Adoption of the Physical Internet in healthcare logistics

(1)

1 Author: Liviu Romeo Racu

Student number: S2365235 Supervisor: Dr. ir. P. Buijs Co-assessor: Dr. ir. S. Fazi Date submission: 26/08/2018

(2)

Abstract

Purpose: The Physical Internet is an emerging concept in logistics that promises significant improvements in the sustainability and efficiency of distribution networks. However, due to its still rather novelty in the literature and in the applied sector, there is little known about factors that can contribute to its implementation. This thesis seeks to study what motivates managers to implement Physical Internet in their companies.

Design/methodology/approach: The thesis has a mixed qualitative and quantitative research design, namely, a combination of semi-structured interviews and a data simulation step. As a first step, interviews with managers of two companies active in the health care sector will be carried out. The findings of these interviews will inform the second step of the present research – data simulation about the here-to-be proposed six mixed classical-physical internet distribution networks. In a third and final step, the results of these simulations will inform once again interviews with the same managers, with the goal of identifying the managers’ interests in implementing the Physical Internet considering concrete strategies.

Findings: From the first round of interviews the following factors are identified as important in the implementation of the Physical Internet: improvement of flows, improvement of the deliveries spatial times and trust between partners. Addressing the issues using the Physical Internet methodology, the following findings resulted: the number of PI-hubs can improve the issues as well as strategical locations where the PI-hubs are placed. A third party organization would also be suitable in managing and operating the Physical Internet network in order to overcome the issue of trust and to increase the willingness of adopting the physical internet structure.

(4)

Practical implications: Organizations, as represented by their managers, seem to be willing to adopt the Physical Internet if a structure is already in place and if a third party organization manages their operations. Companies appear not to find as interesting the control of the Physical Internet and prefer acting passive towards Physical Internet management. The difficulties posed by managing the PI-hubs and the network itself are not motivating for companies to adopt the Physical Internet without public and/or private external support.

(5)

1

Introduction

Drawing inspiration from the principal characteristics of the internet as we know it, scholars in the domain of logistics management have recently proposed a novel system of global logistics: the Physical Internet (PI). The PI is an open global logistics system that brings together various physical, digital and operational actors on a single platform using interconnected hubs and protocols, with the goal to move, store, supply and use physical objects throughout delivery networks in a way that is economical and sustainable (Montreuil, Russell, & Ballot, Physical Internet Foundations, 2013). Montreuil (2009) suggested that the implementation of the PI would improve various aspects of the logistics chain, such as an improved utilization of vehicles and a reduction of CO2 emissions.

When bringing an innovative concept to the market, certain steps have to be followed in order to reduce the costs and risk of failure. The first stage for a successful adoption of an innovative concept is research and development. Moreover, the stage of research and development is divided into 3 phases: basic research, information collation and development testing (Greenhalgh & Rogers, 2010). In the case of PI, the first two phases have been extensively researched and developed in the PI literature (Sternberg & Norrman, 2017). On the other hand, the knowledge surrounding the third point - development testing and implementation - is scarce. Therefore, the goal of this thesis is to investigate the challenges posed by the testing and implementation of PI in order to bring this innovative concept closer to a real implementation scenario. Since companies are the main actors in a possible future implementation of PI, the thesis will research what logistical elements are important for managers so that they will consider implementing PI. Moreover, the thesis will analyse the factors important for managers in the PI implementation from a quantitative perspective by simulating different scenarios.

(6)

2

classical logistical concepts and PI logistical concepts. Due to their explicit interest in the PI, two companies in healthcare material logistics - Vegro and Rentex - have been chosen as case studies. Moreover, these companies have overlapping distribution networks in the healthcare sector, which classifies them as suitable unit of analysis for the case study. The companies are quite similar in their operations and market focus, therefore a cooperation between them is easier to establish than companies that are different in their characteristics. Moreover, since the research of PI implementation is at its beginning, it is important to use similar companies to avoid at this stage the complexities of different customer characteristics and different logistical distribution networks.

In order to conduct the study, logistical issues identified by the managers will be analysed by simulating different PI-classical network designs. The results of the simulated designs will be assessed based on network efficiency and presented to the managers for further remarks. A unique characteristic of this thesis compared to the literature of the PI is the investigation of different design models as well as having a closed cycle of interviews: pre-interviews, results and final interviews. Moreover, the designs proposed follow an incremental PI transition. That is, the first design is based on a classical logistical network and incrementally PI elements are introduced to the new designs until the PI elements become dominant in the network.

(7)

3

1. Theoretical Background

1.1. What is the Physical Internet?

The Digital Internet - where packages of information can travel from one user to another by using different means of the internet network (Montreuil, Physical Internet Manifesto, 2009) – is the main source of inspiration of the Physical Internet. It is defined as

“an open global logistics system founded on physical, digital and operational interconnectivity through encapsulation, interfaces and protocols. It is a perpetually evolving system driven by technological, infrastructure and business innovation” (Montreuil, Russell, & Ballot, Physical

Internet Foundations, 2013). Montreuil (2009) suggested that the PI could be used for the improvement of logistical issues, some examples of which include the better utilisation of vehicles, more efficient use of carriers and facilities, multimodal transportation, city logistics and a reduction of CO2 emissions, as well as a shortage of truck drivers.

Figure 1. Representation of classical distribution network vs PI facilitated distribution network (Yang, Pan , & Ballot, 2017)

(8)

4

and follow a hierarchical structure, while in the PI distribution map, the load flows are not predefined and are all part of a larger interconnected framework that offers more efficient possibilities regarding transportation and storage.

1.2. The PI-hubs and flows

PI-hubs are an important component of the PI system. They are defined as multimodal network nodes locations that facilitate the exchange of PI-loads by performing the unloading and reloading activities efficiently (Ballot, Montreuil, & Meller, 2014). In addition, by facilitating the exchange of loads between different transporters, their function is critical for the success of the PI environment. Due to the variety of types of transportation, difficulties arise regarding the exchange between different transporters; PI-hubs solve this problem due to their specialisation in mixing and adopting loads. Therefore the main benefits of PI-hubs are efficiency with regard to the loading and unloading times, as well as creating cross-points where loads can take different paths (Ballot, Montreuil, & Meller, 2014).

A characteristic of the PI network is the creation of compact standardized flows between the network’s core points. Since flows are a main component of distribution networks, it is important to observe the behaviour of flows in a PI enabled environment.

2A. Flow of goods in classical network 2B.Flow of good in PI network

(9)

5

The graphical representation in Figure 2 depicts two distribution networks in France which outline the difference in flow intensity between the different networks. The flows were derived by Hakimi et al (2012) in a simulation of the Physical Internet in France. The simulation aimed to quantify the impact of the PI in terms of economic, environmental and social effects. The results of the simulation showed a significant decrease of 20% in total kilometres travelled. With regard to the distribution flows, a convergence between different nodes in the network system can be observed, giving a clearer picture of the main routes of the network. Therefore, it can be concluded that the flows of the PI have a clear structure in a large system when compared with the classical logistical network. Another important observation is that routes in a PI enabled environment are more intensive between hubs and less intensive between final customers and hubs.

1.3. The adoption process of the Physical Internet

To understand the complexities of adopting the Physical Internet into the current logistical context, Sternberg & Norrman (2017) developed a PI adoption framework (Figure 3) based on the work of Iacovou, Benbasat, & Dexter (1995). The framework is used to categorise the state of the art PI literature into drivers of PI adoption, which includes business models, technological blueprints, promised effects and adoption of the PI itself. Business models refer to the internal efficiencies and indirect benefits associated with the implementation of a concept in an organization; technological blueprints explain the availability of components in order to implement an innovative concept; and promised effects are essentially what an innovative concept can offer to stakeholders in exchange for funding.

(10)

6

Figure 3. Adoption framework for the Physical Internet (Sternberg & Norrman, 2017)

The other article by Simmer et al (2017) - which looks at how the implementation of the PI affected the logistical environment in Austria - is considered to be the most extensive research on the topic of the adoption of PI. The study investigates the perception of Austria’s transport service providers in the context of horizontal collaboration and the PI. Interviews were conducted with 16 companies about the future of logistics and the challenges the PI poses. Their core findings are that trust and dependencies between partners are important for managers. As a final remark, the research underlines the need to implement horizontal collaboration as a first step towards the implementation of the PI.

(11)

7

be used to create a trust implementation plan for the managers of the companies in order to increase their willingness to implement PI.

Figure 4. Three step framework for trust development. (Pomponi, Fratocchi, & Tafuri, 2015)

1.4. Summary of literature findings

The concept of PI and the elements that form the concept of PI were proposed and elaborated by Ballot, Montreuil, & Meller (2014) which created a foundation for the PI. Regarding the business models, technology blue prints and promised effects, the literature is quite vast. Nevertheless, there is not much known about the adoption of PI and the focus PI research should have in order to implement the concept in the future (Sternberg & Norrman, 2017). It can be also remarked that PI research has reached a level where an implementation model is needed. In order for the test implementation to work, more information is needed about its process.

(12)

8

will involve two round of interviews: before and after the quantitative analysis. The quantitative part will rely on design methods and simulation of the designs.

2. Methodology

The general framework of the study design follows three stages: Stage 1) Pre-interviews; Stage 2) Design, simulation and data analysis; and Stage 3) Interviews. This type of design is suitable for the thesis because it offers insights into the manager’s interpretation of the data analysis. The aim of the interviews is to clarify what incentives might motivate practitioners to implement the PI according to their understanding about the topic. In order to demonstrate that PI is a solution for the incentives required by the practitioners, a simulation of their distribution networks will be developed. Lastly, the managers will view the results and conclude if PI provides enough benefits to be implemented. Due to the exploration phase of the topic, interviews and simulation are the recommended tools to investigate the implications of the topic (Karlsson, 2016) .The research will follow different stages in data collection and different types of data analysis. The motivation behind this type of design is the importance of finding out what managers think about the PI and if they would implement it.

2.1. Company Description

The focus will be on two companies who are active in healthcare logistics: Vegro and Rentex. The first company, Vegro, was chosen due to its active interest in innovation, especially regarding the Physical Internet. They are involved in selling and renting specially designed mattresses for medical purposes, as well as different medical equipment and supplies intended to help patients at home. In total, they own five distribution centres throughout the Netherlands, as well as a large network of shops, used for selling and renting items; approximately 20% of their orders are made in store. The demand for their products comes from both private and business customers, with 90% of the demand being private. The private customers are patients who were discharged from hospitals and are in need of special medical items at home.

(13)

9

which makes it difficult to state the exact number and type of trucks assigned to each DC. The headquarters of Vegro are located in Lisse, where orders and their distribution are planned.

The other company, Rentex, was chosen due to its logistical synergies with Vegro. Rentex is a company active in the healthcare sector and its core activity is providing healthcare institutions with clean textiles. The company has a reusable strategy regarding their textiles: everyday they bring clean textiles to healthcare institutions and pick up the used textiles in order to be cleaned for further use. They are active in the Northern and Western regions of the Netherlands and have a larger number of trucks - which can carry higher quantities of products - when compared to Vegro. The routes of the trucks are planned at their headquarters in Bolswald, Frisland and they operate seven days per week, including night shifts. Their main customers are hospitals and care centres who are supplied with standardised containers designed to facilitate the distribution of textile products.

2.1.1. Vegro demand characteristics

Regarding their logistical operations, Vegro has different flows in their distribution systems (Table 1). Some of the flows are outsourced to a third party logistic company. These outsourced flows mostly concern internal distribution flows. Different types of transportation vehicles with medium and large capacities serve the flows. The frequency of the flows is daily for the customers and two times per week for supplying the distribution centre and shops.

Flows Outsourced/Internal Type trucks Intensity

DC- Customers Internal Medium size 6 days / week

DC-Shops Outsourced Large size 3 days/week

Shops-Customers Internal Medium size 6 days/week

Suppliers- DC Outsourced Large size 3 days /week

Table 1. Flow description Vegro

(14)

10

the three regions, followed by Drenthe with a 31% demand orders. Orders in Frisland cumulate to a percentage of 10%.

Region Two digits postal code

Orders demanded per postal codes in %

Total demand per region in % Groningen 99 11.97% 58.41% 98 4.57% 97 20.90% 96 12.09% 95 8.88% Drenthe 94 10.59% 31.67% 93 5.59% 79 6.20% 78 9.29% Frisland 92 2.30% 9.92% 91 0.59% 90 1.05% 89 1.22% 88 0.59% 87 0.75% 86 0.75% 85 0.66% 84 2.02%

Table 2 Orders distribution per regions and two digits postal codes

Customer orders were further categorized on two dimensions. The first dimension is delivery and pick-up orders while the second dimension is time window of order request. Table 3 offers an overview of the distributions of the dimensions. The data derived for two dimensions was derived to estimate the types of demands per regions; however, due to the small demand of orders in the region of Friesland, orders could not be assign to the time window and type distributions. Therefore, the table is more informative about the type of demand Vegro has.

Order time windows 00:00:00.000 0830-1230 1030 - 1430 1230 -1630 1430 - 1830 Orders demanded per

time window 48% 28% 9% 10% 4%

Type of order Delivery Pickup

Orders percentage per

day 51% 49%

(15)

11

2.2. Research Design

In this thesis there are five stages of research (Table 4). Stage 1 of the research is dedicated to understanding what challenges the managers identify in their companies, by way of interviewing them. The questions for the interviews are divided into three topics: operational issues, managerial issues and the companies’ perception of adopting the PI. In this stage, it is essential to understand on what issues the PI should focus and what the PI should improve for the managers to be willing to adopt it. The interview followed pre-defined protocol (see Appendix 1).

Stage 2 is dedicated to collecting data on the companies’ product distribution systems (networks); access was given to the primary data of Vegro, though approximations had to be made using the available resources on the internet for Rentex. Furthermore, data on the distribution system of Rentex was collected via the interviews with the mangers of Vegro.

In Stage 3, the structure for the original network design was developed using data from the companies. Based on this structure, five new distribution networks designs are proposed using PI methodology. The designs were simulated using a routing solver and data was collected from the companies. The simulation’s replications were validated using a confidence interval method.

Stage 4 of the research is concerned with the analysis of the data outcomes from the design simulations. For the analysis of the designs, a comparison between the original network design and the other designs is used. The results are tested using a paired T-test to confirm that the differences between the designs are significant. A regression model was used to identify the efficiency of the time variables between the original design and the designs proposed.

(16)

12

Stages Type Research Data source Analysis Purpose

1.Gathering information about what issues managers have with their distribution Semi-structured interviews Logistical Manager Vegro Director Vegro Identify managerial and operational issues that PI could solve Understand the challenges of the company and the company perception about PI 2.Gather quantitative data to enable the simulation of a PI environment Data collection from Vegro Secondary data from Rentex Logistical Manager Vegro Website of Rentex Identify distribution of data and characteristics Prepare the set-up for the simulations

3. Develop network designs and simulate the designs

Simulation Data from Vegro

Based on the data, develop five new network designs which incorporate PI elements and simulate them using an excel solver Understand how other distribution network with PI elements will behave 4. Create a regression between the models to see how they defer from the original situation Regression analysis Simulations Run a number of regressions to observe the connections and implications of each design proposed

Find out how PI networks perform in comparison with the original network designs 5.Conclusion of the research Semi-structured interviews Innovation Manager Vegro Logistics Manager Vegro Logistics Manager Rentex

Present the managers with the results of the simulations and according to the data presented asked if they would consider implementing PI and if not why Understand if PI benefits would trigger managers to implement PI and if not, why not

Table 4 Research phases

2.3. Data collection

(17)

13

Due to the research period and the difficulties in gaining access to Rentex primary data, secondary data was used. Some of the characteristics of the data are that the data was requested only for the North of the Netherlands and for the time interval 1st May 2017- 30th November 2017. Since the networks of the two companies had to overlap in order to be able to analyse them, only the north region of the Netherlands was accounted for. The time interval was selected based on the interviews with the logistical manager of Vegro. During the time interval mentioned above, the companies had a stable demand of products and therefore it is the most representative period for how the company deals with distributing goods.

From Vegro, data about the following topics was requested and received: - Flows between all actors involved in the distribution network

- Customers locations and orders behavior (Delivery or Retour; Requested order time) - Number of trucks and capacity of their trucks.

- List of constrains regarding number of kilometers a driver can spend during the day, number of hours a truck is available, working times for the truck drivers

- Location of all facilities of Vegro in the North Netherlands

From Rentex, data about the following topics was collected: - Location of their customers

- Orders behavior (delivery and return characteristics) - Trucks type and capacity

- Location facilities in the North of the Netherlands

The data collected about the second company, Rentex, was done with the help of Vegro as well as the internet. From the interviews conducted with the Vegro employees, it has been concluded that Rentex serve all the hospital in the north of the Netherlands. Regarding their fleet of trucks, from their website and interviews was derived the size and capacity of their carrier’s fleet.

2.4. Quantitative data analysis

2.4.1. Designs

(18)

14

PI element introduced Purpose/Characteristics Design 1(original network

design)

N/A -Have a base to develop the

other designs

-Provide a general overview about the characteristics of the network

Design 2 -Shops of Vegro become

PI-hubs

-The flows are divided in two streams: DC-Shops and Shops-Customers

-The flow to the shops is already existent and not fully utilized due to low demand request from shops

-PI-hubs(shops) offer the company more flexibility to reach customers

Design 3 -The elements introduced in

design 2 are valid also in this design

-Rentex is introduced in the distribution

system-collaboration introduction

-The introduction of Rentex is to underline the importance of collaboration between the companies

-Rentex as a first step will be in charge of the DC-Shop flows to their large trucks ability to transport many loads

Design 4 -The shop PI-hubs are

removed from the design -A central PI-hub is introduced in Groningen

-Collaboration between the two parties is still active

-The introduction of a core PI-hub in Groningen is based on order intensity demand of both companies.

Design 5 -Groningen PI-hub is removed

-Hospitals in the north Netherlands are used as PI-hubs

-Customers of both companies are split according to the order intensity demands of both companies

-Since hospitals are located in the most populated areas, the assumption is that also around those areas the most orders will be demanded.

- Vegro will be in charge of the West regions in the north Netherlands and Rentex in the east regions. This separation is proposed based on DC

distances to customers and demand intensity.

Design 6 -PI hubs are introduced at

strategic locations on the core infrastructure system. E.g. where different highways intersect

-The distribution to the PI-hubs will be managed by Rentex and the distribution to customers will be managed by Vegro

-Strategical place of the PI-hub and impact of this decision -Demand is allocated according to fleet characteristics of both companies

(19)

15 2.4.2. Simulation

The tool used to simulate the designs was the Microsoft Excel solver developed by Professor Güneş Erdoğan (Gunes, 2017). The solver is developed as a solution for the vehicle routing problems and its functioning is based on the algorithm of large neighbourhood search (LNS). The choice of simulating the designs was driven by the models variety and complexity (Robinson, 2004). Moreover, using other methods to analyse the designs would have implied higher costs. Therefore, at this stage of research, a simulation methodology is the logical way of analysing the designs.

The solver aims at improving the efficiency of the routes created as well as calculating the times this routes need. Since the issues described by managers are mainly regarded to flows and time management, a VRP solver combined with the PI designs would aid at generating a set of data from which the effect of PI in comparison with the current logistical network could be determined. Moreover, the solver is a close approximation of the current planning systems Vegro and Rentex are using. Based on the interviews with the planners, the distribution planning follows a similar approach to the LNS.

The following logic flow will be applied for the simulations:

1. Introduce the deposit locations and customer locations using postcodes and location finder from Bing

2. Based on the postcodes, coordinates for each address will be generated.

3. Run the solver which will group customers around the closest deposit location and based on other customer characteristics as time intervals, pick-up or delivery.

4. Finally collect data for each truck and aggregate the data for the day.

The main characteristics of the solver are that it allows the users to introduce up to 20 deposit locations and up to 200 customer locations. Some of the parameters used for the solver are described in detail below.

(20)

16

the intervals become narrow and validate the model. The results of the analysis can be found in Appendix 2. Using this method, it was concluded that 4 days would have been enough for statistical significant results. The data converges after day 4 and remains converged for all the days.

In total, a number of 60 instances were simulated for 6 proposed designs. Each simulation produced a set of data for each vehicle active in the simulation. The data in each simulation was aggregated on four characteristics: kilometres driven, driving time, working time and number of trucks used from the ones available.

In order to run the simulation, the following settings were made: the back-haul option for all the models was activated, that means trucks have to deliver and pick-up orders in the same day; the average speed of trucks was set at 70 km/h for all the model. After the initial setting, the addresses for the distribution centre and customers were inserted as well as the time window in which the customers have to be reached, service time for each of the customers and the whether the item was to be picked up or dropped off. Service times differed between hospitals and private customers as well as capacities. For example, hospitals require higher capacities as well as higher service times. The next step was calculating all possible distances between the parties involved in the model. This setting had to be performed two times due to an error in the software that did not provide all the values for a large series of data. After deriving all the possible distances in the network, the vehicle settings were established. Vehicle settings refer to the capacity of the vehicles, time they start the shift, maximum time allowed in the system as well as the maximum number of kilometres they can drive per day. Once all of the settings of the solver were fixed, the solver was ran and the data received.

(21)

17 2.4.3. Analysis data simulation

For the data analysis, the following methods were used: data comparison method, simple regression model and time variation analysis. The data comparison method was used to assess the difference in total kilometres saved between the original distribution designs and the proposed designs. In order to assess the significance of the differences, a pair t-test was performed. The results and the process of the test will be discussed in the analysis section.

The second analysis investigates the time efficiencies each model produces. In order to do this, a regression model is developed in which kilometres are the dependent variable and driving and working times are the independent variables. The purpose of the models is to observe the impact driving and working times have on kilometres travelled, and compare the models to see which one is most efficient in terms of time impact. In order to test the efficiency, six regression models were derived based on the simulation data.

In order to analyse the driving and working times, a comparison of the driving times between models will be provided. For the analysis, the average driving and working times were derived for all the days in all the designs. The averages are plotted in a box graph that shows the spread of the data, the mean of driving times for each design as well as outliers that stand out.

Furthermore, an analysis of the variation coefficients will be developed. The variation coefficients are used to explain the variation in driving times. A coefficient above 1 will be interpreted as high variation, a coefficient between the values of 0.7 and 1 will be interpreted as moderated variation, while a coefficient lower than 0.7 will signal a low variation (Hopp, 2011). The coefficient is calculated by dividing the standard deviation by the mean of each day from each design. This method is used in order create measurements that are unitless and easier to assess.

3. Interviews

3.1. First round interviews

(22)

18

of flows refers to the performance of trucks in the distribution system and the planning of orders. Other performance characteristics regarding the flows refer to trucks capacity, the overlap of the vehicle routes and inefficiencies regarding delivery and returns. The second operational problem raised by the manager regarded the spatial time of the orders. More specific, due to order delivery time constrain, trucks have to travel on the same location multiple times per day. which creates inefficiencies in delivery and frustrations among the employees. Flow efficiency will be translated into kilometers and spatial time in the system will be regarded as working times and driving times.

During the interviews with the companies, no significant other operational issues were raised concerning PI infrastructure or operational. Managers did not see operational PI implementation as an issue and stated that all the operations could be possible to realize. These operations refer to sharing of trucks, exchange of loads and visiting different customers from their own.

Regarding managerial issues, the topics of IT management, contracts and regulations, ownerships and trust were discussed. Managers did not find IT alignment between companies a critical factor in PI adoption due to the possibilities to explore IT implementation solutions after the adoption. Contracts and regulations were also not mentioned as critical factor for implementation of the PI. Nevertheless, an important factor mentioned by the companies was the adoption of PI by the private sector and not public sector. This topic of private companies adopting first PI was motivated by the intense competition in the logistical markets and the immediate seek for a competitive advantage. A motivation for the adoption of PI proposed by the managers was the opportunities of extending their markets with the aid of other companies involved in the PI system.

The issue of trust was mentioned during the interviews as a main problem at the first stage of PI implementation. Managers perceive trust as the reliability and level of involvement, their collaborative party will dedicate to the cooperation relationship. Managers also mentioned that once trust between different parties is established, other issues that PI implementation poses as software and strategy alignment could be overcome. This was motivated by the interviewee as a main problem due to the differences in company cultures, organizational strategies and human behavioural characteristics.

(23)

19

be trust between partners and the cooperation involvement between parties since the companies have differed organizational strategies.

3.2. Trust creation in a PI environment

Since the companies regarded trust as a critical factor in the adoption of PI, a trust implementation plan is proposed. The aim of the implementation plan is to create clarity for companies in what steps have to be taken in order to develop trust. The implementation plan is developed according to Pomponi, Fratocchi, & Tafuri (2015) which describe trust creation as a three stage process based on shared assets and aims. For the scope of this thesys, only the first stage will be discussed because the thesys analyses the first implementation of PI as a first step. More research is needed in the area of PI trust in order to conclude that the trust framework for horizontal collaboration is similar for the PI trust framework.

Stage 1 Operational

Collaboration Tasks Time Frame

Shared assets

Data Share data about planning and resources available

1st Year of collaboration Information Plan meetings were information

is exchanged in order to create operational alignment

Fleet/carrier Share trucks and create load exchange points

Aims

Cost reduction Become more efficient in delivery by mixing loads Quicker response Exchange distribution plans in

order to be able to react fast Joint distribution and

flows consolidation

Create standard joint flows in the beginning to be able to

consolidate them

Improved productivity Develop common KPI in order to measure and improve

productivity

(24)

20

companies and logistical processes has to the shared in order for the companies to create alignment in their collaboration. Finally, loads have to be exchanged in the system and trucks capacity shared in order to create cost reductions and quicker response for their customers. In the discussion section, the results from the interviews regarding the framework can be view for further insights about the manager’s perception about trust.

3.3. Second round interviews

Based on the results from the design simulations, a second round of interviews was conducted with the innovation and logistical managers of Vegro as well as with the logistical manager from Rentex. The goal of the interview was to assess their willingness to implement a PI network.

During the interviews with the logistical and innovation manager of Vegro, the managers had a positive and optimistic attitude about the results. However, they had some remarks regarding the general setting of the designs. The first issue raised was the difficulties of managing all the shops and hubs by themselves. This was from a financial perspective and operational perspectives. Vegro is active towards the implementation of PI, but they would not consider possible to manage, develop and operate the network by themselves. Regarding the issue of trust, managers understood the framework and were willing to implement it in order to develop a better relationships with Rentex, nevertheless, the resources involved and the uncertainty about the outcome of the relationship made them reluctant.

The hubs in the hospitals were viewed as a good framework for them, nevertheless, comments about the high costs and bureaucracy of operation such hubs were made. Regarding the planning of the loads in the system, Vegro is willing to cooperate with other companies but they would not consider managing the network by themselves, even if the benefits are higher that the costs. The reluctance of managing the PI network was based on the misalignment between their core operations and the operations they would have to perform in the PI environment.

(25)

21

and it would not make companies dependable on each other. Rentex would be willing to implement as soon as possible the PI if the network is already in place and a third party would manage it.

In conclusion, both companies were interested in implementing PI according to the results but only under certain circumstances. The network should be already in place, at a basic level, and a third party should manage it. Both companies seemed interested in the possibilities of extending their markets with the help of the PI logistical collaboration.

4. Experimental Designs

4.1. Current distribution networks

For the representation of the current distribution network of Vegro, original data from Vegro was used. A representation of the network can be observed in Figure 5. In order to validate the representativeness of the emulation, the logistical manager from Vegro confirmed the flow behaviour and customers distribution locations.

(26)

22 Figure 5 Distribution Network Vegro 12 September 2017

(27)

23 Figure 6 Distribution Network Rentex

4.2. Proposed alternatives PI network designs

In the development of the designs, PI elements are introduced incrementally. In the first design, the shops of Vegro are transformed into PI-hubs. The second design uses the framework developed in the second design plus the introduction of Rentex collaboration. In the third design, one main hub is used and the flows of the companies are extended to a more complex collaboration. In the fourth design, hospitals are used as hub locations while the flows of the two companies are mixed. The last design contains a strategically placed hub location, as well as full collaboration between the companies.

(28)

24

The orders will be assigned to the trucks by the planning department, and each truck will return at the end of the working day back to the base in Nieuwleusen. Figure 7 is a simplified representation of the current distribution model used by Vegro. The colour blue underlines that all the flows, central distribution centre and customers belong to Vegro.

C1 C2 C3 C4 C7 C8 C6 C5 DC Nieuwleusen

Original Design Vegro

Figure 7 Design 1- Current distribution network Vegro

A new network redesign solution is proposed in order to facilitate the collaboration between the parties. The proposed network design uses the shops of Vegro as cross-dock point for the orders in each region. A large truck will bring the orders to each shop of Vegro in the north regions and from the shops locations, smaller trucks will deliver orders to the customers in the area of the shop. This design is proposed because there is already a distribution flow to the shops which is not fully utilized. Moreover, shops have the capacity to accommodate the limited number of orders that are distributed each day. A benefit is that the location of the shops next to the customers offers Vegro more flexibility in their distribution times.

(29)

25

is the possession of shops. Vegro has a vast network of shops located in all major cities in the north Netherlands while Rentex do not own any shops. The shops of Vegro are supplier by a third party logistics and are acting as smaller distribution point. Most orders from the shops have to still be delivered and only a small 5% from the total orders are pick-up straight from the shops by customers.

Figure 8 presents a simplification of the design described above where larger trucks supply shops with products and orders. Smaller trucks delivered or picked-up orders from the area closely located to the shops. Only Vegro will conduct the distribution activities in this network. The PI concepts introduced in this framework are hubs and the delimitation between PI core network and the local network of the company.

Distribution Network Vegro with shops as hubs

Shop 1 Vegro Shop 2

Vegro

Figure 8 Design 2- Network design Vegro with shops acting as hubs

(30)

26 C1 C2 C3 C4 C5 C6 C7 C8 DC Vegro Nieuwleuzen DC Rentex Bolsward C1 c2 c3 Shop 1 Vegro Shop 2

Vegro Shop 1 _Vegro

Figure 9 Design 3-Vegro shops acting as hubs and Rentex distributing order to shops

(31)

27

Figure 10 shows an overview of the two combine networks, where the loads of Rentex for the east regions are picked-up and delivered by Vegro and the loads of Vegro for the west regions are picked-up and delivered by Rentex.

Distribution Networks Vegro-Rentex with Cross-Dock

Groningen C2 C3 DC Bolswald C1 Groningen Cross-Dock (Hospital)

Figure 10 Design 4-Combine Rentex-Vegro Design with cross-dock location in Groningen

(32)

28

In Figure 11 it can be observed the closed overlap between the hubs (green) and customers of Rentex (yellow) and the flows from the two companies.

C1 C2 C3 C4 C5 C6 C7 C8 DC Vegro Nieuwleuzen DC Rentex Bolsward C1 c2 c3 Hub 1 Hub 2 Hub 3 Hub 5 Hub 4 c4 C5

Figure 11 Design 5- Hospitals acting as hub locations

The last network design aims to bring the networks as close as possible to a PI network design. This is done by the use of the main infrastructure structure-highways. On the nodes where highways in the north of the Netherlands intersect, three hubs are proposed. There nodes represent the entrance and exit from the PI network. Once orders arrive at this point, they will be delivered with the first available truck to the closest location. In order to be able to simulate this scenario, the assumption that Vegro will take care all the loads from the hubs exit is proposed. Therefore, Rentex will bring all the loads to the hubs and from the hubs; Vegro will bring all the loads in the system.

(33)

29 C1 C2 C3 C4 C5 C6 C7 C8 DC Vegro Nieuwleuzen DC Rentex Bolsward C1 c2 c3 Hub 1 Hub 2 Hub 3

Figure 12 Design 6- Hubs located on the core infrastructure

In summary, the designs presented above are introducing and developing incrementally PI mix network structured up to the last design, which represent an first PI structure implementation design. There can be many variations of the models since the structure of a PI system is dynamic. Nevertheless, a core framework incorporating core elements of PI can still represented by simplifications and assumptions. The next section will discuss the simulation of the designs and how they behave.

5. Results

A detailed representation of the simulation results can be viewed in Appendix 3. Based on the results from the simulation, an analysis was developed regarding the kilometres covered by each design and the efficiency of the working and driving times. The total kilometres for each design were aggregated in order to observe which design is more effective in that matter. Table 7 provides information about the total kilometres. As a first conclusion, design 4 is the weakest performing from all the designs followed closely by the original design, number 1.

(34)

30

4 has the highest number of kilometres used followed closely by the current design, design 1.The best performing designs in term of kilometres are design 3 and design 5.

Regarding design 4, the explanation of its inefficiency is Vegro’s incapability of coping with its own demand plus Rentex demand, which creates inefficiencies in delivery. Trucks of Vegro have to perform multiple transports in order to cover the overcapacity. Concerning the original model design 1, the high number of kilometres can be based on the long distances trucks have to travel in order to reach the customers. The other models all show improvements regarding the travelled kilometres per day.

The low number of kilometres from design 3 are due to the high number of PI-hubs that diminish the number of kilometres travelled between PI- hub and customers, as well as the use of collaboration and specialization in routes between Rentex and Vegro. Vegro will specialize in delivering to customers while Rentex will specialize in suppling PI-hubs with products.

D1 D2 D3 D4 D5 D6

Total KM 36596,33 33108,07 20078,47 39169,95 24981,41 28456,34 Table 7 Total kilometers covered by designs

In table 8, a kilometres based comparison is presented. A paired t-test was performed in order to assess the significance in the difference between the means. Moreover, the Persons r was computed to assess the effect size.

Design 2, 3 5 and 6 present positive results regarding the kilometres saved. Design 4 does not produce any kilometre savings. The highest savings impact is produced by model 3, where shops are used as hubs. This high number of shops used as hubs and the low distances trucks have to travel from shops to customers can explain this saving impact. Design 6 can be regarded as a second best scenario, taking into account the impact of saved kilometres and the effect size. Therefore, it can be concluded that models 3 and 6 have a high-saving kilometre impact of all networks.

Total Km saved Savings % Avg saved per day T stat Significance Persons r

D1-D2 3488,26 10% 232,55 6,23 p<0,01 0,74

D1-D3 16517,86 45% 1101,19 29,51 p<0,01 0,74

D1-D4 -2573,63 -7% -171,58 -4,04 p<0,01 0,57

D1-D5 11614,92 32% 774,33 11,37 p<0,01 0,08

D1-D6 8139,99 22% 542,67 12,13 p<0,01 0,45

(35)

31

Next, the impact of the driving times and working times will be assessed. This was done with the help of regressions. The dependent variable used for the regression of the designs was the travelled kilometres and the independent variables used were driving and working time. The thesis investigates the impact driving and working time has on the number of kilometres driven each day. Moreover, the efficiency of each design regarding the independent variables is being assessed.

D1 D2 D3 D4 D5 D6

Intercept 508,91** -554,62* -515,16** -161,23 -628,92** -1076,98* Driving Time 36,39** 72,10** 72,29** 62,60** 85,90** 58,73**

Working time 8,30** 1,38 1,25 3,55* -1,84 11,46*

Table 9 Results regression models-** significant at p<0.01;*significant at p<0.05

In Table 9, the general results of the regressions are presented. The complete results of the regressions can be viewed in Appendix 4. For the assessment of the driving and working time, the coefficient of each regression was investigated. From the results it can be noticed, that all the designs perform better in term of driving time comparing to the original design. The best performing design, in terms of driving time, is model 5 and 3. That means that an increase of one driving hour will cover 85.9 more kilometres in comparison with design 1, where only 36.39 kilometres will be covered.

In terms of working time, only models 4 and 6 are statistically significant. Moreover, model 4 performs worse in terms of efficiency comparing with the original design. However, design 6 shows a slight increase in the efficiency of working times in comparison with original model.

(36)

32 Figure 13 Average driving times

In Figure 13 a comparison representation between the designs driving times can be seen. From the graph it can be noticed, that the average times decrease for the designs incorporating more PI elements. The best performing design in term of driving time is design 3. Also the spread of the driving times across the data is narrower comparing with the original design, design 1.

Figure 14 Average working times

(37)

33

the longer service time Rentex has in distributing the goods. Nevertheless, the mean of the average working times is lower or equal with the mean of design 1

Figure 15 Variation coefficient driving time

Regarding the variation of the driving and working times, it can be said – based on Figure 15 and 16 - that both have low variation times. However, design 2, 3 and 5 perform worse than the original design in terms of driving variability. Nevertheless, design 5 and 6 perform better in terms of variation regarding the working times.

(38)

34

6. Discussion

First issue discussed in this thesis was trust. Trust appeared to be an important aspect in adopting the P in this setting. There is no literature regarding trust in the PI field. Nevertheless, a trust implementation framework was used from the closest concept to PI, namely, horizontal collaboration. The framework proposes certain steps to deal with the issue of trust, however not all the steps fit the setting of PI due to the different contexts (Pomponi, Fratocchi, & Tafuri, 2015;Simmer, Pfoser, Grabner, Schauer, & Putz, 2017). Horizontal collaboration aims at a long time strategical partnership by following a three step implementation framework. In a PI environment, such a strategic partnership would be difficult due to the multitude of companies involved. Thus, trust in PI is different. Another aspect about trust is the need of a third party partners to manage the PI network. Managers of both companies stated in the interviews that the trust issue would be easier to deal with should a third party organized and manage the PI network and a guarantor of trust were between the members of a network. The importance of a third party company managing the PI network extends to more topics as fair loads allocation, flows planning between companies and costs and revenue management. The topic of a third party company managing the PI implementation is not discussed in the PI literature and might be the key for implementation of PI since it can act as a mediator and facilitator between parties.

Regarding the proposed PI design that incorporate PI-elements, it can be concluded that the classical logistical distribution networks settings have to be changed in order to accommodate PI-elements as PI-hubs and PI-flows. This is necessary in order to facilitate the exchange of loads. The adoption of the classical network distribution varies according to type of customer, trucks fleet and logistical strategy.

(39)

35

more companies can be added to the structure, since is quite an adoptive network due to the two levels it has: main highway network hubs and hubs customers. Another conclusion based on design 3, is that more PI-hubs can increase the efficiency of time and kilometres of a network. A downside of this design however is the costs involved given the multitude of PI-hubs.

Regarding the performance of design 5 that incorporated PI-hubs at the location of hospitals, a reduction by 32% in the covered distance can be seen. The reduction and efficiency of the design can be based on the introduction of more PI-hubs. Design 4 and 5 have the same characteristics in flows. However, due to having more PI-hubs location, design 5 is more performant than the original design, while design 4 not. Design 6 followed a strategical placement of the PI-hubs. From its performance, it can be noticed that only a 22% reduction of kilometres was reached. Since only 4 PI-hubs were introduced in this design, it can be concluded that the design performed the best in terms of ration number of PI-hubs introduced.

7. Conclusion and Limitations

Due to the promised benefits of the Physical Internet (PI), its implementation into the current logistical network can provide organizations a competitive advantage in the industry. This thesis has investigated what incentives will motivate managers to implement the PI of companies in the logistical healthcare sector. Moreover, the thesis also investigates the impact of PI on logistical topics as time spatial management and flow efficiency. The thesis was structured on three operational and managerial KPI. Based on these KPI, six network designs were developed for the distribution networks of Vegro and Rentex.

(40)

36

The number of PI-hubs as well as their strategical location have an important role on the efficiency of a PI distribution network. Moreover, the number of available trucks and their characteristics of each company pay an important role in how the PI is structured.

Regarding the willingness of managers to implement PI, it has been found that the need of a third-party PI manager and organizational is critical to deal with managerial issues. Such issues were identified as trust, as operational issues as well as planning of the collaborative distribution and managing the PI-hubs and flows. Managers seem to be willing to implement PI if all the structure is already in place and if a third party will be in control of it. A public-private third party management was preferred by managers due to the magnitude and complexities of a PI network.

In the current thesis, different placement of PI-hubs, number of PI-hubs and different level of collaboration were used. This can be a limitation of the study since the elements were not introduced on the same structure of the current design but the current design was adopted to different PI-settings. This might have created a large variation in how the designs behaved. Nevertheless, the purpose of the thesis was to prove that PI-elements, even in different scales, will improve the current classical network design.

The thesis did not investigate the impact of the designs on costs due to the lack of data about operational and handling costs as well as profits assigned to orders. However, the topic of costs and profits of the PI-networks is important for managers in their strategical decisions. This is a limitation of the study since it lacks such a cost analysis.

Since PI is regarded as an interconnected network for multiple organizations, the use of only two companies can be regarded as another limitation. Moreover, the location where the companies are located can be also viewed as a limitation because the analysis is limited to a specific area that does not cover large distances. Both companies are active in the field of health care logistics, which limits the scope of the analysis only to this filed of operation.

(41)

37

(42)

38

Bibliography

ALICE. (2017). Alliance for logistic innovation through collaboration in Europe. Zaragoza: ENIDE. Ballot, E., Montreuil, B., & Meller, R. (2014). The physical internet: The network of logistics networks.

La documentation Francaise.

Greenhalgh, C., & Rogers, M. (2010). Innovation, intellectual property, and economic growth. Princeton University Press.

Gunes, E. (2017). An open source Spreadsheet Solver for Vehicle Routing Problems. Computers &

Operations Research , 62-72.

Hakimi , D., Montreuil, B., Sarraj, R., Ballot , E., & Pan, S. (2012). Simulating a physical internet enabled mobility web: the case of mass distribution in France. 9th International Conference

on Modeling, Optimization & SIMulation-MOSIM'12, (pp. 10-p).

Hopp, W. (2011). Supply chain science. Waveland Press.

Iacovou, C., Benbasat, I., & Dexter, A. (1995). Electronic data interchange and small organizations: Adoption and impact of technology. MIS quarterly, 465-485.

Karlsson, C. (2016). Research Methods for operations management. Routledge. Montreuil, B. (2009). Physical Internet Manifesto. Technical Report.

Montreuil, B., Russell, M., & Ballot, E. (2013). Physical Internet Foundations. In T. Borangiu , A. Thomas, & D. Trentesaux, Service Orientation in Holonic and Multi-Agent Manufacturing and

Robotics (pp. 151-166). Berlin: Springer.

Pan, S., Ballot , E., Huang, G., & Montreuil, B. (2017). Physical Internet and interconnected logistics services: research and applications. International Journal of Production Research, 2603-2609. Pomponi, F., Fratocchi, L., & Tafuri, S. R. (2015). Trust development and horizontal collaboration in

logistics: a theory based evolutionary framework. Supply Chain Management: An

International Journal, 83-97.

Prakash, A., & Deshmukh, S. (2010). Horizontal collaboration in flexible supply chains: a simulation study. Journal of studies on Manufacturing, 54-58.

Robinson, S. (2004). Simulation: the practice of model development and use. Chichester: Wiley. Simmer, L., Pfoser, S., Grabner, M., Schauer, O., & Putz, L. (2017). From horizontal collaboration to

the Physical Internet–a case study from Austria. International Journal of Transport,

Development and Integration, 129-136.

Sternberg, H., & Norrman, A. (2017). The Physical Internet–review, analysis and future research agenda. International Journal of Physical Distribution & Logistics Management, 736-762. Yang, Y., Pan , S., & Ballot, E. (2017). Innovative vendor-managed inventory strategy exploiting

interconnected logistics services in the Physical Internet. International Journal of Production

(43)

39

Appendix 1

Interview Protocol

Beginning script:

Introduction:-Hello! My name is Romeo Racu, I’m an Supply Chain Management student from the University of Groningen. I am here to learn about what incentives might trigger managers to implement PI. Thank you for taking the time to talk with me today. The purpose of this interview is to learn in what areas should further research of PI focus in order that managers will be willing to implement PI. There are no right or wrong answers, or desirable or undesirable answers. I would like you to feel comfortable saying what you really think and how you really feel. If it’s okay with you, I will be tape-recording our conversation since it is hard for me to write down everything while simultaneously carrying an attentive conversation with you. Everything you say will remain confidential, meaning that only my professor and I will be aware of your answers the purpose of that is only so we know whom.

1. Could you tell me something about yourself? How did you decided to work in the field of logistics?

2. What is your experience in logistics and what is your general felling about it? 3. How do you view the future of logistics?

4. What can you tell me about the Physical Internet?

If concept is not known describe. Physical Internet is an open global logistics system founded on physical, digital and operational interconnectivity through encapsulation, interfaces and protocols. It is a perpetually evolving system driven by technological, infrastructure and business innovation.

5. How about Horizontal collaboration?

If concept is not known describe. Logistical horizontal collaboration is negotiated cooperation between independent parties in a system for the sharing of risks and improving the efficiency of their operations.

6. What are the most important thinks when you think about the efficiency of logistics? 7. Does your company have a future innovation plan for the logistic department and how

urgent is that plan?

8. Under which circumstance would you consider implementing PI?

9. Which benefits should PI bring on top of your current system in order to implement it? At which level?

 Increase in company productivity-increase profits

 Reduce cost of non-core activities as personnel and consumables.  Increase customer service

(44)

40

o Fuel consumption o CO2reduction

o Logistical costs-loading, unloading  Increase filling rate trucks

10. What do you view as impediments in implementing PI?  Trust partners

 Policies

 Technical issues- software  Different load types  Truck capacity  Schedules

Discuss more about which benefits would trigger an implementation of PI.

11. What factors in the efficiency of logistics you consider critical and non-critical for implementation of PI?

12. Could you tell me more about the process of implementing a new logistical system in your company?

13. Do you find it important to collaborate with other companies in the field of logistics? 14. What do you think is the first step in the implementation of PI?

15. What can be some drawback in implementing PI in the near future?

16. Would a simulation aid in the decision process regarding PI implementation? 17. What type of evidence about the efficiency of PI you would need to try it?

End interview:

(45)

41

Appendix 2

Replications Mean time in system Cumulative mean Standard Deviation Lower interval Upper Interval % Deviation

1 2693,057 2693,06 n/a n/a n/a n/a

(46)

42 Replications Mean time in

system Cumulative mean Standard Deviation Lower interval Upper Interval % Deviation

1 2368,38 2368,38 n/a n/a n/a n/a

(47)

1 1499,74 1499,74 n/a n/a n/a n/a

(48)

1 2465,978 2465,98 n/a n/a n/a n/a

(49)

1 1442,444 1442,44 n/a n/a n/a n/a

(50)

1 1884,879 1884,88 n/a n/a n/a n/a

(51)

(52)

(53)

(54)

(55)

51

Appendix 4

SUMMARY OUTPUT D1 Regression Statistics Multiple R 0,99 R Square 0,98 Adjusted R Square 0,98 Standard Error 23,54 Observations 15 ANOVA df SS MS F Significan ce F Regression 2 362290,85 181145 ,43 326,90 3,42778E -11 Residual 12 6649,54 554,13 Total 14 368940,40 Coefficie nts Standard

Error t Stat P-value

(56)

52 SUMMARY OUTPUT D2 Regression Statistics Multiple R 0,98 R Square 0,97 Adjusted R Square 0,96 Standard Error 43,41 Observations 15 ANOVA df SS MS F Significan ce F Regression 2 625693,5 1 312846 ,76 166,01 1,80139E -09 Residual 12 22614,26 1884,5 2 Total 14 648307,7 7 Coeffici ents Standard

Error t Stat P-value

(57)

53 SUMMARY OUTPUT D3 Regression Statistics Multiple R 0,98 R Square 0,97 Adjusted R Square 0,96 Standard Error 43,47 Observations 15 ANOVA df SS MS F Significa nce F Regression 2 625633,0 2 31281 6,51 165,55 1,83049E -09 Residual 12 22674,75 1889,5 6 Total 14 648307,7 7 Coeffici ents Standard

(58)

54 SUMMARY OUTPUT D4 Regression Statistics Multiple R 0,99 R Square 0,99 Adjusted R Square 0,98 Standard Error 24,98 Observations 15 ANOVA df SS MS F Significa nce F Regression 2 493515,3 1 24675 7,65 395,56 1,11276E -11 Residual 12 7485,81 623,82 Total 14 501001,1 2 Coeffici ents Standard

(59)

55 SUMMARY OUTPUT D5 Regression Statistics Multiple R 0,98 R Square 0,95 Adjusted R Square 0,94 Standard Error 52,55 Observations 15 ANOVA df SS MS F Significa nce F Regression 2 652275,0 2 32613 7,51 118,12 1,27581E -08 Residual 12 33131,89 2760,9 9 Total 14 685406,9 0 Coeffici ents Standard

(60)

56 SUMMARY OUTPUT D6 Regression Statistics Multiple R 0,97 R Square 0,95 Adjusted R Square 0,94 Standard Error 42,46 Observations 15 ANOVA df SS MS F Significa nce F Regression 2 377832,4 7 18891 6,23 104,78 2,5245E-08 Residual 12 21636,08 1803,0 1 Total 14 399468,5 4 Coeffici ents Standard