Designing as trading-off: a practice-based view on smart service systems

ABSTRACT

Posture-related problems, such as back pain, are an increasing global burden. They are deeply intertwined with how humans sit. While the information systems (IS) literature has been relatively silent on this matter, emerging literature in related disciplines has begun to attend to this problem by developing various artefacts. However, researchers have oftentimes done so by basing their artefacts on engineering rationales and attending only limitedly to the interactions between artefacts and humans. These interactions are crucial because data on posture is best collected by placing sensors on humans’ backs. This calls for considering and evaluating how bodies move in relation to sensors, the emotive reactions of humans to sensors and how humans make sense of recommendations emanating from underlying artificial intelligence (AI) technologies. We uncover what these considerations of human-centredness mean for designing smart service systems for posture management and suggest that a core consideration relates to trading-off possibilities of smart technologies and necessities emerging from practices. This study contributes to the body of knowledge on designing smart service systems and responds to calls for more IS research dealing with the prevention of chronic health conditions.

KEYWORDS:

• Smart service systems
• design science research
• posture management
• chronic conditions
• practices

1. Introduction

Approximately eight percent of the global population suffers from frequent back pain, a problem linked to the fact that we sit too much in harmful postures (Seth et al., 2021). While problematic consequences of sitting have been a long-standing concern for professions where office work has always been important such as the law, academia, and consultancy (Schell et al., 2008), this “sedentary epidemic” (Bourahmoune & Amagasa, 2019, p. 5808) has spread to many other occupations given the rise of remote work due to the COVID-19 pandemic (Papalia et al., 2022). Scholars across a variety of disciplines such as computer science, health-related disciplines, and fields related to engineering have begun to work on sensor-based smart technologies in order to design artefacts that could help with solving this problem (Ahmad et al., 2021; Bourahmoune & Amagasa, 2019; Ho & Ismail, 2021; Seth et al., 2021; Zhao et al., 2021). Progress notwithstanding, an important conceptual concern that has been voiced in this context is that these literatures are largely based on engineering rationales that barely consider how technologies relate to humans and vice versa (Bourahmoune & Amagasa, 2019; Roggio et al., 2021). However, if smart technologies are supposed to live up to their potential to improve how we as humans sit, designing them needs to take into account how the practice of sitting unfolds and how humans relate to smart technologies while sitting. The reason is that we need to attend to sitting in detail in order to generate detailed insights about whether and how real-time data can be collected about posture and how this data can be used in order to improve sitting.

The growing amount of literature on smart service systems in information systems (IS) research is helpful for solving the aforementioned problem (Beverungen, Breidbach, et al., 2019; Beverungen, Matzner, et al., 2017; Yang et al., 2021). Smart service systems are capable of reconfiguring and self-organising because learning algorithms adapt services to customers’ behaviours and demands in real time (Beverungen, Breidbach, et al., 2019; Maglio, 2015; Maglio et al., 2015; Medina-Borja, 2015). This is why smart service systems are typically seen as human-centred (Beverungen, Matzner, et al., 2019; C. Lim, M.-J. Kim, et al., 2018; Lim & Maglio, 2018; Lim et al., 2016) and IS researchers are increasingly working on designing algorithms and computational operations that create such human-centredness (Beverungen, Lüttenberg, et al., 2017; Huber et al., 2019; Klör et al., 2018; Knote et al., 2021). Thus far, two ways to conceptualise the human-centredness of smart service systems have emerged: the elements view and the practice view. The elements view draws from service science the idea that “humans” are to be seen as one element which interacts with other elements, such as data, algorithms, machine learning (ML) models, and physical objects, in smart service systems (Beverungen, Breidbach, et al., 2019; Beverungen, Matzner, et al., 2017; C. Lim, Kim, et al., 2018; Lim & Maglio, 2018). Such conceptualisations have served as effective justificatory knowledge for designing self-learning algorithms that let the human-centredness of smart service systems emerge (Beverungen, Lüttenberg, et al., 2017; Huber et al., 2019; Klör et al., 2018). However, the conceptual challenge with directly applying the elements view to our setting is that primarily attending to self-learning algorithms reiterates the problem that practices remain unconsidered, as the elements view treats humans as one of many elements, thus offering little granularity to unpack practices in detail (Maglio, 2014, 2015; Maglio et al., 2015).

The practice view of smart service systems offers a different approach to conceptualising the human-centredness of smart service systems as it attends to the particular details of what humans actually do and how they interact with smart technology in situ (Barnes, 2001; Feldman & Orlikowski, 2011; Orlikowski, 1996; Reckwitz, 2002). In terms of smart service systems, research has shown that smart objects interact with human bodies, knowledge and emotions in practices which can become smarter or break down entirely, depending on how these interactions unfold (Wessel et al., 2019). This practice view of smart service systems argues that the human-centredness of smart service systems is not only a question of how algorithms and models process data but, most importantly, a question of how smart objects interact with humans (Wessel et al., 2019). Yet, the practice view of smart service systems has so far not been used to design smart service systems and for unpacking fresh takes on design knowledge about smart service systems more generally (Drechsler & Hevner, 2018). This is important because the societal problem that we aim to tackle, i.e., improving how we as humans sit, calls for designing smart service systems with a strong emphasis on the practice of sitting. This is why there is great promise in building on the practice view when designing smart service systems, which is why we ask how should human-centred smart service systems for posture management be designed?

In light of this research question, we designed and evaluated a smart service system to improve posture management as part of a multiyear project. Our eventual smart service system combined a T-shirt with sensors, microprocessors, ML algorithms, and a mobile application that enabled continuous interaction with humans. We combined naturalistic and artificial methods to formatively evaluate our smart service system (Venable et al., 2016). Naturalistic methods like interviews and prototype testing with end-users, physicians, and physiotherapists aimed at iteratively assessing how well the smart service system contributes to improving posture and reducing human risks from user interaction. We complemented the formative evaluation with artificial evaluation methods involving ML optimisation and sensor positioning to assess the technical performance of the smart service system and mitigate technical risks (Venable et al., 2016). In continuous cycles between design and formative evaluation, we iteratively formulated and refined design principles that show how a smart service system should be designed to make the practice of sitting smarter.

We offer three contributions to the literature. First, our work extends design knowledge for smart service systems by suggesting design principles based on the practice view of smart service systems. This is an important contribution because it sensitises designers and researchers to the fact that smart service systems may develop in unintended ways. Our evaluation attended to how smart service systems played out in practice, disclosing how various design decisions that were technically justified had to be adapted due to unintended impacts on the underlying practice. These observations led us to revise our design principles as trade-offs. Thus, designing for human-centredness is a process of trading-off, in which the pursuit of some design goals may directly counteract the achievement of others. Second, we advance research on the role of digital technology in the self-management of chronic conditions in IS, a key societal challenge of our times (Bardhan et al., 2015, 2020; Dadgar & Joshi, 2018). Our smart service system is designed to avoid potentially harmful behaviour in that it speaks to the prevention of conditions before they appear or before humans become aware of them. Lastly, we contribute to ongoing research seeking solutions to posture management. Specifically, we present an artefact as an instance of a design that illuminates our understanding of both the problem space and the solution possibilities for dealing with posture-related problems. As Baskerville et al. (2018) argued, such insights are valuable knowledge contributions that justify the prescriptive knowledge advanced by our study.

2. Relevant literature

Prolonged sitting has induced problems for human well-being on a global scale, and solutions to this problem are beginning to emerge in various scientific disciplines and from various perspectives. In this section, we review this work in order to arrive at our problem statements and situate our specific contributions within this context.

2.1. The importance of smart technologies for posture management: interdisciplinary perspectives

Researchers in medicine have studied the consequences of sitting for many years (De Rezende et al., 2014; Dunstan et al., 2011; Harvey et al., 2013). They have consistently found that this sedentary behaviour – that is, “a cluster of behaviors adopted in a sitting or lying posture where little energy is being expended” (Harvey et al., 2013, p. 6645) – is related to various chronic conditions (Booth & Lees, 2007; Chau et al., 2013; De Rezende et al., 2014; Dunstan et al., 2011; Panahi & Tremblay, 2018; Petersen et al., 2014), particularly back pain (Hartvigsen et al., 2000; Korshøj et al., 2018; Lis et al., 2007; Mahdavi et al., 2021; Roffey et al., 2010). This is a timely problem, as the COVID-19 pandemic has led to steep increases in sedentary behaviour due to people spending prolonged periods of time at home (Ammar et al., 2020; Qin et al., 2020; Smith et al., 2020; Stockwell et al., 2021). Given the extensive remote work policies that many employers now have (Microsoft, 2021), this problem is likely to worsen in the future.

Researchers in a variety of disciplines have started to address this problem. Scholars in public health have focused on effective policies intended to enable individuals to better self-manage their posture, potentially, by means of smart technologies (Ammar et al., 2020; Dunstan et al., 2011; Mahdavi et al., 2021). However, the technologies themselves were not in the foci of these studies. Researchers in healthcare informatics have put smart technologies centre-stage and have attended to how particularly fitness trackers can prevent prolonged sitting (see, e.g., Stark et al., 2022). Fitness trackers measure physical activity objectively and hence surpass the accuracy of patient self-reported measures of sedentary behaviour (McLaughlin et al., 2021). Furthermore, evidence of potentially positive effects of fitness trackers on physical activity has been accumulated recently (Abedtash & Holden, 2017; Chan et al., 2022; Daryabeygi-Khotbehsara et al., 2021; Kheirkhahan et al., 2018; Kim et al., 2022; Müller et al., 2018; Sanders et al., 2016). Also, data collected by fitness trackers can be used to improve diagnosis and clinical advice through the application of ML (Goh et al., 2022; Mönninghoff et al., 2021).

Another line of research has focused on predicting posture-related problems by using ML as it helps to accurately represent physical activity captured via various types of sensors (Liaqat et al., 2021). For example, ML algorithms from computer vision can accurately represent bodily movements captured by noncontact sensors in cameras (Ho & Ismail, 2021; Kumar et al., 2021). Likewise, ML can also be used to model movements based on direct-contact sensors, such as gyroscopes or accelerometers in fitness tracking devices (e.g., Stark et al., 2022), pressure sensors in wheelchairs (Ahmad et al., 2021), car seats (Zhao et al., 2021), or cushions (Bourahmoune & Amagasa, 2019).

The abovementioned works have established that smart technologies can serve as important means for trying to improve how we as humans sit. However, challenges remain. One is that general motion sensors, such as those in fitness trackers, only offer a rough approximation of posture as they do not collect data directly from the human back, even though such data would measure posture more precisely (Seth et al., 2021). Particularly, researchers in material sciences and embedded systems design have argued that smart garments are a more effective means of collecting data from an individual’s back (Loncar-Turukalo et al., 2019; Spanakis et al., 2016). In contrast, fitness trackers are typically located elsewhere on the body. Similarly, sensors built into smartphones or cushions also do not directly measure posture. In turn, attaching sensors to a human’s back promises to generate finer-grained data and improve our predictions of posture problems as well as generable advisories to sit differently.

Researchers in the field of human-computer interaction (HCI) have also begun to study how interactions between wearable technologies and humans could promote physical activity (Chignell et al., 2022; Fennell et al., 2019; Ferreira et al., 2021; Pinder et al., 2018; Rapp & Cena, 2016). The general take-away from these studies is that designing such wearables needs to account for the specifics and potential idiosyncrasies of human bodies that differ vastly across individuals (Berrett et al., 2022; Chignell et al., 2022; Mueller et al., 2021; Williams et al., 2017). In this context, Kuijer et al. (2013) argued that sociological practice theory would be a particularly powerful conceptual toolkit to capture how human bodies move in response to technological prototypes and that the latter should be designed by attending to these responses.

From a practice perspective, humans and smart technologies are actors in their own rights who configure one another. A practice-based understanding of smart service systems thus requires that the design and engineering of smart technologies pay close attention to the interactions between smart technologies and human movements, emotions and thinking that occur during sitting (Beverungen, Müller, et al., 2019; Wessel et al., 2019). While attention to these situated details may seem irrelevant and mundane at first sight, it matters significantly because changes in posture are sometimes barely noticed by observers. Furthermore, if smart technologies are to deliver their promise to assist with posture management, their design needs to be derived from the practice that they are supposed to make smarter (Wessel et al., 2019). However, policy implications derived in medicine and public health typically focus on the policy level and thus “zoom out” from the situated practice of sitting. Likewise, fitness trackers are unlikely to generate the data needed for strong problem prediction and adequate posture modelling. Smart technologies developed in computer science are typically built on engineering rationales that foreground technological functioning over interactions between humans and smart technologies during sitting. While useful for optimising the accuracy of measurements and prediction, this leaves open questions surrounding how models and technologies play out in practice and what this may mean for the design of smart service systems. In fact, computer scientists are beginning to recognise this problem by calling for more work involving humans in designing and evaluating smart technologies (Bourahmoune & Amagasa, 2019; Roggio et al., 2021). However, even these works do not explicitly recognise the conceptual and practical ramifications of taking seriously the notion of practice(s) in designing these technologies in a human-centred way. Conversely, the HCI literature paves the way towards accounting for practices, but then attends to the underlying smart technologies in limited ways.

2.2. A practice view of smart service systems

Placing interactions between humans and smart technologies at the centre of designing smart service systems calls for a lens that accounts for these interactions. Recent literature on smart service systems has made important progress in this regard. Smart service systems foreground that physical objects become augmented with sensors that feed data into analytics architectures (Beverungen, Matzner, et al., 2019; Beverungen, Müller, et al., 2019), which renders the overall technological setup context-aware and capable of self-organisation (Lim & Maglio, 2018; Maglio, 2015; Medina-Borja, 2015). Humans and smart technologies thus interact in smart service systems – formally defined as “system(s) capable of learning, dynamic adaptation, and decision-making based upon data received, transmitted, and/or processed to improve its response to a future situation” (National Science Foundation, 2014, p. 5).

Thus far, research on smart service systems has highlighted that such systems are human-centred because they learn from data how to deliver highly individualised services to humans (Beverungen, Breidbach, et al., 2019; J. Y. H. Lee et al., 2020; Maglio, 2015; Maglio et al., 2015; Medina-Borja, 2015). These technological potentials call on organisations to innovate in value propositions (Chen et al., 2020), business models (Dreyer et al., 2019; Keskin & Kennedy, 2015), pricing strategies (Kennedy & Keskin, 2016), approaching customers more dynamically (Albani et al., 2017; Demirkan et al., 2015; Massink et al., 2010; Peng et al., 2017; Wiegard & Breitner, 2019; Wuenderlich et al., 2015), and have inspired an emerging stream of design science research (DSR) studies. Beverungen, Lüttenberg, et al. (2017) developed design principles for recombinant service engineering, where innovation in smart service systems occurs due to recombinations of existing resources such as humans, data or technological objects. For example, services related to the reuse of existing batteries for electric vehicles (Klör et al., 2018) and other products based on layered modular architectures (J. Y. H. Lee et al., 2020; Yoo et al., 2010) can be effectively designed via recombinant service engineering. Likewise, Knote et al. (2021) focus on the material properties of smart personal assistants to derive their design implications. Huber et al. (2019) propose a domain-specific modelling language suited to the design and engineering of smart service systems, as it captures the dynamic interplay between humans, data and algorithms (Beverungen, Lüttenberg, et al., 2017; Geum et al., 2016; C. H. Lee et al., 2019; Moldovan et al., 2018).

These DSR studies commonly conceptualise smart service systems as comprising of different elements such as objects, data, humans and algorithms that can be integrated with one another via service engineering (Beverungen, Lüttenberg, et al., 2017; Huber et al., 2019). One important goal of studies such as these then is to leverage self-learning algorithms that create human-centredness of smart service systems through learning from data. The elements view of smart service systems thus provides a powerful theoretical framework for designing technological objects but it has been criticised as capturing humans only as a broad and generic category, offering little room to delve specifically into emotions, knowledge and bodily movements of humans (Maglio, 2014, 2015; Maglio et al., 2015). However, putting humans at the centre of designing smart service systems demands an exploration of these details. Some progress in this regard has been made by approaches that highlight the centrality of smart objects that collect and process data from human practices in smart service systems (Beverungen, Matzner, et al., 2019; Beverungen, Müller, et al., 2019). Wessel et al. (2019) built on this view and explicitly integrated it with the sociological notion of practice(s) (Barrett et al., 2012; Oborn et al., 2011; Reckwitz, 2002) in order to arrive at a different conceptualisation of human-centredness in smart service systems. In this practice view, the human-centredness of smart service systems does not rely on the smart object only, but it depends on practices within which smart objects and humans interact. How smart objects interact with human emotions, knowledge and bodily activities is seen as a key explanation for why smart service systems may live up to their technological potential or break down entirely (Wessel et al., 2019). In turn, the practice view is a particularly powerful theory for designing smart technologies because it stresses that interactions between smart technologies and humans will frequently develop in unexpected ways that are important to evaluate and account for while conducting DSR (Holeman & Barrett, 2018; Leonardi & Rodriguez-Lluesma, 2013).

The wider notion of practice(s) builds on the sociological school of thought concerned with what human and technological actors actually do (Barnes, 2001; Feldman & Orlikowski, 2011; Orlikowski, 1996). Similar to Wessel et al. (2019), we rely on Reckwitz’s (2002) understanding of the term practice because it offers established conceptual categories that aid with the design of human-centred smart service systems: practices are “a routinized type of behaviour which consists of several elements, interconnected to one other: forms of bodily activities, forms of mental activities, ‘things’ and their use, a background knowledge in the form of understanding, know-how, states of emotion and motivational knowledge” (p. 249). In this sense, practices refer mostly to mundane activities that individuals carry out regularly without necessarily reflecting on them (Polanyi, 1966; Schatzki, 1996). For example, when we ride a bike, run a trail or sit in front of a computer, we do not rationally think about how we move our bodies. Carrying out these activities is deeply routinised, oftentimes unconscious and highly consequential for posture-related problems.

The practice and elements views of smart service systems differ in important ways in terms of their design implications. The practice view directs the attention of researchers towards designing and evaluating smart objects based on the practices within which they interact with human emotions, knowledge, and bodies. The elements view directs attention to designing learning algorithms without building on the human details disclosed by the practice view. However, attending to how smart objects play out with regard to emotions, knowledge and bodies is important for the overall effectiveness of smart service systems. However, a gap prevails in leveraging these important insights in a DSR study and in designing a human-centred smart service system based on an understanding of practice(s) (Wessel et al., 2019).

Taken together, two problem statements emerge from our transdisciplinary assessment of relevant literature. The first captures the societal problem that we as humans sit too much. Sitting and associated sedentary behaviours are linked to adverse health outcomes that scholars across disciplines have recognised and begun to respond to. The second problem statement is conceptual, particularly, that research about the design of smart technologies in particular and smart service systems in general has a tendency to account for technological considerations and less for how smart technologies interact with humans. This means that there are shortcomings in terms of the designs of particular artefacts, for example the use of sensors that only approximate posture, as well as in terms of the design knowledge about smart service systems more generally. While the practice view of smart service systems is potent to respond to these problems, a gap prevails in terms of leveraging this view as justificatory knowledge for designing smart service systems.

Against this background, we develop our meta-requirements and initial design principles based on earlier work related to the practice view of smart service systems. Particularly, we rely on the conceptual categories of smart objects, human bodily activities, knowledge and emotions as they provide a powerful toolkit to organise our thinking and guide our design and evaluation (Beverungen, Müller, et al., 2019; Oborn et al., 2011; Reckwitz, 2002; Wessel et al., 2019). To exemplify how these categories relate to the practice of sitting, it is important to note that an object might influence sitting. For instance, a shirt that scratches may cause micromovements that affect posture, which indicates an intimate relationship between objects and bodily activities. When working, we may read an email that affects our knowledge. News that excite us often lead to changes in postureand deep concentration can cause us to drop our shoulders unconsciously. Therefore, knowledge is closely intertwined with objects and bodily activities. In addition, emotional triggers, such as excitement, may generate changes in posture and thus affect bodily activities. In short, these parts of a practice come together in how the practice of sitting affects one’s posture. Our meta-requirements and design principles are built on this thinking.

2.3. Designing for smart sitting: meta-requirements and initial design principles

Our research question centres on investigating how to design human-centred smart service systems for posture management. This means that we are asking a question concerned with the design of smart service systems in our chosen context. DSR is a well-established paradigm in IS research that is uniquely positioned to tackle such a research question (Baskerville et al., 2018; Hevner et al., 2004; Peffers et al., 2018). Furthermore, DSR has the potential to advance theoretical and practical knowledge via the creation and evaluation of new and innovative artefacts as it attends to questions that have practical and theoretical relevance at the same time (Gierlich-Joas et al., 2021; Hevner et al., 2019; Vom Brocke et al., 2020; Walls et al., 1992). Our study addresses the societal problem of sedentary behaviour, where it is necessary to assist users in better managing their posture. The fact that back pain is a global burden linked to prolonged sitting and that people are likely to spend considerable time sitting at home corroborates the importance of our work. We work on this societal problem by tackling a conceptual problem existing in the fact that extant work on smart technologies to better manage posture is largely build on engineering rationales while our design knowledge about smart service systems mentions “humans” mainly as one category but does not open up practices. However, opening them up matters for designing smart service systems that help with managing posture and improving sitting. Sitting needs to be closely monitored and data need to be fed back to humans so that they can adjust bodies and movements. This means that we need an artefact capable of collecting data from an individual’s back and to design this artefact for deeply routinised interactions with humans.

To integrate practical and conceptual knowledge, DSR scholars link their situated artefacts to more general meta-requirements that translate the concrete needs that an artefact is supposed to fulfil into more general goals that a class of artefacts should cater to (Walls et al., 1992, 2004). Following this approach, we first identify the requirements for our artefact, which guide its iterative design. Second, we specify our artefact design in terms of design principles that describe whom the design serves, under what circumstances and why (Gregor et al., 2020). We summarise our initial meta-requirements, design principles and how the evaluation of our artefact is related to both in Table 1. The table thus provides a conceptual thread that weaves together the remainder of this paper.

Table 1. Meta-requirements and initial design principles.

#	Initial design principles	Meta-requirements from knowledge base	Evaluation focus
1	Human-centred smart service systems require technologies (e.g., sensors, machine learning models, wearable devices etc.) that predict practices based on measuring bodily activities comprehensively, accurately, and reliably.	MR1: Provide effective “backstage analytics” Beverungen, Müller, et al. (2019) to make a practice capable of self-organising and reconfiguring.	Accuracy and reliability of instant feedback as delivered by machine learning model and sensors. Captured via: summative artificial evaluation (see below)
2	Human-centred smart service systems integrate sensors with bodily activities in ways that users perceive as unobtrusive.	MR2: Smart service systems need to operate as “invisible computers” Beverungen, Müller, et al. (2019), that are unobtrusive to users.	Whether, and if so, how, smart service systems’ operations that generate instant feedback become “invisible computers” (Beverungen, Müller, et al., 2019). Captured via: formative naturalistic evaluation (see below)
3	Human-centred smart service systems provide feedback that users can easily understand.	MR3: Feedback and advice on changing practices must be easy for users to understand.	Whether users understand instant feedback and consider it valuable to change their practice of sitting. Captured via: formative naturalistic evaluation (see below)
4	Human-centred smart service systems provide a functional setup that avoids triggering negative emotions in users.	MR4: Smart service systems need to avoid stimulating negative emotions from their usage.	Affective reactions of users to the instant feedback provided by smart service system. Captured via: formative naturalistic evaluation (see below)

2.3.1 Meta-requirement 1 (MR1)

The elements and practice view of smart service systems consider smart objects as artefacts that generate and collect data from users and feed this data into computational processes that render smart service systems capable of self-organising and reconfiguring (Beverungen, Müller, et al., 2019; Wessel et al., 2019). Empirically speaking, smart objects often involve wearable devices, ML models, and physical devices equipped with sensors (Lim & Maglio, 2018). Beverungen, Breidbach, et al. (2019) conceptualise the underlying computational processes as backstage analytics that process data in adequate ways. Our MR1 thus calls for effective backstage analytics (Beverungen, Müller, et al., 2019).

MR1: Provide effective “backstage analytics” to make a practice capable of self-organising and reconfiguring.

2.3.2 Meta-requirement 2 (MR2)

Bodily activities are concrete movements performed by humans (Oborn et al., 2011; Reckwitz, 2002) that provide data in smart service systems (Beverungen, Breidbach, et al., 2019; Beverungen, Matzner, et al., 2017; Lim & Maglio, 2018), particularly in the area of posture management (Seth et al., 2021). However, smart service systems can fail when users perceive smart objects as obtrusive and thus stop interacting with them (Wessel et al., 2019). This explains why effective smart service systems operate as invisible computers (Beverungen, Müller, et al., 2019).

MR2: Smart service systems need to operate as “invisible computers” that are unobtrusive to users.

2.3.3 Meta-requirement 3 (MR3)

Humans draw on their knowledge when acting in smart service systems, so these systems need to be designed for smooth integration with a user’s knowledge base (Beverungen, Müller, et al., 2019; Reckwitz, 2002; Wessel et al., 2019). For example, users need to understand how to change posture correctly (Bourahmoune & Amagasa, 2019). Therefore, backstage analytics that generate feedback from data, e.g., as a digital representation of a practice, must provide guidance that users easily understand.

MR3: Feedback and advice on changing practices must be easy for users to understand.

2.3.4 Meta-requirement 4 (MR4)

Humans exhibit emotions when interacting with a smart service system, such as fear, disgust, sadness or happiness (Beverungen, Breidbach, et al., 2019; Maglio et al., 2015; Wessel et al., 2019). Negative emotional reactions to a smart object cause smart service systems to break down (Wessel et al., 2019).

MR4: Smart service systems need to avoid stimulating negative emotions from their usage.

Following the DSR methodology (Peffers et al., 2007), we designed, demonstrated, and evaluated a smart service system to render the practice of sitting smart. We combined a T-shirt with sensors that continuously collected data about bodily activities to train a ML model that represented posture in real time and provided suggestions to change posture via a mobile application. Our design was guided by our initial design principles, a reflection on how we planned to make the practice of sitting smarter, and the fulfilment of our meta-requirements.

2.3.5 Design principle 1: smart objects

Backstage analytics build on configurations of sensors, ML models, wearable devices, and other artefacts (Beverungen, Matzner, et al., 2017; Beverungen, Müller, et al., 2019; C. Lim, Kim, et al., 2018; Lim & Maglio, 2018). Designing them in a human-centred way implies that these configurations need to measure ongoing practices and predict related problems comprehensively, accurately and reliably.

DP1 _initial: Human-centred smart service systems require technologies (e.g., sensors, ML models, wearable devices, etc.) that predict practices based on measuring bodily activities comprehensively, accurately and reliably.

2.3.6 Design principle 2: smart objects and bodily activity

Human-centred smart service systems operate invisibly in the background of the technical configuration (Beverungen, Müller, et al., 2019). Sensors are critical in this regard because they collect data from concrete practices that are supposed to become smarter (Lim & Maglio, 2018; Wessel et al., 2019). This is particularly important when collecting data from an individual’s back, where the sensors are placed close to the body (Kumar et al., 2021; Seth et al., 2021), making humans aware that their otherwise routinised behaviour is being monitored. For this reason, DP2 stresses that the integration of sensors alongside individuals’ bodily activities needs to be done unobtrusively.

DP2 _initial: Human-centred smart service systems integrate sensors with bodily activities in ways that users perceive as unobtrusive.

2.3.7 Design principle 3: smart objects and human knowledge

Human-centred smart service systems provide users with feedback that they easily understand based on their subjective knowledge (Albani et al., 2017; Breidbach & Maglio, 2016). In the healthcare context, medical terminology needs to be translated into everyday language that enables users to monitor and change posture (Bourahmoune & Amagasa, 2019; Ho & Ismail, 2021). For this reason, DP3 emphasises the need to design the feedback provided by our app in a way that users can easily understand and perceive as reliable and accurate (Breidbach & Maglio, 2020; Clarke, 2016).

DP3 _initial: Human-centred smart service systems provide feedback that users can easily understand.

2.3.8 Design principle 4: smart objects and emotions

Emotions are deeply human; therefore, taking them into account is an essential part of designing human-centred smart service systems (Beverungen, Müller, et al., 2019; Maglio, 2015; Maglio et al., 2015; Wessel et al., 2019). Prior research has shown that users can react negatively to smart service systems and that fear, anxiety or frustration can cause smart service systems to break down (Wessel et al., 2019). These emotions arise from how users perceive interactions between themselves and the smart object in day-to-day activities. Negative emotional reactions need to be avoided. This includes attention to designing remedies against annoyance over ill-suited clothing, fears of being continuously surveilled or resentments over information that is perceived as being misguided.

DP4_initial: Human-centred smart service systems provide a functional setup that avoids triggering negative emotions in users.

3. The DSR process

Following the DSR methodology (Peffers et al., 2007, 2018), we (a) designed a smart service system that contributes to dealing with posture-related problems, and in doing so, (b) expanded our knowledge of smart service systems (Gregor & Hevner, 2013; Gregor et al., 2020; Gregory, 2011; Gregory & Muntermann, 2014). Figure 1 shows in detail how we engaged with this DSR study. Our approach is consistent with the idea that DSR is less sequential but involves various iterations between design and evaluation (Hevner et al., 2004; Peffers et al., 2007, 2018). The box on the left side indicates how we conceptualised our artefact, formulated meta-requirements, and proposed initial design principles. The white cycles for demonstration and evaluation underscore the formative evaluation that accompanied the development of our prototype (Venable et al., 2016). Finally, the arrows pointing up and down in the lower part underscore the coevolution of the different cycles in our project. Next, we discuss how we carried out the DSR methodology in our study (see also Table 2).

Figure 1. The logic of our DSR study.

Table 2. Overview of our DSR project (following Peffers et al., 2007).

Identify problem and motivate	Sitting is a practice with a relationship to posture-related problems. It demands the design of human-centred smart service systems that focus on practices.
Define objectives of a solution	To improve the practice of sitting and to develop design principles for human-centred smart service systems that are useful also in other contexts.
Design and development	Design and development of a shirt, a machine learning algorithm, and a mobile app that provide real-time feedback for improving posture to humans.
Demonstration and evaluation	Naturalistic approach to evaluate the artefact in a human-centred manner through several rounds of expert interviews and user testing. The machine learning model was also evaluated via artificial evaluation.
Communication	Findings were communicated via research papers, workshops with designers and health insurance providers, as well as public relations and social media more broadly.

Problem formalisation and motivation were driven by a practical and societal problem. Sitting is a practice that is deeply connected to posture-related problems (Kumar et al., 2021; Seth et al., 2021; Zhao et al., 2021). From a conceptual point of view, designing a solution for this problem is interesting, because little work has been done to centre the design of smart service systems around a practice, i.e., make sitting smart (Wessel et al., 2019).

The definition of our design objectives followed a twofold rationale. First, we aimed to improve the practice of sitting. Second, we aimed to do this by developing design principles for human-centred smart service systems that had wider appeal and would also be useful in other contexts.

Interplays of design and development, as well as demonstration and evaluation, unfolded between 2016 and 2019. Specifically, a functional prototype of a T-shirt to be augmented with sensors for collecting data on an individual’s posture was designed and developed at a university in eastern Germany. The T-shirt is intended to feed data into a ML algorithm that predicts whether wearers will likely experience back problems given their sitting positions. Feedback is then provided to users via an app connected to the T-shirt.

As depicted in Figure 1, demonstration and evaluation went hand in hand with efforts to design and develop the artefact. We implemented the human risk and effectiveness strategy to evaluate our artefact (Venable et al., 2016) because most of the risks that threatened the effectiveness of our artefact were rooted in the interactions between humans and smart technologies (see MR2–MR4). Therefore, we evaluated interactions between smart objects and human emotions, knowledge and bodily activities through naturalistic evaluation. Furthermore, we evaluated the accuracy and reliability of our ML model by means of an artificial evaluation (MR1). We provide a more detailed account of the demonstration and evaluation of the artefact below, given the criticality of these issues for DSR (Venable et al., 2016).

We communicated continuously about the project by presenting it at our university and at various academic and practitioner conferences. To this end, we also set up an official website and arranged for media coverage during the project. In addition, the results of this study were further disseminated in the form of a university spinoff.

Theoretical contributions arising from this study are depicted in the box in Figure 1: grounded in our naturalistic evaluation, we fundamentally revised initially derived design principles. As we gradually learned over time, trade-offs affected our design choices and we gained insights about these trade-offs that we explicate in this article. The trade-offs are meaningful for making practices smart and guide thinking about preventing healthcare-related problems. This contributes to recent discussions about the self-management of chronic conditions in IS. Finally, our artefact contributes to discussions about digital solutions to posture management and provides design knowledge to further research initiatives.

4. A smart service system to improve sitting: key technological components

We designed several functional prototypes of our smart service system between 2016 and 2019. The project began with a core team of five people conceptualising a smart service system for making sitting smarter through engaging with experts in technologies for posture management, establishing a formal collaboration with a large public health insurance provider and founding a venture working on the artefact in 2019, with substantial funding from the German government.

DP1 focused on the backstage analytics underlying the smart object to capture and process data from an individual’s back that feeds into a digital representation of a user’s posture (DP3). We took several measures to design the smart service system accordingly. First, we selected a physical object that continuously interacted with users while they were sitting. Because users not only sit on chairs, let alone the same chair repeatedly, we opted against cushion-based solutions involving pressure sensors (Ahmad et al., 2021; Bourahmoune & Amagasa, 2019; Zhao et al., 2021). Instead, we opted to equip clothing with sensors, since it achieves close skin contact, regardless of the sitting surface. Consistent with the work by Seth et al. (2021), we equipped a T-shirt with five inertial measurement unit (IMU) sensors that capture posture by transmitting values for their position in space (i.e., the x, y and z axes, yielding a total of 15 data points). Second, DP1 relates to measuring current posture, while DP3 relates to representing current posture and predicting problematic posture. We trained a ML model with sensor data to measure posture and predict problematic posture accurately and reliably. To determine optimal sitting, we generated a test dataset by holding four eight-hour workshops with physiotherapists in December 2017. They helped us define categories of sitting positions that individuals occupy and label our data accordingly, yielding 59 possible postures (see Appendix 3). The combinations of these positions define an individual’s posture in a current situation.

To ensure the reliability of the model we developed per DP1, we standardised the data captured by the sensors. We developed an application that would capture sensor data in a one-second rhythm. To develop our model, in late 2017 and early 2018, one person adopted every posture (see Appendix 3) and we captured each posture for 30 seconds, yielding 30 entries for each posture. Furthermore, to ensure that our model generalises well, we decided to focus on users without severe health issues, such as osteoarthritis or multiple sclerosis. This was appropriate because even these persons showed remarkable differences in posture while standing or sitting.

Further implementing DP1, we used so-called gyroscope sensors because they can measure the rotation of a sensor on its x-, y- and z-axes (Freifeld & Black, 2012; Seth et al., 2021; von Marcard et al., 2017). This was promising because such measurements have been found to provide far-reaching insights into how a person is sitting, which is relevant to DP1 and DP3. As people often make small movements while sitting, it was also important that our sensors have high accuracy and detect micromovements (DP1 and DP3). Finally, our sensors would have to be small and lightweight to ensure minimum nuisances for users (DP2). Given these considerations, we opted for the MPU6050 sensor and the Arduino Pro Mini microprocessor to retrieve data from the sensors (Seth et al., 2021). The technology-driven rationale was that it would be compatible with the Arduino platform, which hosts a large developer community and offers libraries that we could use for our hardware and source code. We aimed to transmit data from the sensors to the microprocessor via Bluetooth because a Wi-Fi connection would consume too much power (for further technical information, see Appendix 1).

5. Demonstration and evaluation of our smart service system

Evaluation in the context of DSR comprises the subactivities of demonstration and evaluation (Nguyen et al., 2021; Peffers et al., 2018; Tuunanen & Peffers, 2018; Venable et al., 2012, 2016). Demonstration is a proof-of-concept evaluation that shows that an artefact can solve a problem in principle (Peffers et al., 2007; Tuunanen & Peffers, 2018; Venable et al., 2012). Evaluation is then supposed to reveal how well an artefact solves the problem (Peffers et al., 2007, 2018; Venable et al., 2012). In this section, we report on the demonstration and evaluation of our artefact (Nguyen et al., 2021; Tuunanen & Peffers, 2018; Venable et al., 2012) after we explicated three criteria that our artefact was supposed to meet while using multiple evaluation methods (see, e.g., Nguyen et al., 2021; Venable et al., 2012). Following Venable et al. (2012), we relied on three criteria:

• Does the artefact show utility and efficacy in terms of meeting its goals: Does the smart service system live up to the meta-requirements (see, e.g., Table 1)?

• Identification of the artefact’s weaknesses and areas for improvement: Can we identify potential areas in need of further development of the smart service system to meet the meta-requirements?

• Identification of unintended consequences of using the artefact: Do our efforts to meet meta-requirements disclose unexpected problems?

With these three derived guiding questions, we then formulated our evaluation strategy by relying on the “FEDS” framework introduced by Venable et al. (2016). The framework provides guidance on how researchers evaluate artefacts depending on the circumstances of a focal project. This implies that choices for evaluation strategies follow a four-stage choice process: (1) definition of evaluation goals, (2) choice of an evaluation strategy from a set of established evaluation strategies, (3) determination of properties to evaluate, and (4) design of the evaluation episodes. Table 3 summarises how we applied this four-stage choice process to our project.

Table 3. Choice process for ‘human risk and effectiveness strategy” (Venable et al., 2016).

(1) Explication of goals	(2) Choice of evaluation strategy	(3) Determine properties to evaluate	(4) Design individual evaluation episodes
Rigour:Major: improvement of social practice upon intervention (MR2-MR4) Minor: technical considerations, i.e., reliability and accuracy (MR1) Uncertainty and risk reduction: Major: social Risk residing in the breakdown of social practice (MR2-MR4) Minor: technical risk rooted in the functionality of the artefact (MR1) Ethics: substantial risk to users (MR2-MR4) Efficiency: easy access to students continuously engaged in sitting	Prioritization of risks: Risks associated with MR2-MR4 are social risks outweighing technical risks associated with MR1 This calls for human risk and effectiveness strategy	Major (MR2-MR4): Interactions between smart objects and (1) bodily activity, (2) knowledge, and (3) emotions of humans Minor (MR1): Accuracy and reliability of machine learning model and sensors	Major (MR2-MR4): Naturalistic evaluation via seven workshops with physiotherapists (2017/18); 12 user tests at our institution (2018), 19 field interviews (2018) Minor (MR1): Artificial evaluation via conceptualisation of our artefact together with technology experts (2016); user tests and experiments (2017); field experiments via nine workshops with physiotherapists

We opted for a human risk and effectiveness evaluation strategy involving a major share of formative naturalistic evaluation methods (of MR 2–4) and a minor share of artificial evaluation (Venable et al., 2016). In terms of defining our evaluation goals (see, e.g., Table 3), this strategy promised to be rigorous. We had to ensure that the data collection and processing in the ML model were accurate and reliable (MR1). We expected standard statistical procedures to help establish the proper technical functionality of our ML model, sensors and microprocessor. A major part of our evaluation demanded that we assess how well our artefact related to humans interacting with smart service systems in practice (MR2–MR4). The main sources of uncertainty and risk thus lay in how the social practice would unfold upon our intervention. Ethical considerations further amplified our choice, because our smart service system could imply substantial risks to users (MR2–MR4), so it was crucial to include users early in our evaluations. Furthermore, we consider any smart service system that seeks to intervene into a practice like sitting inherently human-centric. For that reason, we prioritised the risks associated with insufficiently fulfilling MR2–MR4 over the risks associated with MR1, which justifies the human risk and effectiveness strategy (Venable et al., 2016).

Against this background, we designed our evaluation periods to mitigate the risks associated with MR2–MR4 to the best possible extent (see Table 3). The major share of our efforts was devoted to naturalistic evaluation involving a prototype, user tests and interviews to assess the feasibility of our artefact (March & Storey, 2008). Figure 2 places our demonstration and evaluation efforts on a timeline, and Table 4 explicates the steps we took. The demonstration phase of our functional prototype was carried out between winter 2016 and 2017. For artificial evaluation, we interviewed posture management experts to understand the problem space and to consider potential solutions. We then furthered interaction with the empirical field through initial prototype and intensified user testing at our university, where our research team members wore the designed T-shirt to assess its function. Finally, we conceptualised which postures our artefact should have been able to measure via workshops with physiotherapists. We began to evaluate our functional prototype shortly before the end of 2017 via naturalistic evaluation. To evaluate whether our smart service system satisfied MR2–MR4, we engaged in nine workshops with physiotherapists, 12 user tests with members of our institution involving observation with interviews, as well as 19 field interviews that involved testing our prototype, as well as founding a spinoff venture in 2019 where we worked further on the evaluation of our smart service system.

Figure 2. Timeline of our demonstration and evaluation.

Table 4. Details about demonstration and evaluation, including primary data collected for these purposes.

Time	Role	Primary Data Sources
2016	Demonstration	Conceptualisation with technology experts (multiple interviews with researchers in the domain of posture management to understand the solution space)
2017		Prototype testing within own institution. User tests with three different members of the team (two male, one female) who assessed general functionalities for a proof-of-concept
2017		User tests on multiple prototypes across nine workshops with two physiotherapists (41 hrs of material)
2017/18	Evaluation	User tests on multiple prototypes across seven workshops with two physiotherapists (41 hrs of material)
2017		12 user tests on multiple prototypes with users in an office environment of a co-working space involving observatory techniques, think-aloud method and subsequent interviews
2018		19 field interviews
2019		Founding a venture in order to conduct user tests with prospective clients, as well as to improve posture models and mobile app.

5.1. Demonstration of the smart service system prototype: emphasis on MR1

The demonstration of our operational prototype began in the winter of 2016 and lasted for about a year. Developing a proof of concept for our smart service system took place from 2016 to 2017. During 2017, the key task was to technically implement the smart service system and evaluate its functionality using statistical means (MR1; DP1).

5.1.1 Learning from demonstration: utility and efficacy

Our smart service system was supposed to make the practice of sitting smarter. Therefore, it was necessary to collect data about how people sit and to be potent in intervening in sitting as a practice. Technical risks included whether our sensors, ML model, and microprocessor would measure and predict posture accurately and reliably in real time (DP1), thus displaying easily accessible resulting representations in an app (DP3). To get these issues right, we needed to centre the practice of sitting in our demonstration. Thus, we asked a male person (183 cm; 90 kg) to take all postures mentioned in Appendix 3 for five seconds. Two problems occurred: first, our initial training data did not generalise well to this person; second, he had limitations in his upper-body flexibility and some problematic preconditions.

Our second training cycle followed suit. We manufactured two T-shirts and gave them to the initial male and a second male (179 cm, 74 kg). We initially sampled only males, because we wanted to keep our test group small and anatomically comparable. Having these two individuals wear the T-shirts revealed 40 postures that were too difficult to track and hence were excluded from consideration (see Appendix 3). Finally, we developed two separate training and testing sets, which allowed us to combine the sets later and test the generalisability of our model (DP1). The third and last training cycle for this prototype occurred when we had our model’s final configuration in place and sought to generate more training data to improve accuracy. We included a third male person and one female person to test whether our model could be generalised to females (DP1). All of these efforts demonstrate how we positioned humans and their practices in a process of designing a smart object that is intended to learn gradually from human data.

Furthermore, the position of sensors for collecting posture-related data on the garment was important to obtain accurate measurements (DP1), effective interaction with the human body (DP2), and an accurate representation of posture (DP3). An eight-hour workshop with physiotherapists in October 2017 focused on this goal and enabled us to place sensors in those areas of the back where harmful movements were known to cause posture-related problems, i.e., the lower (lumbar spine) and upper back (cervical area). In addition, we learned that how individuals move their chests (thoracic area) provides a good approximation of how they move their entire bodies. Furthermore, we learned that our digital representation needed to cover more than just the lower back because office workers tend to hunch their shoulders when sitting in front of computers for prolonged periods. This also causes posture-related problems. Subsequently, we placed sensors across these broader areas.

5.1.2 Weak spots and areas for improvement in data collection and processing

Our efforts to train an ML model and position sensors revealed that some changes were due. To measure posture during the practice of sitting, we obtained data from five sensors (labelled a, b, c, d, e) that helped to assess movements of the back positions along the x-, y- and z-axes. However, the data showed high standard deviations in height (z-values), so we excluded z-values from our model and ended up with ten instead of 15 data points. The exclusion of z-values meant that our model could only measure posture in a two-dimensional space. We responded to this problem in multiple steps. First, we tried to measure how a person sat by positioning sensors in particular areas of the T-shirt and approximate posture in a three-dimensional space by combining the two-dimensional data provided by the sensors. Data were narrowed down to three important postures captured in a first back posture model: “neutral position” (Post_neutr), “both shoulders forward” (Post_shoulder) and “neck forward” (Post_neck). These were chosen because they are probably the most common positions when sitting in front of a computer. In other words, the positions are indicative of the potential harm that might arise from widespread sitting practices. Appendix 1 provides a detailed explanation of how we improved these models through an iterative process that eventually led to testing multiple ML methods that were able to predict posture. However, we realised that the models could be further improved. This is why our second step was to adopt a technique introduced by Freifeld and Black (2012): we began working with a simulation of a three-dimensional human skeleton upon which we placed virtual sensors. We then ran a series of simulations that generated data on posture and used this data to train our ML model. The advantage of this technique was that it allowed us to collect training data much faster and cheaper, and that it enabled us to shift sensors to different parts of the body within the software.

5.2. Naturalistic evaluation of a smart service system: emphasis on MR2–MR4

5.2.1 Learning from the evaluation: utility and efficacy

A naturalistic formative evaluation of our smart service system was carried out between late 2017 and throughout 2018–2019 (see Table 4). Consistent with our approach to the keep practice centre stage, we carried out twelve user tests. We conducted the first tests in a start-up co-working space in December 2017, where we observed users wearing our prototype in a natural environment. At that time, our prototype was a combination of an initial ML model, Arduino hardware and a T-shirt. In line with our practice-based perspective, we applied the think-aloud method, which is suitable for obtaining verbal reports on problem-solving (van Someren et al., 1994). Following this method, we asked users to continuously vocalise their thoughts and emotions freely and without interruptions as they engaged with our prototype for over 45–90 minutes. Users were asked to continue regular tasks in their offices during that time. To refrain from interrupting the thought processes and to mitigate observer biases, we provided only guidance on what the prototype was supposed to do and not how it was done. We took notes on the thoughts and interviewed users only afterwards to gain a better understanding of what they expressed while using our prototype. Wearing the T-shirt made us aware that the positioning of the sensors had an impact on the accuracy of our measures, because our sensors could only measure posture accurately when they were positioned vertically on the garment. This problem emerged as we observed users’ bodily movements and assessed how these movements affected the computational processes performed by the smart object. Users mentioned irritations while wearing the T-shirt, which affected how they moved. If bodily movements shifted the sensors, this could cause inaccurate posture measurements (DP1), leading to an unreliable representation of the practice that could not provide effective guidance (DP3).

The above-mentioned problem also affected DP2, which addresses the interaction between the smart object and the bodily activity of humans. User tests showed that users would note the materiality of the sensors because the sensors touched the users’ skin. Sensors were considered as an intrusion into the practice of sitting almost by default as users were not accustomed to feeling sensors. For example, when the test subjects sat on a chair or wore a backpack, both items pressed against the garment from the outside, pushing sensors against the skin and causing discomfort. Hunching or scratching, in return, would frequently displace the sensors, so they could no longer deliver accurate measurements (DP1 and DP2). Certain bodily movements also accidentally cut the connections between the sensors, power bank and microprocessor. In short, these user tests showed that the smart object’s computations would frequently come to a halt due to its interactions with the human body. To mitigate interruptions in computation caused by user’s actions in practice, we needed to stronger elaborate on the underlying technical service-oriented architecture of the smart object (by adopting a bounded service¹ —a binder – provided by Android that was then more loosely coupled with other services or activities). Furthermore, we followed users’ demands to choose artefacts of the smallest possible size to ensure that users did not recognise them (DP2).

We tested a second version of the prototype with physiotherapists. These tests revealed further limitations arising from bodily movements (DP2) in that they affected the reliability of how our ML model measures posture (DP1) and its resulting digital representations (DP3). In particular, the reliability of posture measures and associated predictions can only be determined relative to body weight and size. However, individuals differ in these regards, so we responded to this problem by integrating a calibration function (see Figure 4 in German language: Kalibrieren) that was implemented behind a button on the home screen of our app which would allow our ML model to determine a user’s normal posture. The function requests that users relax, sit straight and push the button. Additionally, prototyping and user tests in 2018 sensitised us to the importance of enabling individuals to start and stop measurements. To initiate this function, we implemented a “start/stop” button on the home screen that provided feedback to users on whether or not our smart service was active. When starting measurements, users would receive feedback about their current postures on the home screen regarding shoulders (Figures 3 and 4: Schulter), cervical spine (Figures 3 and 4: Halswirbelsäule), thoracic spine (Figures 3 and 4: Brustwirbelsäule), and lumbar spine (Figures 3 and 4: Lendenwirbelsäule). Our user tests revealed further limitations. Users pointed out that not holding their smartphones in one hand when calibrating it would make calibration easier. We responded by implementing a countdown that enabled users to put the smartphone aside and then perform calibration. As a result, users better understood what was being measured and that this measurement was specifically fine-tuned to them (DP3). Further feedback on this change from five informants convinced us that the function was perceived as informative and valuable. Figures 3 and 4 show the user interface of our app at different stages of design and development.

Figure 3. Front-end of our mobile application.

Figure 4. Screens of our mobile application.

Another problem that emerged from our user tests was related to how easily users understood the feedback that our app was giving to them. This was crucial, as both MR3 and DP3 require users to easily access the information displayed to them. However, our user interface lacked a clear and easily understandable statement that the current posture could have problematic consequences. Thus, we implemented a dashboard that would aggregate the time users spent in certain postures and allow users to alter the period reported (last hour, day, week or total time of use). A timeline feature visualised a user’s stack of motions on the x-axis and duration on the y-axis, with the diagram using different colours to highlight potential triggers of back pain. For example, yellow bars indicated that an individual’s head stayed in the same position for too long and red bars indicated that specific limits had been exceeded. We designed the app to send users a push notification once they reached a critical threshold. The notification came in two forms. One was a short message calling for a change in posture. The other was more detailed, explaining the reason for the warning and indicating the relevant area of the back and posture. Figure 4 shows the resulting app screen displays, and Figure 5 shows the push notifications (in German).

Figure 5. Push notifications sent by our mobile application.

MR4 and DP4 target the relationship between smart objects and emotions, a critical relationship for the human-centredness of smart service systems (Wessel et al., 2019). Our initial smart object was designed to avoid negative emotions linked to posture-related problems in the long run and to avoid negative emotions that would occur while using our app. However, interviewing users about their emotional reactions to wearing the T-shirt and engaging with the mobile application revealed striking insights. Most users did not perceive a need to prevent posture-related problems because they had never experienced them before. Only one person had experienced them and they saw clear value in our artefact. This problem surprised us, as our project was built around the idea that the prevention of posture-related problems would be valuable.

Additionally, users reacted with hesitancy towards the smart object because it did not evoke positive emotions. It turned out that our focus on avoiding negative emotions, which led us to design transparent information, was insufficient for users to see value in our smart service system. Users perceived notifications sent out by the system as useful, but as “no fun at all”. This inhibited users’ willingness to adhere to the recommendations. Thus, we redesigned the user interface to evoke positive emotions by enabling individuals to specify an ideal posture based on their measurements, which we implemented through backstage analytics (DP1). Furthermore, we incorporated game-based elements that provided feedback into the mobile application’s user interface so that users could see how close they were to attaining an ideal posture (DP3 and DP4). Instead of merely reducing the amount of information sent to users and thereby reducing the potential for negative emotions, we learned that this information had to be sent in a different format so that it would trigger positive emotions. These positive emotions are intended to motivate users to act in accordance with the guidance of our smart object.

Further interviews conducted throughout 2018 then revealed problems with how individuals cognitively processed information provided to them, despite above-mentioned changes. We learned that descriptive statistics on posture were perceived as too technical. Instead, the users recommended that the app resemble the user’s perception of their physical body as closely as possible when presenting information about posture. Therefore, we decided to display a 3D image of the body that would move according to how the individuals moved (see Figure 6). We implemented a digital representation in which historical data (in the form of a heatmap) depicted areas of the back at risk. Additionally, feedback indicated a lack of mental stimuli. Our interviews revealed that the mobile application should provide features that suggest to users what to do next (DP3).

Figure 6. 3D image produced by our mobile application.

5.2.2 Identification of unintended consequences of using the artefact

Through evaluating our artefact, we encountered several unintended consequences arising from the practice of sitting while wearing the T-shirt with its sensors and while engaging with the mobile application. Interviewing and observing users disclosed important insights that reinforced some DP while undermining others. We could detect these issues because our naturalistic evaluation helped us to unpack partially problematic relationships between our design principles that emerged only after we had evaluated how our artefacts affected the practice of sitting. The ways in which humans related their emotions, knowledge, and bodies to the artefacts (Reckwitz, 2002; Wessel et al., 2019) disclosed that the implementation of certain design principles placed constraints on the implementation of other design principles (see below). These constraints made us aware that designing human-centred smart service systems based on a “practice view” (see above) was important, but also characterised by a process of trading-off various technical considerations against different elements of practices (Reckwitz, 2002; Wessel et al., 2019). Against, this background we revised our design principles as trade-offs and elaborate on our revisions next. Similar to Seidel et al. (2018), we summarise our revisions in Table 5.

Table 5. Revised design principles and justification for revisions.

Revised design principle	Trade-offs	Justification for revision
DP1rev (initial DP 1 and DP 2 were merged into DP1) Human-centred smart service systems should balance between the goals of producing complete data and disrupting practices with data collection.	Performance trade-off: Trading off the necessary completeness and accuracy of required data with possibilities to collect that data.	Evaluation showed that the pursuit of DP2 constrained pursuit of DP1. The solution was to merge DP1 and DP2 in order to address a trade-off that matters for designing smart service systems.
DP2rev (formerly DP 3) Human-centred smart service systems should provide a digital representation of practice resembling a user’s visual perception of that practice instead of providing encompassing information requiring expert knowledge.	Knowledge representation trade-off: Trading-off subjective visual perception of practice with objective assessment through complete and accurate data.	Evaluation showed that users demanded a digital representation without expert terminology. This was striking in our context as it contradicted our initial DP3, where it seemed important to tell users exactly what the healthcare-related problem is.
DP3rev (formerly DP 4) Human-centred smart service systems require triggering positive human emotions in the short-run in order to avert adverse consequences in the long-run.	Temporal trade-off: Trading off the long-term avoidance of negative emotions with triggering positive emotions in the short term.	Evaluation showed that without prior encounters with posture-related issues, users do not typically perceive the avoidance of posture-related problems as important in the long run. Instead, our app should be fun to use in the short-run in order to avoid posture-related consequences in the long-run.

DP1 initially highlighted that human-centred smart service systems would require technologies capable of predicting practices based on measuring bodily activities comprehensively, accurately, and reliably (see Table 1, above). However, our naturalistic evaluation of practices of sitting made us aware that DP2 and DP3 placed important constraints on the potential to implement DP1. In particular, DP2 prescribed that designing human-centred smart service systems calls for integrating sensors with human bodily movements unobtrusively. In fact, the underlying meta-requirement called for smart service systems to act as “invisible computers” (Beverungen, Müller, et al., 2019, p. 9). The naturalistic evaluation of practices of sitting revealed that implementing DP2 constrained implementing DP1 for several reasons. Sensors, power banks, and microprocessors placed close to the human body to collect data were frequently perceived as nuisances. Likewise, bodily movements could displace sensors or disrupt connections between different parts of the configuration of our smart service system which stopped data collection. However, DP1called for collecting data about the sitting practice as completely as possible in order to train our model and accurately predict posture. This required placing as many sensors as possible close to the body. However, the more sensors there were, the larger was the risk that users would perceive them as disturbances or that bodily movements would displace sensors. In conclusion, the first trade-off that we developed from our study and that matters for designing human-centred smart service systems relates to the performance of the artefact as there is a performance trade-off between the necessary completeness and accuracy of the required data and the possibility of collecting that data from practices. As the trade-off relates to crucial prescriptions made by DP1 and DP2, we merged these DP into our revised DP1_rev addressing the performance trade-off.

DP1rev: Human-centred smart service systems should balance between the goals of producing complete data and not disrupting practices with data collection.

Furthermore, our initial DP3 called for human-centred smart service systems to provide feedback that users could easily understand via a user interface, in our case, a mobile application. Our naturalistic evaluation made us aware that while this prescription might seem straightforward, it is hard to achieve in practice. Our artefacts collected data on posture and our model predicted problematic posture based on a ML model that was developed with the help of user data and expert knowledge from physiotherapists (see above). It was our expectation that this knowledge would be valuable to users in order to adapt their practices of sitting in order to prevent chronic back pain. It turned out, though, that users would not understand expert terminology and considered descriptive statistics about their posture less valuable. Instead, they suggested that seeing 3D images of their bodies would make it easier to understand what was problematic about how they were sitting. This struck us as we were designing in a health care-related context where expert knowledge matters for helping humans change their behaviours. However, it was particularly this expert knowledge that made it challenging for users to understand what they could do differently. Instead of using expert language in textual form, users demanded visual depictions of their body that closely resembled their subjective perceptions of how they were sitting. This made us aware that while the knowledge that underlies a human-centred smart service system needs to be comprehensive and based on expertise, the representation of such knowledge should be as easily understandable as possible. Therefore, we understood that there was a trade-off between the knowledge base that mattered for our human-centred smart service system versus how that system would represent the knowledge to humans. In turn, a second trade-off that matters for designing human-centred smart service systems is a knowledge representation trade-off between collecting complete data on posture (DP1) and providing a subjective visual perception of the practice to the user via digital representations. Therefore, we revised DP3 into DP2_rev.

DP2rev: Human-centred smart service systems should provide a digital representation of practice resembling a user’s visual perception of that practice instead of providing encompassing information requiring expert knowledge.

Finally, our initial DP 4 stressed that human-centred smart service systems would provide a functional setup that would not trigger negative human emotions. In the long run, our smart service system was supposed to avoid negative emotions of back pain. However, concrete interactions between our artefacts and humans should also not cause anger, sadness, or resentment in the short run. Our naturalistic evaluation revealed that our considerations were proverbially “off point” as humans were expressing major emotional issues arising from how our artefact affected the practice of sitting. Somewhat ironically, informants did not see value in preventing back pain in the long run through sitting differently as long as they would not acutely suffer from back pain. Counter to our expectations, prevention itself was not perceived motivating enough to changing posture through interacting with our artefacts. Informants also asked to make our app more fun and integrate gamification into our app with the aim of spurring engagement with our smart service system. We learned multiple lessons related to our initial DP4 from our naturalistic evaluation. First, our initial DP4 had a one-sided focus on avoiding negative emotions (see above, Table 1) while our evaluation made us aware that focusing on promoting positive emotions was seen as more valuable by humans instead. Second, not only did it matter to attend to promoting positive emotions instead of avoiding negative, it also mattered to understand the temporal horizons linked to emotions. If our human-centred smart service system was supposed to contribute to avoiding negative emotions in the long run, designing it needed to attend to promoting positive emotions in the short run. Taken together, we developed these observations from our naturalistic evaluation into a temporal trade-off that matters for designing human-centred smart service systems because it highlights the importance of triggering positive emotions in the short run in order to avoid negative emotions in the long run. This was why we revised DP4 into DP3_rev.

DP3rev: Human-centred smart service systems require triggering positive human emotions in the short run in order to avert adverse consequences in the long run.

6. Discussion and implications

Our research question was: “how should human-centred smart service systems for posture management be designed?” To answer this question, we designed a T-shirt augmented with sensors measuring posture, feeding data into a ML model, and providing feedback to humans about how to change posture via an app. Conceptually, we built on a practice-based approach to smart service systems (Beverungen, Müller, et al., 2019; Maglio & Lim, 2016; Wessel et al., 2019) because our study addressed health problems resulting from the practice of sitting. This “practice view” of smart service systems enabled us to locate our meta-requirements and design principles on the level of concrete interactions between smart objects and the human body, knowledge, and emotions which were key analytical categories drawn from practice theory (Reckwitz, 2002). Going beyond earlier work on posture management in health care related fields, computer sciences, and human-computer interaction literatures that highlighted primarily technological considerations of designing posture management systems, we could unpack general design implications that arise once we take practices seriously in designing human-centred smart systems for posture management. Multiple rounds of naturalistic and artificial evaluation (Venable et al., 2016) of our smart service system revealed that design principles for smart service systems are best considered as “trade-offs” due to several unanticipated consequences arising from practices. Our study accentuates, for instance, how an expanded collection of user data would increase the performance of smart service systems in terms of high-performing self-learning algorithms, but it would also disrupt the underlying practice and, therefore, counteract the performance improvements rooted in technology.

Our study highlights three trade-offs that matter for designing human-centred smart service systems: a) performance, b) knowledge representation, and c) temporal trade-offs. The performance trade-off highlights the potential necessity to degrade the technological performance of smart service systems to enable unobtrusive integration of sensors with a given practice. The knowledge representation trade-off highlights situations where a smart service system has to be built on encompassing expertise, however, the integration of the system in a practice demands to represent that knowledge in very simple ways. Designing in this way may mean to design representations that leave out important information that doctors would tell their patients in a personal conversation or a video call. Nevertheless, as our naturalistic evaluation showed, humans are unlikely to understand or even value the suggestions of a smart service system if it displays expert terminology that humans have difficulties to understand. Finally, the temporal trade-off brings a long-term perspective to designing smart service systems. Particularly, our smart service system was supposed to avoid negative emotions both momentarily and in the future. However, evaluating how the system integrated with users’ practices of sitting showed that humans demanded triggering of positive emotions instead of avoiding negative emotions.

These trade-offs and our design knowledge, more generally, are based on a “practice view” of smart service systems. Therefore, they offer an important complement to design knowledge of smart service systems based on an “elements view” (Beverungen, Müller, et al., 2019; C. Lim, M.-J. Kim, et al., 2018; Lim & Maglio, 2018; Maglio & Lim, 2016). Designers are best advised to consider either or both of them in lieu of the smart service systems that they have at hand. This leads us to discuss our contributions to scholarship in the IS field and beyond.

6.1. What the ‘practice view’ means for designing smart service systems

Our first contribution is to extend the stock of design knowledge about smart service systems in IS research. Extant design knowledge has been built largely on an elements view that enabled scholars to fruitfully advance our knowledge about how human-centredness of smart service systems can emerge from machine learning, federated learning and self-learning algorithms in general (Beverungen, Lüttenberg, et al., 2017; Huber et al., 2019; Klör et al., 2018; Knote et al., 2021; J. Y. H. Lee et al., 2020). Such algorithms help to configure various other elements of smart service systems, such as humans, machines, organisations, and others, for mutual fit and benefit (Badinelli et al., 2012; Barile & Polese, 2010; Calza et al., 2015; Demirkan et al., 2016; Polese et al., 2016; Spohrer et al., 2012, 2017). Smart service systems are seen to become human-centred from data being fed into computational operations that learn and become smarter over time (Huber et al., 2019; J. Y. H. Lee et al., 2020).

Against this background, our design knowledge complements these earlier works in that it emphasises that human-centeredness of smart service systems arises from interactions between humans and artefacts in smart service systems. Designing them, therefore, demands considering interactions between computational operations and human bodies, knowledge, and emotions (Oborn et al., 2011; Reckwitz, 2002; Wessel et al., 2019). This is important because, as our study showed, these interactions bring to the fore “trade-offs” that matter for designing smart service systems. Attending to these trade-offs and to practice theory more generally is helpful when designing smart service systems that closely interact with human bodies, knowledge, and emotions. For example, while the elements view of smart service systems would likely underscore our initial DP1, our DP1_rev highlights that designing for completeness and accuracy of data is important for designing self-learning algorithms, but this goal has to be traded-off against potential disruptions of a practice arising from collecting data through sensors. Our evaluation showed that humans tend to consider sensors a nuisance when sensors make body contact as humans are likely to terminate wearing a shirt or moving their bodies in ways that lead to misrepresenting the practice in the data. This is why DP1_rev is particularly well suited to inform designing smart service systems that have many touch points with human bodies, such as those involving sensors in the garment or close to the skin. In contrast, other smart service systems, such as those involving sensors in machinery, may well benefit from the “elements view” discussed above (Beverungen, Müller, et al., 2019; Klör et al., 2018) as the sensors in these smart service systems have little body contact.

Furthermore, DP2_rev reveals a trade-off between expert knowledge that underlies smart service systems and the representations of conclusions drawn from that knowledge. Humans have difficulties in understanding and processing expert terminology that they do not understand, even when this knowledge might provide comprehensive information about their practice. Humans involved in our evaluation requested visual representations that were easy to understand and did not contain many details. This was striking as one might assume that patients would expect a doctor in a physical setting to explain these details to a patient. However, health care is not the only domain where expert knowledge matters and smart service systems are expected to greatly benefit humans (Demirkan et al., 2015, 2016). Finance, the law, smart contracts, and other areas of life also involve expert knowledge and are industries experiencing an upsurge in smart service systems (Lim & Maglio, 2018). Therefore, the general takeaway from our study is that designing smart service systems in these areas can strongly benefit from considering our DP2_rev. This is important because far-reaching transparency and the need to explain the output of sophisticated technologies to humans can bring legal demands that designers must carefully consider when designing smart service systems.

Finally, DP3_rev relates to emotions, particularly to trading-off avoidance of adverse consequences in the long-run with stimulating positive emotions in the short run. For example, the user interface of our smart service system frequently interacts with users by offering them advice on how to adapt the daily practice of sitting. While this is an important practice, it is largely subconscious. Adding artefacts to this practice makes users conscious of sitting. Designing a smart service system that seeks to adapt practices in this fashion requires cautiously considering emotional responses in the short- and long run. Other smart service systems seek adaptations of more conscious human practices and/or require lesser interactions with user interfaces. A smart washing machine, for instance, only occasionally interacts with a human as these machines can automatically purchase washing powder, thereby rendering purchasing as conscious human practice largely subconscious (Beverungen, Müller, et al., 2019). Therefore, DP3_rev is particularly meaningful for designers of those smart service systems where human interaction with the smart object is frequent and the smart service system makes users conscious of a practice that was more or less subconscious before.

To summarise, our revised design principles highlight the importance of attending to practices when designing smart service systems. This “practice view” of smart service systems offers an important complement to design knowledge based on the “elements view” of smart service systems by showing the trade-offs that designing smart service systems involve and that are rooted in practices. A general takeaway for designers of smart service systems is that they can choose between the “practice” or the “elements” view according to three criteria: (a) the degree to which the smart service system has close contact to human bodies, (b) the degree to which the smart service system is supposed to translate expert knowledge into easily accessible user advice, and (c) the frequency by which users will interact with smart objects via a user-interface that makes humans conscious of a practice that was carried out subconsciously before.

6.2. How the ‘practice view’ of smart service systems helps with designing for the prevention of chronic conditions

Our second contribution is to literature on digital solutions for the management of chronic conditions that are becoming increasingly recognised in the IS field. Patients need to manage chronic conditions such as diabetes, overweight, or chronic back pain by generating and acting upon data in their everyday lives outside hospitals (Bardhan et al., 2015, 2020; Dadgar & Joshi, 2018). Thus far, the emphasis in this literature is on how digital solutions help with the self-management of chronic conditions (Brohman et al., 2020; Dadgar & Joshi, 2018; Savoli et al., 2020). Our study advances this literature in a meaningful way as it offers design knowledge for systems potent of prevention. This is important for various reasons.

First, prevention follows a logic that emphasises avoidance of patients developing a problematic condition, whereas earlier studies attended to scenarios where patients need to come to terms with a condition that they already have (Barrett et al., 2016; Dadgar & Joshi, 2018; Q. B. Liu et al., 2020; Liu et al., 2020; Thompson et al., 2020; Wessel et al., 2019). The design challenges of the associated technologies are different because users need to see value in avoiding something that is not perceived as an acute and continuously pressing problem but instead requires functions that trigger positive emotions that motivate adaptation of the underlying practice.

This is why, secondly, leveraging a practice-based view for the design of smart service systems helps to unpack shifts in design principles. In our case, prevention had to be designed through entertaining user interfaces using gamification that could not rely on displays of medical information. By leveraging a practice-based view and our design principles in particular, researchers and practitioners can better understand how they can design smart service systems that help address these new challenges that IS research is well positioned to speak to.

6.3. Advancing digital solutions for posture management through designing smart service systems based on the ‘practice view’

Research on technologies that speak to the improvement of sitting or other forms of sedentary behaviour has been thus far largely developed in health-related disciplines, engineering, computer science, and associated fields. Our final contribution to this research is specifically residing in the artefact we designed together with the design principles and meta-requirements developed around it. Our practice-based perspective sensitised us to choose particular sensors better suited to capture data from the human body than pressure sensors, which were widely used in this work (Ahmad et al., 2021; Bourahmoune & Amagasa, 2019; Zhao et al., 2021). Likewise, leveraging a practice-based view informed choices to design a shirt in order to reach the best possible coverage of the data needed to occasion changes in posture, and not a cushion or a single wearable that other works relied on (Seth et al., 2021). In short, while much of the earlier work in this area offers largely technology-based design knowledge, our study opens a new terrain by highlighting practices as important and deriving human-centred meta-requirements and design principles that earlier studies could not offer. In addition, we also offer a concrete approach to evaluating these systems more naturalistically, a call that has been voiced in this literature as well (Bourahmoune & Amagasa, 2019; Roggio et al., 2021).

6.4. Implications for policymakers and practitioners

Our study also holds several implications for policy and practice. In terms of policy, researchers in medicine and public health have highlighted the need of policies that would enable individuals to better manage their posture (Ammar et al., 2020; Dunstan et al., 2011; Mahdavi et al., 2021). However, the interest here is typically in nationwide policies that abstract from the details of the concrete practice of sitting. Our study puts the practice of sitting centre stage and highlights how important it is to understand the minute details of this practice in order to arrive at effective interventions. This means that it would be valuable to develop policies that take the situated details of the sitting practice seriously. For example, complementary to formulating nationwide guidelines, it could make sense to allocate budgets more strongly to micro-level studies that explore sitting and other problematic practices in order to understand in detail how these practices play out. This matters strongly because health care is typically an area where the potential of smart technologies to positively affect health outcomes remains under-exploited. One reason for this could be that policymakers and practitioners need to attend more to practices as many health problems play out through what people do daily. After all, if health care is to improve nationwide or even globally, this change will start with people changing what they do in their everyday lives. Designing smart service systems based on the “practice view” will be instrumental in supporting such change.

Furthermore, our study calls for a shift in perspective when discussing the design of smart technologies for posture management. Computer scientists have paid important attention to how to develop particular components of such smart technologies, such as underlying models (Ho & Ismail, 2021; Kumar et al., 2021; Liaqat et al., 2021). Our practice-based approach highlights that the technological rationales underlying these technologies may oftentimes generate a potent smart object; however, it may under-deliver in practice. Because these smart technologies are supposed to be implemented in practices, the analysis and consideration of practices matter strongly for smart service systems in posture management. As our evaluation showed, taking the very notion of “practice(s)” seriously uncovers many operational aspects of smart service systems that engineering rationales typically do not consider. Thus, like researchers, industry practitioners will benefit from paying close attention to the concrete practices in which humans interact with smart service systems. While this may demand complementing engineering and science-driven rationales that are important in this industry, it will likely be beneficial for the products and services that can be offered on the markets.

6.5. Limitations and future research

Despite the rigour of our work, it is not without limitations, which provides ample opportunities for future research. First, we acknowledge the situatedness of the phenomenon that we study and the limits to the generalisability of our design principles. In consonance with van Aken et al. (2016), while not subscribing to the notion of universal generalisability, the design principles we advanced are transferrable as potential design knowledge for a similar class of problems within a corresponding problem space. The design principles thus enable analytical transferability but not statistical generalisation (A. S. Lee & Baskerville, 2003, 2012). We nonetheless call on future research that explores the applicability of insights from this study in related contexts (for example, Meth et al., 2015). Such endeavours have the opportunity to sharpen, extend or improve the fidelity of the knowledge advanced by our study for future scholarship.

Second, we have made pragmatic choices regarding the demographics of the sample of subjects that we engaged in developing and evaluating our smart service system. While this does not invalidate our findings, it points to an opportunity for future research to expand the scope of scholarly knowledge in this area by engaging with a wider spectrum of subjects. The room to further illuminate the calibration of knowledge in this regard is broad as different variables can be considered ranging from gender, age and health profiles to geographical distinctions. We believe this is still a greenfield that is ripe for future research to explore.

A third limitation that calls for future work relates to designing artefacts that meet the highest data privacy standards. This is particularly important because devices that collect data on the everyday practices of individuals make these individuals transparent to service providers, who may or may not act in their best interests. We addressed this by providing users with a consent declaration according to which their data would be processed solely by our technologies. As our artefact, unlike other apps, did not process user profiles or data in other ways than to provide feedback on posture, further technical implementation of data security was not a central concern for our project at prototyping stage. This, however, does not rule out the general importance of data privacy, which future studies should address (Dickhaut et al., 2023). Relatedly, it is also important to note that studies such as ours lend themselves to multidisciplinary collaborations between scholars across different fields as it draws on multiple expertise and domains. The fact that our study emphasises a balancing between human-centredness and technology-centredness in a field of interest to healthcare professionals points to the need for such study designs. We therefore call for future research to draw on the complementarity of insights that can be gained from multidisciplinary collaborations attending to pertinent issues such as chronic diseases in moving the frontier of practical and scientific knowledge forward.

7. Conclusion

We all sit too much, in this day and age, where remote work is increasingly becoming the norm. Our comprehensive design science research study charts new territory in terms of designing smart service systems and digital solutions for posture management as we designed and evaluated a T-shirt enhanced with sensors and a mobile application that together produced measures and recommendations to change posture. In doing so, we learned that any attempt to change a practice like “sitting” asks for a “practice view” on designing smart service systems. We lay out how the design knowledge resulting from this view differs from design knowledge drawn from the established “elements-view” and discuss how and when to decide between them.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/0960085X.2024.2308541.

Notes

1. https://developer.android.com/guide/components/bound-services (Accessed December 11^th 2023)