

B. J. Dietvorst, J. P. Simmons, and C. Massey. Algorithm aversion: People erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General, 144(1):114–126, 2015. ISSN 00963445. doi: 10.1037/xge0000033.

M. Drexl. Rich vehicle routing in theory and practice. Logistics Research, 5(1-2):47–63, 2012. ISSN 1865035X. doi: 10.1007/s12159-012-0080-2.

R. Elshaer and H. Awad. A taxonomic review of metaheuristic algorithms for solving the vehicle routing problem and its variants. Computers and Industrial Engineering, 140(December 2019): 106242, 2020. ISSN 03608352. doi: 10.1016/j.cie.2019.106242. URL

https://doi.org/10.1016/j.cie.2019.106242.

M. M. Flood. The Traveling-Salesman Problem. (March 2020), 1956.

P. Fontaine, F. Taube, and S. Minner. Human solution strategies for the vehicle routing problem: Experimental findings and a choice-based theory. Computers and Operations Research, 120, 2020. ISSN 03050548. doi: 10.1016/j.cor.2020.104962.

B. Fynes, P. Coughlan, H. Boer, M. Holweg, M. Kilduff, M. Pagell, R. Schmenner, and C. Voss. Making a meaningful contribution to theory. International Journal of Operations & Production Management, 2015.

S. P. Gayialis, G. D. Konstantakopoulos, and I. P. Tatsiopoulos. Vehicle routing problem for urban freight transportation: A review of the recent literature. In Operational Research in the Digital Era–ICT Challenges, pages 89–104. Springer, 2019.

J. R. Hackman and G. R. Oldham. Development of the Job Diagnostic Survey. Journal of Applied Psychology, 60(2):159–170, 1975. ISSN 00219010. doi: 10.1037/h0076546.

Y. He and Z. Yang. Parcel delivery by collaborative use of truck fleets and bus-transit vehicles. Transportation Journal, 57(4):399–428, 2018. ISSN 00411612. doi:

10.5325/transportationj.57.4.0399.

W. Hofmann, T. Assmann, Z. D. Neghabadi, V. D. Cung, and J. Tolujevs. A simulation tool to assess the integration of cargo bikes into an urban distribution system. 5th International Workshop on Simulation for Energy, Sustainable Development and Environment, SESDE 2017, Held at the International Multidisciplinary Modeling and Simulation Multiconference, I3M 2017, pages 11–20, 2017.

(46)

G. Kim. Optimal Vehicle Routing for Parcel Delivery with Considering Two Time Periods. IEEE International Conference on Industrial Engineering and Engineering Management, 2019-Decem: 918–922, 2019. ISSN 2157362X. doi: 10.1109/IEEM.2018.8607385.

G. Kovács and K. M. Spens. Abductive reasoning in logistics research. International journal of physical distribution & logistics management, 2005.

N. Lin, H.-d. Liu, and C.-q. Gong. Research and Simulation on Drivers Route Choice Behavior Cognition Model. International Journal of Computer Science Issues, 9(6):210–215, 2012. ISSN 1694-0784.

J. N. MacGregor and Y. Chu. Human performance on the traveling salesman and related problems: A review. The Journal of Problem Solving, 3(2):2, 2011.

K. Martins-Turner and K. Nagel. How driving multiple tours affects the results of last mile delivery vehicle routing problems. Procedia Computer Science, 151:840–845, 2019. ISSN 18770509. doi: 10.1016/j.procs.2019.04.115. URL https://doi.org/10.1016/j.procs.2019.04.115.

R. Masson, A. Trentini, F. Lehuédé, N. Malhéné, O. Péton, and H. Tlahig. Optimization of a city logistics transportation system with mixed passengers and goods. EURO Journal on

Transportation and Logistics, 6(1):81–109, 2017. ISSN 21924384. doi: 10.1007/s13676-015-0085-5.

B. McCrea. State of Transportation Management Systems: Trends to track in 2017 - Logistics Management, 2017. URL

https://www.logisticsmgmt.com/article/state_of_transportation_ management_systems_trends_to_track_in_2017.

N. H. Muslim, A. Keyvanfar, A. Shafaghat, M. M. Abdullahi, and M. Khorami. Green driver: Travel behaviors revisited on fuel saving and less emission. Sustainability (Switzerland), 10(2): 1–30, 2018. ISSN 20711050. doi: 10.3390/su10020325.

P. Pahlavani and M. R. Delavar. Multi-criteria route planning based on a driver’s preferences in multi-criteria route selection. Transportation Research Part C: Emerging Technologies, 40: 14–35, 2014. ISSN 0968090X. doi: 10.1016/j.trc.2014.01.001. URL

http://dx.doi.org/10.1016/j.trc.2014.01.001.

(47)

W. Salter. Inbound logistics: transportation planning a manual headache or an optimized dream., 2014. URL http://www.inboundlogistics.com/cms/article/ transportation-planning a-manual-headache-or-an-optimized-dream/.

A. Samimi, A. Mohammadian, and K. Kawamura. Behavioral freight movement modeling. The 12th International Conference on Travel Behaviour Research, (October):1–30, 2010.

M. D. Simoni, P. Bujanovic, S. D. Boyles, and E. Kutanoglu. Urban consolidation solutions for parcel delivery considering location, fleet and route choice. Case Studies on Transport Policy, 6 (1):112–124, 2018. ISSN 22136258. doi: 10.1016/j.cstp.2017.11.002. URL

https://doi.org/10.1016/j.cstp.2017.11.002.

S. Srivatsa Srinivas and M. S. Gajanand. Vehicle routing problem and driver behaviour: a review and framework for analysis. Transport Reviews, 37(5):590–611, 2017. ISSN 14645327. doi: 10.1080/01441647.2016.1273276.

Y. Sun. Decision-Making Process and Factors Affecting Truck Routing. Freight Transport Modelling, (2011):233–249, 2013. doi: 10.1108/9781781902868-012.

T. F. Tan and B. R. Staats. Behavioral drivers of routing decisions: Evidence from restaurant table assignment. Production and Operations Management, 29(4):1050–1070, 2020.

W. ter Haar. Pakketbezorger zijn is meer dan alleen 120 keer busje in, busje uit, 2018. URL https://www.ad.nl/ad-werkt/pakketbezorger-zijn-is-meer-dan-alleen-120-keer-busje-in-busje-uit a81eb41f/#: :text=Niels Bouman%2C sinds 2002 op,180 pakjes op een dag.

A. L. Tucker. An empirical study of system improvement by frontline employees in hospital units. Manufacturing & Service Operations Management, 9(4):492–505, 2007.

C. Voss, M. Johnson, and J. Godsell. Research methods for operations management. Chapter 5. Routledge, 2016.

J. Xu, F. Yan, and S. Li. Vehicle routing optimization with soft time windows in a fuzzy random environment. Transportation Research Part E: Logistics and Transportation Review, 47(6): 1075–1091, 2011. ISSN 13665545. doi: 10.1016/j.tre.2011.04.002. URL

http://dx.doi.org/10.1016/j.tre.2011.04.002.

A case study on human VRP solving practices in a parcel delivery setting