Voice assistants and railway passengers : A user-centered exploration of value creation opportunities in a railway service context

Linköping University | Department of Computer and Information Science Master Thesis, 30 credits | Master of Science - Design and Product Development Spring Term 2020 | LIU-IDA/LITH-EX-A--20/023--SE

Voice assistants and railway passengers

A user-centered exploration of value creation opportunities in a railway

service context

Gustav Backhans

Douglas Driving

Supervisor: Vanessa Rodrigues Examiner: Johan Blomkvist

Linköpings Universitet SE-581 83 013-28 10 00, www.liu.se

i

Copyright

The publishers will keep this document online on the Internet – or its possible replacement – for a period of 25 years starting from the date of publication barring exceptional circumstances. The online availability of the document implies permanent permission for anyone to read, to download, or to print out single copies for his/her own use and to use it unchanged for non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional upon the consent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security, and accessibility. According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement. For additional information about the Linköping University Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its www home page:

http://www.ep.liu.se/.

ii

Abstract

There is a current rising trend of using voice assistants (VAs) to perform tasks in new ways, and various companies are considering introducing them as part of their service propositions. One such company is the railway service provider SJ AB that is interested in understanding how a VA may benefit their passengers. To better understand the utility of a VA in the railway service context, this thesis aims to explore the value creation opportunities that the introduction of such a technology presents. This exploration is done through identifying the potential functionality of such a VA, what value that functionality can create for passengers, and what barriers exist for creating that value. This identification is done through a user-centred research process, during which design probes and experience prototyping were carried out with railway passengers and analysed through qualitative content analysis.

The results show that a VA presents fast, convenient, and intuitive access to a wide set of functions but is hindered by its invisible affordances and the user’s preconceptions. It shares several functions with existing channels, presented in a new way, that span the entire course of the train journey. The functionality and the identified functions primarily create utilitarian values, which connects to the passenger’s pragmatic view of the service. Furthermore, the passengers perceived the VA as an agent able to take over responsibilities and tasks from the passenger, resulting in a peace of mind but also a diminished sense of control. Finally, the railway service context affected what functionality is suitable, what values can be created and what barriers need to be considered. Some of the value creation possibilities and barriers are also contemporary and might change with shifting social norms and further technological development.

iii

Acknowledgements

We would like to sincerely thank SJ AB for the opportunity to conduct this project, as well as for all the knowledge they have shared with us regarding the context of this study. We would specifically like to thank our contact person Emma Christoffersson for her engagement and support of the project, as well as Johan

Beskow for the insights he has given us into SJ’s perspectives on the subjects discussed in this thesis.

We would also like to thank our supervisor Vanessa Rodrigues for her active participation in the project and her endless stream of helpful advice regarding the research area and how to write a proper paper, advice which have all contributed to the quality of this thesis. We also want to thank our examiner Johan

Blomkvist who, time and again, has shown interest in our work, steered us in the right direction and

supported us whenever we have felt lost and confused. We also want to thank our opponents Jeff Pertot and Pernilla Sandén for their feedback on the work, as well as all the friends and family members with whom we have discussed our ideas and methods.

We would also like to thank all the participants of the studies conducted in this project, whose thoughts and ideas the results are built upon, and finally the SJ train staff who assisted us in carrying out research activities on the train and gave us each a tangerine.

iv

Table of contents

Functionality, values, and barriers ... 19

Method ... 21

Pre-study ... 21

Experience prototyping ... 24

Qualitative Content Analysis ... 30

Results ... 32

Functionality ... 32

Values ... 36

Barriers ... 41

Discussion ... 48

Sonja compared to other channels ... 48

VAs and responsibility ... 48

Door to door with Sonja ... 49

The effect of the service context ... 49

The contemporary nature of this work ... 51

Utilitarian values in a practical service ... 52

Method discussion ... 53

Result in a larger context... 54

Conclusions ... 56

Future research ... 56

References ... 58

v

Appendix

Appendix 1 Design Probes ... 60

Appendix 2 Ideas generated in contextual design probes... 61

Appendix 3 Design probe codebook ... 63

Appendix 4 Bodystorming scenarios ... 65

Appendix 5 Prototyping tasks... 67

Appendix 6 Interview Guide ... 70

Appendix 7 Functions list ... 72

Appendix 8 Leroi-Werelds customer value typology ... 75

Figures

Figure 2: Google home smart speaker ... 11

Figure 3: Interacting with Siri through an iPhone ... 11

Figure 4: User commands through GUI vs VUI ... 13

Figure 5: Simplified project structure ... 19

Figure 6: Design probes ... 21

Figure 7: Affinity diagram sorting ... 24

Figure 8: Experience prototyping setup ... 24

Figure 9: A conversation with Sonja ... 26

Figure 10: Training phrases for Sonja ... 27

Figure 11: Sonja does not understand the command ... 28

Figure 12: Sonja identifies the keyword "genomgång" ... 28

Figure 13: The first instruction slides of scenario 1 ... 29

Figure 14: Quote analysis example ... 31

Figure 15: Functions identified through different methods... 32

Figure 16: Interacting with a VA in a service context ... 50

Tables

Table 2: Experience prototyping participants ... 25

Table 3: Experience prototyping scenarios ... 29

Table 4: Function categories ... 33

Table 5: Interaction types ... 34

Table 6: Identified values ... 36

vi

Nomenclature

Table 1: Nomenclature

Phrase Short Definition

Automatic Speech Recognition

ASR The technology making it possible for a computer to transcribe human speech.

Voice Assistant VA Software agents that can interpret human speech and respond via synthesized voices (Hoy, 2018).

Voice-User Interface

VUI An interface that a person interacts with using spoken language (M. H. Cohen, Giangola, & Balogh, 2004).

Graphical-User Interface

GUI “A visual way of interacting with a computer using items such as windows, icons, and menus, used by most modern operating systems.” (lexico, 2020).

User-centered design

UCD A broad philosophy of how to approach design with a varied collection of methods and tools, with the common denominator of focusing on users and their needs throughout the design process (Abras,

Maloney-Krichmar, & Preece, 2004). Interactive Voice

Response units

IVR Earlier applications of VUIs, mostly in the shape of telephone response systems, when calling service providers such as banks.

7

Introduction

Since the birth of the computer, the relationship between them and us humans has been strained as it was expected of us to learn their language and not the other way around. How we have managed to communicate so far have mostly been through keyboards, and if the sound of the industrial revolution was the steady chuffing of a steam engine then the noise of the digital revolution must be the smattering of keys. But technological development has come far, and the computers are now ready to instead listen to us. If you want to speak to one you can begin by checking your pocket. There is a good chance that in your pocket there is a phone, and that in that phone there is a voice assistant (VA) ready to listen to you. Using VAs to perform tasks in new ways is a rising trend (Hoy, 2018). An example pointing to this trend is that 24% of the U.S. population today owns a smart speaker such as google home or amazon echo (NPR, 2019), and are using it for things like searching the web, control home automation devices and manage their calendars (Hoy, 2018). Although VAs are becoming increasingly popular, a clear description of what values they can create for a customer is lacking. The lack of such knowledge makes it troublesome for companies to understand what the technology is useful for and why they would want to introduce it. A better understanding for the value of VAs would help companies and organizations in determining whether it can be beneficial for their customers, leading to a higher potential in achieving competitive advantage (Robert B. Woodruff, 1997).

There are ideas about introducing VAs into various service processes to improve them from a customer perspective. The railway service company SJ AB is interested in finding out what a VA could do for their customers. What values a VA could create in the specific service context (i.e. train travelling) is however unknown. In order to make an informed decision about whether (and how) SJ and other railway service companies should introduce a VA in their service offering, they need to be aware of what values the VA could create for the users, and what barriers exists in creating those values.

Purpose

The purpose of this thesis is to explore the value creation opportunities of introducing a voice assistant in the service context of train travelling. This is done by investigating the possible functionality of a voice assistant in the train traveling context, what values it could assist in creating and what barriers exist in creating those values.

Research Questions

The study aims to answers the following research questions:

• What functionality does a voice assistant offer in the railway service context? • What values can a voice assistant create for railway passengers?

8

Delimitations

The following delimitations are made in the study:

- The study is being done in a Swedish context. This both because of the project being a

collaboration with SJ, and for the ease of access to Swedish passengers.

- The study is not exploring how functions can be technically implemented. This delimitation is

made as including technical aspects, such as modern AI and machine learning, to this project would make the scope too wide.

- The contact with users is taking place over skype. This delimitation is made to limit the spread of

Covid-19.

Thesis structure

The thesis is structured into the following chapters:

Background

Describes the background of the study and the brief given to the authors by the project owner.

Theory

Prior research and other important literature related to VAs, user-centered design, service design, prototyping and value creation.

Approach

Contains a high-level description of the approach taken to achieve the results, as well as the definitions used for functionality, value, and barrier.

Method

A description for the methods used in the process, including a pre-study, experience prototyping, interviews, and qualitative content analysis.

Results

The results in the form of a description of the functionality of the VA and a list with values and barriers identified.

Discussion

A discussion of how these results relate to prior research, what stands out, what contributions they make to the field, how the results were impacted by the choice of method, the ethical implications of the work.

Conclusion

9

Background

This study was conducted in collaboration with the train service company SJ AB who were interested in learning what value a voice assistant could add to the customers of their train services.

On their website, SJ writes: “SJ is a Swedish travel partner that offers sustainable train travel, both independently and in collaboration with others. Every day, 140,000 people choose to travel on one of SJ's 1,200 departures from 284 stations”. The company was founded in 1856 and is today the market-leading railway company in Sweden. It is owned by the Swedish state and is tasked with operating profitable passenger rail services. SJ primarily focus on long-distance travels between the major cities in Sweden. (SJ AB, 2020a)

SJ strives to integrate new technologies into their service processes. They have identified a trend of voice-related technologies becoming increasingly common as a tool for communicating, gathering information, and interacting with systems. To meet this trend, SJ wants to be at the forefront of utilizing the technology. To be able to do this, they want to understand how voice technology can be further integrated into the experience of railway passengers before, during and after a train ride in a way that adds value. This led to the idea for this project which was given to the authors as a proposed master’s thesis. The purpose of the proposed project, as stated in a project brief provided by SJ, was to learn how passengers could benefit from voice-related technology in the future, and how that technology could add value both short-term and long-term. The work presented in this thesis took that project brief as a starting point.

10

SJ’s voice assistant

In 2018, SJ participated in the launch of Google Assistant in Sweden. This participation led to the development of Sonja, SJ’s voice assistant. To access Sonja, a user starts their google assistant app and says the key phrase “Prata med SJ” (speak to SJ). A snapshot of a conversation with Sonja is shown in Fel!

Hittar inte referenskälla..

Figure 1: Speaking with SJ through Google Assistant

At its current stage, a user can speak with SJ to: (SJ AB, 2020b) ● Get information about when their next train leaves ● Find out if a specific train is on time

● Save their favorite trip so they can ask about it quicker next time ● Learn fun facts about SJ and trains

Currently, the VA is still in an early development stage and provides user with only a narrow set of functions. SJ, however, want to investigate the potential of further developing the VA by understanding how it could contribute to the service experience of their customers.

11

Theory

This chapter presents theory used in the thesis. First, voice assistants (VAs) and their strengths and weaknesses are described. After that theory related to the approach, including user-centered design, service design, and prototyping, is presented.

Voice Assistants

VAs are “software agents that can interpret human speech and respond via synthesized voices”. In the last couple of years, several inexpensive VAs have been brought into everyday use. Examples of popular VAs are Apple’s Siri and Google Assistant, which are embedded into smart speakers (Figure 2) and smartphones (Figure 3). When used on a smartphone, Siri and Google Assistant can be interacted with through both voice and text chat. Compared to earlier voice-activated technologies, a VA can respond to a larger variety of commands as it is connected to a server and therefore does not have all commands programmed from the beginning. (Hoy, 2018)

Figure 2: Google home smart speaker

Figure 3: Interacting with Siri through an iPhone

With a VA, a user can do things like get responses to different questions, control home automation devices and manage basic tasks such as calendar apps. Additional features can be added to the assistant, enabling the control of other programs and apps. These additional features are usually developed by third party actors making it possible for the assistant to communicate with their apps or systems. (Hoy, 2018)

12

2.1.1 Voice-user Interfaces

A voice-user interface (VUI) is an interface that a person interacts with using spoken language. The reason behind why a company should introduce a VUI is that it can save money, improve the potential to reach customers, extend brands, solve problems and increase customer satisfaction. (Cohen et al., 2004). VAs, such as Google assistant and Apples Siri, are examples of VUIs. Other examples of applications of VUIs include TV controls, handwriting assistance, and speech transcription.

2.1.2 Interacting with a VUI

Interaction with VUIs is primarily done by voice, speaking into a microphone placed in a cell phone, a smart device etc. A user interacting with a VUI typically expects to be able to use natural language, meaning that they can use “ordinary linguistic constructs in a conversational manner” where “each party’s utterance sets the context for interpreting subsequent utterances” (Cohen & Oviatt, 1995). The users expect to be able to talk normally.

A VUI system can be designed to ask specific questions to a user that they answer. Although this makes user responses predictable and easier to interpret, it also limits the possibilities for interaction. A different approach is therefore to base the interaction in human conversation in which dialogue enhances the richness of the interaction making it possible to convey more complex information. The VUI can be designed for tasks with a large range of complexity, from scripted prompts that the user follows to agent-based models in which operations are monitored and executed in a dynamically changing world. (Allen et al., 2001)

The structure of the terminology behind the VUI´s conversation abilities is not standardized. For example, Cohen and colleagues (2004) describes that a VUI is built up of prompts (recorded or synthesized speech played to the user during interaction), grammars (words, sentences and phrases a user can say in response to the prompts), and dialog logic (the action the system is taking in response to user inputs). Google presents another view, talking about Intents, responses and prompts in their development platform of Google assistant, Dialogflow. This platform was the one used to construct the prototype and will be more clearly described in 4.2.4 Creating the prototype.

13

2.1.3 VUI advantages

For a user, the advantages of a VUI is that it can be intuitive, efficient, ubiquitous (i.e., the user does not have to actually be at the device to interact), enjoyable, and hands/eyes-free (Cohen et al., 2004).

Figure 4: User commands through GUI vs VUI

In a graphical-user interface (GUI), a designer can either provide users with simple commands that requires them to use long sequences of commands when doing complex tasks, or to provide more higher-level commands in all situations which tends to make the interface become cluttered with options (Allen et al., 2001). In natural language interaction, however, spoken commands can “shortcut the navigation of a menu hierarchy to invoke known commands” (Cohen & Oviatt, 1995), illustrated in Figure 4 above. This is because conversational interaction provides the opportunity for users to state what they want in their own words (Allen et al., 2001). Results from several studies has indicated that using VUIs have a strong potential in increasing efficiency that stems from the use of spoken language communication (Cohen & Oviatt, 1995). For example, Wang & Van De Panne (2006) found out that using a VUI to create animations can be faster than using a GUI, especially when doing complex tasks that required pushing several buttons in the GUI. Thus, one of the advantages of using a VUI is that it is much faster (Dasgupta, 2018).

A good VUI with a large vocabulary should also require minimal training in the system domain (Cohen & Oviatt, 1995), as users can utilize their existing knowledge of the language during interaction. In this way, using voice as a means of interaction can be seen as more intuitive as it is more natural for the user (Dasgupta, 2018).

A voice interface can also allow for hands-free and eyes-free interaction. Situations where a voice interface is favored are therefore when a user’s eyes or hand are busy performing other tasks, as using their voice to interact allows them to focus their attention on the task at hand instead of breaking away to use a keyboard or a screen. Many studies have shown that using voice interfaces in domains including hands/eyes-busy tasks leads to higher productivity and accuracy. (Cohen & Oviatt, 1995) Situational impairments that leads to voice-interaction being the most efficient option is, for example, when having limited visual attention or hand availability while driving (Corbett & Weber, 2016). Other suitable applications found for VUIs are when options for using a screen and other input devices are limited, when the user is suffering from disabilities, and when the subject matter of interaction is pronunciation, such as in a software for learning new languages (Cohen & Oviatt, 1995).

14

To summarize, the strengths of VUI’s are that they allow for eyes and hands-free interaction, that they

shortcut the navigation of hierarchical menus making them more intuitive and efficient, and that they

are easy to learn.

2.1.4 VUI Weaknesses

The strength of a GUI is that its use of visible icons and menus allows the user to easily see and discover what possible actions they can take, and to learn the system through exploration (Norman, 2010). The property of how easy it is for a user to understand what a system can do has been referred to as discoverability (Corbett & Weber, 2016). Yankelovich (1996) suggests that discoverability is the most challenging issue of designing VUIs. This problem of discoverability stems from the user having an inaccurate mental model of the systems capabilities and from the system not being able to visually present its affordances and metaphors as guidance (Corbett & Weber, 2016). In other words, the functionality and boundaries of the system is hidden (Yankelovich, 1996).

Additionally, this might make the user not know what words or phrases the system can interpret (Cohen & Oviatt, 1995). Multiple failed attempts to find the right query or command for the system can lead to frustration and disillusionment (Cohen & Oviatt, 1995). Depending on what words a user discovers that they can successfully use, they might start expecting that the system will also understand other, similar words and assume they can say things the system might not actually support (Yankelovich, 1996). A solution to this problem is for the system to spell out every prompt the user can give it, but the problem with this approach is that the speech output is slow and temporal, costing time and making it hard for a user to remember (Yankelovich, 1996). Although this problem exists, exposure to a VUI tends to make users better at knowing what they can say to it (Yankelovich, 1996). Research on VUI has found that, because of a lack of a visual interface, users are more prone to taking a trial and error approach in learning to use a system using a VUI (Corbett & Weber, 2016). After just 10 minutes of interacting with a system, the amount of errors might be cut in half as users changes their vocabulary, grammar and manner of speaking to fit the capabilities of the system (Yankelovich, 1996).

To summarize, the drawbacks of VUI’s are they cannot visually show what a user can do, or how they can do it.

User-centered design

User-centered design (UCD) is a broad philosophy of how to approach design with a varied collection of

methods and tools, with the common denominator of focusing on users and their needs throughout the design process (Abras et al., 2004). The term was popularized by the book User-Centered System Design: New Perspectives on Human-Computer Interaction (Norman & Draper, 1986) and was a reaction to the complexity of the new technologies at the time leading to new design challenges (Abras et al., 2004). Donald Norman would go on to write the book The design of everyday things (1988), an international bestseller with the core message that more needed to be done to design the world around us for humans and their needs. By designing a product with a user centered approach, more effective, efficient and safer products could be created that are better accepted by its users (Preece, Rogers, & Sharp, 2002 trough Abras et al., 2004).

The early tools of UCD were focused on understanding the users through interviews and observations early in the design process and gather their opinion through usability testing of a prototype in the later stages (Abras et al., 2004). But designers have been moving closer and closer to the users, sparking a family of

15

approaches to design such as participatory design, experience design, service design, UX design and more, each bringing in a new set of tools and perspectives (Sanders & Stappers 2012). Mapping or comparing these domains is a daunting task since they overlap and borrow from each other, while others exist purely as a response to some other field. As an example of its difficulty, Stickdorn, Hormess, Lawrence, & Schneider (2018) notes in the beginning of their book This is service design doing:

“The splitters will talk about the differences between service design, design thinking, holistic UX, user-centered design, human-centered design, new marketing and even more. The lumpers will point out that these approaches have far more in common than they have differences, and suggest names matter far less than the principles that these practices all share. [...] Honestly, we don’t care what you call it, as long as you are doing.”

Still, there is a common trend in that design is moving from the making of stuff (buildings, products, images) to the making of stuff for people in the context of their lives. The problems that design tackles are becoming increasingly complex, involving more stakeholders using more advanced technology. This increase in scope has led the design effort to lend resources and people from other domains such as sociology and psychology, building multi-disciplinary teams that could use their varied perspectives to explore this more complex context. (Sanders, 2012) These multi-disciplinary teams then engage the users and the context with a broad set of tools ranging from strategies to more hands-on methods.

A common characteristic of these design processes are that they are iterative, meaning that the steps on the way are not thought of as a linear process, but instead a series of attempts that you learn from, applying your knowledge again and rethinking your approach. These iterations can be large (redoing a design from its beginning) or small (quickly trying out a sketch on some paper). By iterating you will more quickly do the inevitable mistakes that you can learn from, increasing efficiency. (Stickdorn et al., 2018) Another trend among these different design approaches is the involvement of users as co-creators in the design process. This was also born from necessity since design is tackling more advanced problems (Sanders, 2012). It takes the form of involving users at several stages of the process as experts on their own lives, lending their experience to the designer through activities such as workshops, make toolkits and design probes (Sanders, 2012). The common practice is putting the users in a position where they can make their own solution, giving the designer information both on the what user needs are to be met and how they are to be met. In this thesis, co-creation is viewed as mindset guiding the approach.

Service design

This thesis focusses on a part of this cluster of design approaches, namely service design, which is considered a mindset, process, or toolset. Service design is iterative, multi-disciplinary, user-centered and uses visualizations, prototyping, user research and other activities to meet the needs of all stakeholders within a service, be it users, businesses, or others. Service design focuses on solving the right problem, meaning that it starts by using qualitative research methods to investigate what a customer really needs. Once the right business opportunity has been identified, service design adopts an iterative approach that focuses on doing experiments and using prototypes to generate new insights and ideas for how to solve a problem. (Blomkvist et al., 2018)

Further, the processes of designing a service is, in contrast to a product, usually ongoing as the provider keeps investing in physical assets, training new workforce, and making changes and improvements to front

16

stage service encounters and back stage service support processes (Goldstein, Johnston, Duffy, & Rao, 2002). An organization can thus continue to improve their service by, for example, applying new technology to its value proposition (Stickdorn et al., 2018).

A service consists of components such as processes, people skills, and materials that must all be integrated. Deconstructing a service into its components allows designers to identify elements of the service concept so that they can be measured against customer needs and delivered appropriately. To design a service means to define an appropriate mix of such components, both physical and non-physical. However, it should be noted that having a reductionist view and focusing on the details may ignore the fact that a customer may instead see a service as a “whole experience”, and not a collection of components. Customers tend to have an internal image of the service concept already before they have experienced it, which might come from word of mouth, prior service experiences or other sources of information. The customers image of the service concept creates expectations for it, which needs to be understood and met by the provider. (Goldstein et al., 2002)

Prototyping

Prototyping is an activity commonly used in the process of developing services. Houde & Hill (1997) define a prototype as any representation of a design idea, and an artifact as the interactive system being designed. When it comes to service prototyping, Blomkvist & Holmlid (2010) suggests that it is not defined as a single thing, but rather as a variety of approaches. Interviews with service design practitioners indicates, however, that service prototyping is (1) a central activity to the service design process, (2) about making services visible to learn and communicate about them, and (3) collaborative (Blomkvist & Holmlid, 2010).

To make prototypes generate valuable insights, it is essential to explicitly define its purpose and use different tools depending on the defined purpose (Houde & Hill, 1997). The purpose of service prototypes can be to explore, evaluate and/or communicate ideas and concepts (Blomkvist & Holmlid, 2010). Role prototypes are specified as “those which are built primarily to investigate questions of what an artifact could do for a user” (Houde & Hill, 1997).

Experience prototyping is a form of prototyping that enables people to gain “first-hand appreciation of

existing or future conditions through active engagement with prototypes”. It could be any type of representation that is designed to understand, explore, or communicate what it might be like to engage with a product. One of the basic tenets behind using experience prototyping is that experiences are subjective, and that the best way to understand experiential qualities of an interaction is to therefore experience it subjectively. Experience prototyping can be used to explore possible solutions so that a design team can get inspiration, confirmation or rejection of ideas and conduct a more informed development of the user experience. (Buchenau & Fulton Suri, 2000)

Value in services

To properly understand service as a perspective, it is necessary to have an understanding of value creation that is based on a clear definition of value (Grönroos & Voima, 2013). Here, prior research on value, value creation and value typologies are discussed.

17

2.5.1 What is value?

There are plenty of ways to define value. It can be ethical, ideological, sociological, economical, mathematical etc. Almost every field has a way of separating that which is valuable from things which are not. In service design, there are two literary streams that discuss value creation: service dominant logic (SDL) and service logic (SL) whose main contributors are Stephen L. Vargo and Robert F. Lusch for SDL and Christian Grönroos for SL (Overkamp, 2019). These literary streams present two similar models.

Grönroos & Voima (2013) explains how the creation of value can, on a general level, be explained as a process that increases a customer’s well-being, meaning that they become “better off” in some respect. The created value is commonly seen as a trade of between the benefits that the customer receives and whatever he/she gives up to receive those benefits (Woodruff, 1997). For example, the value of a first-class train ticket could be a trade-off between the additional monetary costs and the benefits the passengers get from sitting in the first-class wagon (such as higher comfort).

In the concept of “value in use”, value is realized at the moment that a service or product is used (Vargo & Lusch, 2004). This implies that value does not inherently exist in a product or service itself, but rather that it is created because of a process. Additionally, in this process users can be seen not just as the receiver of value, but also as a co-creator of it (Vargo & Lusch, 2004). This implies that users are a part of, and can have an impact on, the value creation process.

Furthermore, Vargo & Lusch (2004) argue that users of a service are the judges of the service value. In this view, a provider cannot themselves determine what value they are creating for their customers with a service or product. Value hence become somethings subjective that exist in the mind of the customer. Consequently, the experience of a customer is crucial to the value being created (Sandström, Edvardsson, Kristensson, & Magnusson, 2008). As a customer must determine the value and participate in creating it, a company can only offer value propositions (Vargo & Lusch, 2004). Part of such a value proposition are physical/technical enablers, i.e. products or attributes necessary to create the conditions for a service experience (Sandström et al., 2008).

As customers are the ones who determines what value is created through a product or service, it is important for a company to understand their customers. Vargo & Lusch (2004) explains how:

“The service-centered view of marketing is customer centric (Sheth, Sisodia, & Sharma, 2000) and market driven (Day, 1999) This means more than simply being consumer oriented; it means collaborating with and learning from customers and being adaptive to their individual and dynamic needs”.

Spending time learning about customer needs will therefore help a company in creating value for them. One way that SL and SDL differ are in their view of the product’s role in the service. In SDL, a product is seen as the service it provides, for example the razor essentially being the hair cutting service it provides. While in the SL view, the product is one of the resources integrated in a process, which the customer consume as a service (Grönroos, 2006). In the context of this project, the VA is seen as a technical product that is being integrated at several service moments. While it is clear that the service provided by the VA is central to its existence, understanding the process in which the value is created is fundamental for applying the technology in a beneficial way.

18

An increasing number of services are based on technology, so called technology-based services, and are embedded in cars, phones, radios etc. (Sandström et al., 2008). In these services, it can be more difficult to offer value propositions that create favorable service experiences, as there is no service-oriented employee who will interact with the customer (Sandström et al., 2008). This might soon be turned on its head with the development of new technologies based on artificial intelligence, speech recognition, cloud technology and big data. Wirtz et al. (2018) paints a picture of the future possibilities of service robots, “system-based autonomous and adaptable interfaces that interact, communicate and deliver service to an organization’s customers”. The article claims that the service robots will affect all key actors on a micro, meso and macro level. While this surely sounds promising for VAs, especially when the article notes that text- and voice applications basically have zero marginal costs, there is still a lot of academic research to be done (Wünderlich et al., 2015). Several aspects of the integration of smart objects needs to be explored: how and when customers can co-create value with these objects, what influence the perception of control has on the experience and how the integration of smart devices in the lives of the users affects the users’ attitudes and behaviors (Wünderlich et al., 2015).

2.5.2 Different types of value

To specifically understand what values can be created for a customer, values have been structured into typologies. One of the most well-known typologies for values is the one presented by Holbrook (1994), where values are divided into 8 different types. As the research field and the industry have evolved since this typology was presented, an updated typology that is based on contemporary views on technology, products and services has been presented in which values are divided into positive and negative values (emphasizing the existence of trade-offs). Examples of positive values in this typology include convenience, control, and novelty, while examples of negative values include price, privacy risk, and societal costs. (Leroi-Werelds, 2019)

19

Approach

This project was approached from a user-centered service design standpoint, laying emphasis on exploring the value creation possibilities of the technology within its service context. This was done by involving users in the creation of a VA prototype experienced by users. Insights were collected throughout the project, and the results are presented as a description of the possible functionality of a VA, the values it can create and the barriers to creating it.

Process

In Figure 5 below, a simplified visualization of the approach is displayed. The project is divided into a pre-study and an experience prototype.

Figure 5: Simplified project structure

The pre-study aimed to collect insights about the railway passengers, the railway service context, and what railway passenger would expect of a virtual voice assistant in the context. These insights were mainly collected directly from railway passengers through design probes, but also through discussion based on analyzing customer journeys from SJ, bodystorming during a train trip and ideation based on desk research.

The results from the pre-study were then fed into the formulation of the experience prototype. The prototype was then “experienced” through individual sessions with nine participants over video chat. The transcriptions from the sessions were analyzed qualitatively, inductively producing a list of values and barriers that can be viewed in the Results chapter. The results also include a description of the possible functionality of a VA constructed for the train traveling context.

Functionality, values, and barriers

The results of this thesis are divided into values, barrier, and functionality, of which the latter was further divided into function categories, interaction types and personality. Except for values, the definition of which is brought up in the theory chapter, the rest of the terms were a result of the attempt to formulate the insights gained throughout the project into something that could be discussed, essentially establishing a sort of taxonomy making it easier to discuss the phenomenon that is VAs.

20

3.2.1 Functionality

The function categories are an attempt to describe the nature of the functions a VA can have. The functions essentially describe what Sonja can do for the user. These functions had the simple format of a statement describing Sonja’s abilities in a form that is similar to that of describing the ability of a person, for example “Sonja remembers your seating preferences” and “Sonja can order from the bistro”. This format of describing functions was chosen since it let the authors find and formulate the function fast. It allowed a rather abstract phenomenon to be quickly written down, without explaining the individual steps of the interaction or what is needed to be known by whom. Of note is that this way of describing a function does not address any of the technical aspects that needs to be sorted out for it to work, since this was not within the scope of the project.

To describe how the users interacted with Sonja, interaction types were formulated. Interaction types includes descriptions of how, when and in what order the user or VA speaks to each other. It can also include ways in which Sonja is accessed or how she is integrated into the larger phone interface.

Finally, the functionality also included description of Sonja’s personality. The personality as a phenomenon becomes relevant to discuss as Sonja simulates a conversation between humans, and because her choice of words and tone of voice has an impact on the way that a user relates to her.

3.2.2 Value

As this study is also concerned with the value that a VA can create for train passengers, it is important to define how value can be understood in this context. Value is, in this study, understood as whatever

passengers perceive to be positive improvements to their train journeys that can be created through them interacting with a VA. This understanding captures several of the aspects discussed in theory about

value, including how Grönroos & Voima (2013) talks about value making customer become better off, as well as how Vargo & Lusch (2004) talks about value being both co-created and determined by the customer at the moment of a product or service use. An important note is that the view on value in this study implies that the VA does not contain any value in and of itself, but that it is rather something that gives the passenger the potential for value creation. The study aims to investigate what those values are.

3.2.3 Barriers

Finally, this study is concerned with the barriers that exist for using the VA in the railway service context. Barriers, in this context, describes anything that raises the difficulty or limits the possibility of passengers using a VA to create value. This might include things that are both intrinsic to the nature of the VA itself, or things that are external such as context or user related aspects. This broad definition is used as the study aims to create a holistic view of the things that might be of concern when aiming to introduce a VA into a train-travelling context.

21

Method

In this chapter, the methods used included design probes, experience prototyping, interviews, and qualitative analysis are described and motivated.

Pre-study

The pre-study aimed to collect insights and ideas about what railway passenger would expect of a VA in a railway service context. The ideas and insights were mainly collected from railway passengers through design probes, but also through various forms of ideation. The collected ideas were structures using the affinity diagram method.

4.1.1 Design probes

Design probes are a user-centered method aimed to gain understanding for people and explore design opportunities (Matterlmäki, 2004). A design probe commonly exist in the form of a kit handed out to users, containing tasks that lets them document insights, thoughts, feelings and ideas that arises in certain contexts (Matterlmäki, 2004). The design probes usually have an exploratory character, and helps designers and users to interpret their situations and explore designs (Matterlmäki, 2004).

In this study, a design probe was developed and carried out to passengers on a train going from Stockholm to Linköping. The design probe aimed to:

- Collect ideas directly from passengers on how VAs could help them during their train journey - Collect insights about user’s understanding of and attitudes towards VAs

4.1.1.1 The Probes

The design probes, displayed in Figure 6 below, were designed to be self-explanatory meaning that all instructions that the participants required to fill it in was included in the pamphlet.

Figure 6: Design probes

The probe contained a series of tasks that aimed to let the user think about what they thought a VA could do for them on the train and was centered around on two main tasks. The first task concerned a VUI that

22

would be placed somewhere on the train, and the second task concerned a VA connected to the participants cellphone. The reasoning to explore both a static VUI and a mobile VA was to understand the users’ general needs that could be met through voice, and by connecting the VUI to the train, the users could think more freely about its possibilities. The VUI was represented by a wooden button (a fake microphone) fastened on the pamphlet by blue tack, as can be seen in Figure 6 above. Images of all the pages of the design probes can be found in Appendix 1.

The design probes also contained some warm-up questions such as “have you ever spoken to a virtual assistant (yes / no / I do not know)”, as well as some quantitative questions regarding the participants feelings regarding using a VA in different places.

4.1.1.2 Process

The design probes were developed in iterations where feedback from friends of the authors, supervisors and SJ employees were used to make sure the tasks could be understood and generated the intended type of answers. Once the probe had been iterated to a stage where people could understand them, they were printed on good quality paper and prepared for handout.

The design probes were handed out on a train going from Stockholm to Linköping. With permission from the train staff, the train callout system was used to announce that the design probes would be handed out and what their purpose was. The design probes were then handed out to 49 passengers who mostly sat in the first-class wagon of the train, together with a piece of candy as a thankful gesture for the ones willing to participate. The reason to why they were mostly handed out to first-class, was that the authors expected that people were going to be reluctant to help and therefore the probes would cover the whole train. The authors then started from the front of the train, where the first-class carriage were located, and realized toward the end of the carriage that the probes had almost run out. The participants then got 20 minutes to fill in the probes after which they were collected.

4.1.1.3 Analysis

The probes were analyzed through a qualitative content analysis. A more thorough description of the theoretical background to this can be found further down in 4.3 Qualitative Content Analysis. The probes were analyzed in the following steps:

1. The results on the quantitative questions were compiled

2. All generated ideas for what a VA could do were collected in a spreadsheet. For each idea, the number of times it was generated was noted. The ideas can be found in Appendix 2.

3. All text responses were collected in a spreadsheet (including what people wanted to say to a VA, and what answer that VA would give)

4. The responses were analyzed through a coding process, resulting in a list of insights.

The coding was first done by the two authors separately, who each went through the list of responses and coded each response inductively. The list of codes was then compared, and a common codebook was created listing for every code a corresponding definition. For example, the code “place info - train” had the definition “asks for information related to where things are located on the train”. All codes were then structured in categories to visualize how codes were related. The codebook can be found in Appendix 3. Once the codebook was finished, the codes were re-applied to the responses, and the number of times each code appeared was counted.

23

4.1.2 Ideation

The pre-study phase started with the authors writing down all ideas, thoughts, and questions they could think of concerning the technology, the users, and the context. This was to get the things in their heads on paper, so they could focus on collecting new information. Throughout the pre-study phase, different design research techniques were then employed to gain more insights about the technology and the context.

As a more active way of sparking creativity, a bodystorming session was held. In the bodystorming session, the authors placed themselves in the context of study and tried to generate ideas for what a VA could help them with in that context. More specifically, the bodystorming session took place on a train journey to Stockholm where the authors were going to meet the project owners, SJ. By employing this method, the authors gained a better understanding and empathy for the service context which the VA would be integrated in.

The session began at a train station, and then continued during a train ride. It followed the structure that is presented in This is service design doing (Stickdorn et al., 2018), basically enacting six pre-written scenarios with connected design questions (see Appendix 4), discussing how a VA could be used. Insights and ideas were written down.

Some ideas were also generated using a customer journey (provided by the project owner) as inspiration. The customer journey was based on interviews with the project owner’s customers and showed touchpoints and emerging customer needs throughout the train journey. For each need in the customer journey, the authors tried to apply the VA as a solution to see if it could fill the need. The possible applications of the technology were saved in the list of ideas.

4.1.3 Affinity diagram

To get a better overview of the ideas from the ideation and the design probes, they were sorted in an affinity diagram. This method was used several times during the project, to sort insights, concepts, themes, and ideas. An affinity diagram is a way of bundling and grouping information (Dam & Siang, 2020). During the exercise, all collected ideas were printed on pieces of paper and put in a pile on the table. The first idea was placed at the center of the table. The authors then picked up one idea each at a time in silence. If the participant thought the idea in his hand resembled one on the table, he put it next to that idea. If not, he put it separately. If duplicates were discovered, they were removed. When all ideas had been placed on the table, the participants discussed the categories, re-arranged the ideas until they were satisfied, and then named each category. The affinity sorting session is displayed in Figure 7 below.

24

Figure 7: Affinity diagram sorting

Experience prototyping

In experience prototyping sessions, potential future users of a VA got to interact with a prototype of a VA called Sonja. The prototyping sessions were held remotely with individual participants through video calls, as shown in Figure 8 below. During the sessions, participants were presented a scenario and followed a set of tasks in which they interacted with the prototype on their phone. The facilitator asked questions regarding the participants experience during and after the sessions.

Figure 8: Experience prototyping setup

The aim of letting participants interact with the prototype was to make them experience what it could be like to use a VA during a train journey. This experience would then make the participants able to reflect on how the VA may be able to help them within the context and what they thought about its functions. The purpose of such a reflection was to enable an interview in which the participants could more easily give their opinions of the future potential for a VA in their train journeys. The experience prototyping

25

aimed to result in insights about the functionality, values, and barriers that user might perceive as possible in such a future scenario.

4.2.1 Participants

Because the study was explorative and aimed to collect a broad range of insights, participants were selected to achieve a high variation in thoughts and opinions expressed. The participants differed in age (between 20 and 70), had different train-travelling habits (from daily commuters to people who travel about once a year) and had different levels of experience in using VAs. The motivation behind using participants that varied in these parameters was that the authors (with recommendations from SJ) thought that these parameters would give the most variance in opinions expressed. A table of the participants can be found in Table 2 below. All participants that participated signed an informed consent and the names used for the participants in this report are pseudonyms.

Table 2: Experience prototyping participants

Scenario Name

Approx.

Age Gender

How often do you travel by train?

Have you ever used a voice-controlled assistant?

Owns a smart speaker

1 Lovisa 67 Female Less than once a month It has happened No

1 Alice 64 Female A few times each month Never No

1 Annika 33 Female A few times each week Never Yes

2 Roger 25 Male Less than once a month A few times each month Yes

2 Elise 28 Female A few times each week Never No

2 Bengt 53 Male A few times each week Never No

3 Adam 55 Male A few times each month It has happened No

3 Paul 28 Male Less than once a month A few times each month Yes 3 Lukas 22 Male A few times each month It has happened No

Participants were recruited through friends and social media channels. A brief text explaining what the test was about and why participants were wanted was sent out together with a link to a google form. Those who were interested in participating could then fill in the form where they entered their age, how often they travelled with train and their experience in using VAs. From the google form answers, participants where then screened so that those who were selected for participation could be contacted so that a time for prototyping could be scheduled.

4.2.2 Process

Once a participant had been identified, a time was scheduled for when they would participate in the prototyping. Before the scheduled prototyping session, preparatory material was sent out to the participant. The material included a set-up guide that would enable them to install the prototype on their phone, an informed consent form, and a PowerPoint-file that would be used during the actual session. The following steps where they executed during the prototyping sessions:

1. The facilitator called the participant through video chat.

2. The participant was asked for consent for recording the video call, and the test began. 3. The participant opened the PowerPoint-file and started doing the presented tasks. 4. In between the tasks, the facilitator asked questions about the functionalities tested.

26

5. When the whole scenario was complete, the test leader interviewed the participant about the experience.

All sessions were audio-recorded and lasted for about 45 minutes, where about 15 minutes was spent completing the tasks and the rest was spent interviewing. The prototyping was done with nine participants at different times during a three-week period.

4.2.3 The prototype

The prototype, called Sonja, was accessed by the participants through the google assistant app. A screenshot from a conversation with Sonja is displayed in Figure 9 below.

Figure 9: A conversation with Sonja

Sonja would give pre-written responses to prompts given by the user. This means that Sonja only worked if the user approximately followed the tasks presented in the PowerPoint-files, as Sonja would not understand questions or commands which she did not already have a planned response for. In this sense, Sonja gave the illusion of functionality that she did not actually have. When interacting with Sonja during the prototyping session, the user was introduced to a scene, for example as in scenario 3:

You are going to take a trip from Stockholm to Gävle. Ask Sonja what the weather is like at your destination.

Sonja was then programmed to respond to questions about the weather at the destination with: The weather in Gävle will be clear but cold with 4 degrees Celsius at your arrival time.

27

This allowed the user to image that Sonja knew they were going to Gävle, when they would arrive, and what the weather was going to be like.

4.2.4 Creating the prototype

Sonja was created using Dialogflow from the google cloud platform. Dialogflow is an “end-to-end, build-once deploy-everywhere development suite for creating conversational interfaces for websites, mobile applications, popular messaging platforms, and IoT devices” (Google, 2020). Dialogflow was used as it enabled a cheap way of creating a voice assistant prototype that could understand and give responses to user input. This in turn made it possible to simulate a future scenario in which such an assistant is present. Each response Sonja could give was packaged as an intent, which is basically Sonja's response coupled together with what prompt the user would use for that intent. If a user’s intent were to know what seat she had, Sonja would listen to prompts like “Where is my seat?”. This would activate the intent, where Sonja's response would be “Your seat is number 32.” Sonja was trained to recognize the intent of the user by giving her “training phrases”, displayed in Figure 10 below, which was different examples of how someone would formulate the prompt. These variations were then abstracted through machine learning, so that Sonja could respond to prompts like those that were listed.

Figure 10: Training phrases for Sonja

How the machine learning was done is unfortunately not presented, but a tendency to respond to keywords like “booking” and “trip briefing” was noted, leading to conversations as pictured below in Figure 11. When Sonja did not understand what you said, or lacked an intent that matched the interpreted prompt, she responded with a “fallback intent”, essentially responding with different versions of “Can you say that again?”.

28

Figure 11: Sonja does not understand the command

Figure 12: Sonja identifies the keyword "genomgång"

The first couple of phrases was written in by the programmer, but since each test of Sonja was saved to Dialogflow it was easy to add phrases that some tester had said during the tests. This meant that every test of Sonja could lead to an update of her ability to understand you, leading to better performance. This was done during tests of the prototype before the experience prototyping session, giving Sonja some real word experience before the sessions. These tests were iterated a couple of times, each time increasing the amount of training phrases will also rewriting the scenarios and some of her responses to fix misunderstandings between Sonja and the participants.

4.2.5 Scenarios

The participants were divided into three groups, with each group presented with one of three scenarios. The reason for dividing the participants into different scenarios was to be able to try as many different ideas from the pre-study as possible, which would generate a wide variety of insights. The scenarios presented different parts of the train journey, as different parts of the journey were related to different functionality. For example, it is unlikely that a passenger would ask Sonja which platform the train was leaving from, while sitting on the already moving train. An overview of the three scenarios, including their

29

name, aim and how they were presented to participants during the prototyping sessions, are shown in Table 3 below.

Table 3: Experience prototyping scenarios

Nr Scenario Aim Scenario presentation text

1 Trip planning To explore if a VA is suitable for

planning a trip

“You are at a work meeting in Linköping and are planning your trip home. You expect to be done at 16.45 and are planning to take the train back to Södertälje. You decide to use SJ’s voice assistant Sonja that you heard about. You are used to paying about 125 SEK for your trip.”

2 Guidance To explore if a VA is suitable for guiding the user to their train

“You are at Stora Torget in Linköping and have just heard that your favorite band will be playing in Stockholm tonight. As you do not have anything else planned, you decide to take the train there to go to the concert. You have your earphones plugged in and use SJ’s voice assistant Sonja to be guided along the way.”

3 On the train To explore how a VA could be helpful to a passenger being on the train

“You have settled down on the train and relaxed for your journey. You saw an ad for Sonja, SJ’s virtual assistant, on the platform before boarding the train, and decide to play around a bit. … The train has stopped moving and you decide to ask Sonja for an explanation for the delay.”

For each scenario, there was a PowerPoint-file presenting the scenario that the participant looked at during the session. Each scenario was divided into several acts. Each act had tasks connected to it that would be solved using the prototype. The idea was that each task would present a way in which the VA could be used. For example, in the first act from scenario 1, presented in Figure 13 below, it was presented how the VA could be used to book a train ticket. The full list of texts and tasks for each scenario is presented in Appendix 5.

30

4.2.6 Interviews

The participants were interviewed during and after the prototyping sessions. The interviews were constructed on guidelines from This is service design doing (Stickdorn et al., 2018) and Convivial toolbox (Sanders, 2012). After each act there were pauses in which the facilitator asked the participant about what they just experienced and the functions that had been presented. When the session was finished, a longer interview (about 30 minutes) was conducted. The interviews were semi-structured and conducted so that the participants would be able to express their thoughts about the prototype and its potential value in a future train-travelling scenario.

The facilitator had a prepared interview guide (see Appendix 6) containing information to present the participants with, as well as a list of questions to ask them. The participants were asked about what they liked and disliked, if they thought the technology could help them in the future, if they had any issues speaking to Sonja, and how they would like Sonja to be in the future. The interviews were done in a semi-structured way, meaning that the list of questions was used as a starting point for discussion and a support for the interviewer, but that the participant was also allowed to talk more freely about their experience. During the interviews, a second facilitator was also in the video call to listen. The second facilitator could ask additional questions at the end of the interview.

Qualitative Content Analysis

Qualitative content analysis is a way of analyzing text data (Hsieh & Shannon, 2005). Hsieh & Shannon (2005) defines qualitative content analysis as “a research method for the subjective interpretation of the content of text data through the systematic classification process of coding and identifying themes and patterns”. The goal of using qualitative content analysis is to gain an understanding of the phenomenon being studied (Hsieh & Shannon, 2005).

The interview transcripts were analyzed using what Hsieh & Shannon (2005) describes as “conventional content analysis”, in which themes and categories emerge directly from the studied data. Such an approach is suitable when prior theory on the studied phenomenon is lacking (Hsieh & Shannon, 2005), and was used here as prior knowledge about what functionality, values and barriers could emerge could not be found in prior research.

The qualitative content analysis aimed to serve two purposes. Firstly, it helped the authors in familiarizing themselves with the material and building empathy for the thoughts expressed by the participants. Secondly, it created a structure by highlighting the important parts of the texts and grouping the information into different areas, making it easier to discuss and draw conclusions from.

4.3.1 Process

During analysis, the interview transcriptions were first read through by the authors, with the aim of getting a general understanding for its content. Then, each passage was highlighted, coded, interpreted, categorized, and compared. Through the analysis, a list was created containing values that participants expressed that the technology could create for them, identified barriers for creating those values, and functionality aspects of the VA that was discussed.

31

More specifically, the analysis contained the following steps:

1. Codes were written for each passage in the transcripts, highlighting the important parts of each passage.

2. Themes highlighting commonalities between passages were identified and written down in a spreadsheet. The passages in the transcripts were labelled using the themes.

3. Themes were organized into categories, and duplicates were removed 4. A definition was written for each theme

5. The organized themes were re-applied to the transcriptions to make the labelling coherent 6. Through informal discussion, higher level themes describing values, barriers and functionalities

were identified.

7. All quotes related to each high-level theme were collected in a list An example of the analysis process is displayed in Figure 14 below.

32

Results

In this chapter, the results from the study will be presented. First, how respondents of the design probes and the participants of the experience prototyping perceived the VA’s functionality. Then, the values and barriers of having such a VA in a railway service context.

Functionality

Through the pre study and the experience prototype, different versions of a VA have been presented, evaluated, and discussed. This chapter presents a summarized description of the potential functionality that a VA could have for the railway passenger. Again, the taxonomy is created by the authors, for the purpose of discussing the value creation possibilities. The functionality of Sonja was divided into three aspects: functions, interaction types and personality.

5.1.1 Functions

In total, 55 unique VA functions was identified. The functions that was represented during the experience prototyping came from the design probes and the ideation. During the experience prototyping interviews, additional ideas for functions were identified, some of which had already been addressed in the design probes but had not been put into the prototype. A flowchart describing what functions came from what activity are shown in Figure 15 below.

Figure 15: Functions identified through different methods

These 55 functions were categorized, using the affinity diagram method, into five categories presented below in Table 4 below. The complete list of functions, including their source and category, can be found in Appendix 7.