Abstract
Behavioural design has emerged as an important means of shaping behaviour change. Realising such change often necessitates complex combinations of multiple behaviour change interventions and multiple design artefacts. Yet, it is unclear how current practices should be adapted for such contexts. Hence, we ask: How can complexity be addressed in a behavioural design process? Through an in-depth case study, we develop several propositional insights. We theorise Complex Behavioural Design (CBD) in systemic terms and differentiate this from typical behavioural design processes by: (i) the progression of design at multiple abstraction levels requiring different design capabilities (high dealing with vision and goals; mid dealing with the behavioural/technical system; and low dealing with interventions and artefacts), (ii) the presence of both top-down and bottom-up reciprocal interactions across levels, and (iii) the need for mid-level design coordination. These significantly extend current discussions in behavioural design and point to propositional directions for future research.
Behavioural design has emerged as a distinctive approach to shaping positive, ethical behaviour change, in areas ranging from health to security and sustainability (Niedderer et al., 2017; Voorheis et al., 2022). Realising change in such areas often necessitates complex interventions with multiple behaviour change techniques and designed artefacts (Niedderer et al., 2014; Schmidt, 2022). In this context, researchers in design (Bay Brix Nielsen et al., 2021), persuasive systems (Oinas-kukkonen & Harjumaa, 2008), and implementation science (Pfadenhauer et al., 2017) have all emphasised the need for coordination across design elements over time.
In response to this need, authors such as Pfadenhauer et al. (2017), Movsisyan et al. (2019), Schmidt (2020, 2022), and Maier and Cash (2022) call for specific behavioural design processes tailored to complex intervention development. For simplicity, we refer to such processes as Complex Behavioural Design (CBD), where complexity is defined by the number of elements and interactions to be considered. Here, Pfadenhauer et al. (2017), Schmidt (2022), and Maier and Cash (2022) provide general principles, emphasising the need for flexible, longitudinal design processes, where interventions can be iteratively developed, implemented, integrated, and refined over time in context. However, current behavioural design theory, processes, and practices both limit their discussion of complexity (Bay Brix Nielsen et al., 2024) and typically emphasise more deductive, linear approaches (Hagger and Weed, 2019; Schmidt and Stenger, 2021; Voorheis et al., 2022). For example, while seminal guidance by Michie et al. (2015) and the OECD (2019) acknowledge complexity, their process discussion is focussed on specific interventions. Bay Brix Nielsen et al. (2024) highlight how this creates a disconnect between often complex behavioural problems/solutions and current behavioural design theory and processes. More pragmatically, behavioural design teams are responding to complex problems in practice despite this gap in current process theory. It is thus ambiguous how current practices could be adapted to or implemented in CBD processes. Hence, we formulate the Research Question (RQ): How can complexity be addressed in a behavioural design process?
Due to the lack of theory and empirical accounts of CBD, we adopt an explorative, qualitative, theory building approach, supported by a single in-depth case study. More specifically, we introduce the case of the Samsung Electronics design team as a representative CBD environment. Here, the team confronted complexity in coordinating interdependent interventions and human behaviours as well as in coordinating their own development processes. As such, this case provided a critical site for developing theoretical propositions, by allowing us to examine the intersection of taskwork (the specific actions taken to design the intervention) and teamwork (how different team members interact) in CBD. Through this approach we develop several significant contributions. Specifically, we theorise CBD in systemic terms and differentiate this from typical behavioural design processes in three main ways: (i) the progression of design at multiple abstraction levels requiring different design capabilities (high dealing with vision and goals; mid dealing with the linked behavioural and technical system; and low dealing with specific interventions and artefacts), (ii) the presence of both top-down and bottom-up reciprocal interactions across levels, and (iii) the need for mid-level design coordination in order to realise system outcomes and leverage the diverse capabilities within the team. These significantly extend current discussions of CBD and point to propositional directions for future research.
Background and initial conceptual framework
To ground our investigation, we develop an initial conceptual framework based on current understanding in behavioural design and then contrast this with more general discussions of complex design processes.
Behavioural design, complexity, and initial conceptual framework
Behavioural design is characterised by a focus on behaviour as the primary object of design, coupled with an explicit grounding in behavioural theory and demand for explicability in both process and outcomes (Khadilkar and Cash, 2020; Michie et al., 2015). In terms of the design object, complexity is typically acknowledged in four main areas: (i) behaviour and technology in context (Maier and Cash, 2022; Pfadenhauer et al., 2017); (ii) the combination of behaviour change techniques in interventions (e.g. combining feedback and priming into a single app) (Bohlen et al., 2020; Michie et al., 2015); (iii) the combination of interventions, potentially across multiple users (e.g. intervening with managers, nurses, patients, and others) (Aunger and Curtis, 2016; Michie et al., 2015: p. 131); and (iv) the evaluation and maintenance of intervention(s) and behaviour(s) over time (Movsisyan et al., 2019; Schmidt and Stenger, 2021). For example, West and Gould (2022, p. 38) emphasise the need ‘to create prototypes and evaluate these before going into full scale production’ because it is not feasible to predict the impact of all the factors involved in complex behaviour change. In addition, the process of design can itself be complex, based on the number of activities and interactions being coordinated and how these relate to the problem and solution (Eppinger et al., 1990). This has resulted in growing calls for processes tailored to CBD (Pfadenhauer et al., 2017; Schmidt, 2022).
However, most processes, practices, and theory in behavioural design are still geared towards more deductive, linear approaches, which are difficult to align with the greater flexibility needed to deal with complexity as discussed by Bay Brix Nielsen et al. (2024, p. 521) and others (Schmidt, 2020; Voorheis et al., 2022). This takes three main forms. First, many of the most widely adopted processes and models limit complexity by tightly bounding their scope to increase explicability and scientific rigour. This is evident in, for example, the work of Michie et al. (2015) and the OECD (2019) that focus on specific interventions but position their processes with respect to a wider cycle of development, piloting, evaluation, and implementation. Second, there are several more iterative or systemic processes that accommodate complexity, such as those provided by Fogg (2009), Wendel (2013), Aunger and Curtis (2016), and Oinas-Kukkonen and Harjumaa (2009). Yet, here, questions remain as to how these processes should progress and how teams should coordinate the multiple elements needed to address complex contexts. For example, Fogg’s (2009) eight-step process allows for iteration between all steps but provides little insight into how this might then progress considering differing levels of complexity. Third, Aunger & Curtis (2016) and Pfadenhauer et al. (2017), both highlight the increased importance of team coordination as well as integration of diverse stakeholders and capabilities when dealing with complex behaviour. Further, Schmidt (2020) and Cash et al. (2022) highlight challenges regarding the need to balance design and scientific concerns, and how this can create tension amongst involved stakeholders. However, such issues are not addressed in current process literature. Thus, there is a need to better understand both what the team is doing (i.e. taskwork) and how the team is interacting (i.e. teamwork) (Crawford and Lepine, 2013) in CBD.
Drawing current understanding together, Figure 1 illustrates the progression of a ‘typical’ behavioural design process. Here, the process is formulated as a response to a definable behavioural problem and results in a specific behavioural solution, which is complete and testable. The process itself is characterised by a sequence of steps balancing design and scientific concerns but otherwise not differentiating team roles, tasks, coordination, or capabilities. Thus, while complexity is sometimes acknowledged in the object of design it is not typically acknowledged in the process of design. This provides an initial conceptual framework for our investigation. Here, it is important to note that many practitioners are already engaged in CBD. Hence, Figure 1 highlights a disconnect between current process theory and lived practice, where processes fail to capture the ‘messy’ realities of process coordination. Initial conceptual framework illustrating current thinking regarding the progression of a ‘typical’ behavioural design process.
Complex design processes
While behavioural design is a distinct context, initial directions for investigation can be drawn from the wider literature on complex design processes. These have long been discussed in the context of products, services, and systems (Haberfellner et al., 2019; Tromp and Hekkert, 2018; Ulrich and Eppinger, 2015). For example, human–systems integration (HSI) (e.g. Boy, 2020) and social design (Tromp and Hekkert, 2018) explicitly address design processes that coordinate human and technological elements. Here, all processes respond in some way to the underlying issue that object and process complexity reduce individual’s and team’s ability to understand the totality of the problem or solution (Flager et al., 2014; Yu et al., 2016).
This has led to several recurring process elements in this context. First, bounding process complexity by differentiating developmental task- and teamwork across abstraction levels. For example, Graessler et al. (2018) illustrate this in their discussion of the multi-level, systems engineering v-model. Second, outcomes are achieved over time through coordination of processes across abstraction levels. For example, this is typical of service design where user experiences, products, and systems must work in synergy (Ulrich and Eppinger, 2015). Finally, technical development work is inextricable linked to the coordination of the team itself (teamwork) (Collopy et al., 2020). For example, Kreye et al. (2022) describe how complex engineering teams split work across different aspects of the project, whilst retaining higher level coordination based on the overall aim and understanding of the system. However, it is not obvious if or how these process elements might develop in the behavioural design context or how they might interact with the distinct behavioural design practices outlined in Introduction. Hence, it is necessary to ask (our RQ): How can complexity be addressed in a behavioural design process?
Methodology
To answer our RQ, we use an exploratory, in-depth, single case study for four main reasons. First, CBD processes, especially dealing with interactions between developmental and team coordination, have not been previously conceptualized. Hence, there is a need for in-depth insight to support of theory development (Eisenhardt and Graebner, 2007; Yin, 2018). Second, such processes are relatively rare, complex, and reflect an extreme context for behavioural design. Hence, they require the focused, in-depth investigation made possible by a single case approach (Eisenhardt and Graebner, 2007; Flyvbjerg, 2006). Third, understanding coordination of diverse teams across potentially different abstraction levels requires rich contextual information, which is often only accessible via case study (Yin, 2018). Finally, case study methodology has formed the basis for numerous prior studies of more typical behavioural design processes (e.g. Niedderer et al. (2014)). Our approach thus allows for comparison to, and the theoretical development from, these prior works.
Case selection
We purposively sampled a single project in Samsung Electronics based on theoretical and practical criteria (Yin, 2018). First, due to the nature of CBD the number of organisations with experience in this area is relatively low. Thus, we identified Samsung Electronics as a world leading organisation versed in the development of complex interventions with multiple behavioural and technical components. The established nature of Samsung Electronics helped minimise possible confounds associated with technical difficulty, organisational commitment, or understanding of the intervention context commonly affecting innovative projects (O’Connor and Rice, 2013).
Second, the identified project provided a strong example of CBD linked to changing behaviour around smart products. Specifically, we identified a project conducted by the Design Corporate Centre at Samsung Electronics in Seoul, Korea, over the course of 16 weeks. Importantly, one target of this project was improved quality of life via helping users leverage the capabilities of smart products. For example, one element of the project involved help users understand how they could improve their home air quality through purchasing and/or bundling combinations of products compared to individual products on their own. Here we opted for a case focused on general population users as most representative and least likely to introduce additional confounding variables. This project had many of the characteristics associated with CBD and even focused on aspects of behaviour, such as home air quality found in some prior studies (Bay Brix Nielsen et al., 2018; Pfadenhauer et al., 2017). It presented significant manifestations of all four areas of object complexity identified in Introduction as well as to a more limited degree process complexity in the number and organisation of tasks in the development work over time. Further, the project was successful, and its outcomes were published (Hwang et al., 2022, 2025), mitigating possible confounds associated with practices leading to project failure. Hence, while the focus of the project is a-typical for behavioural design research (which often focuses on applications such as sustainability or health (Voorheis et al., 2022)) the characteristics of the project support external validity.
Third, we had in-depth access to all documentation, team members, and the second author acted as a team member, providing behavioural science capabilities. In total, five designers and eight project members from Kookmin University collaborated on the project. While individuals had differing capabilities (focused on management, design, engineering, and behavioural science), they together formed an experienced team. Importantly, this diversity of capabilities is typical when dealing with complex behavioural problems (Introduction) and helps provide a context in which team coordination is both necessary and visible in the process. This provided rich, consistent data access across the whole pre-, during-, and post-project timeline. Thus, by minimising possible confounds and providing an outstanding example, this case helped increase external validity and provide a foundation for analytical generalisation and theory development (Robson and McCartan, 2011) relevant to other CBD contexts.
Data collection
We collected empirical data about the project from multiple sources concurrently during the project as well as after completion (Yin, 2018). To ensure internal validity and reliability we used a protocol to help elicit evidence and analyse the data (Gibbert et al., 2008). First, the second author kept detailed notes throughout the project as part of their role in the team.
Interviewee summary.
Third, secondary data was collected, including the Request For Proposal (RFP), eight weekly meeting slide decks, a midterm presentation slide deck, and a final presentation slide deck. Other secondary documentation and informal discussions with key participants before, during, and after formal data collection were further added. All data collection was structured across the project timeline to give insight into temporal dynamics within the project and facilitate process analysis. This not only allowed us to triangulate insights from multiple sources (Yin, 2018), it also allowed us to reach saturation in data collection and analysis (Yin, 2018).
Data analysis
Following Langley (1999), we adopted a three-step, iterative approach to building theory from process data.
First, we divided the project into five stages emerging from temporal bracketing based on the tasks performed. In the first stage – project mobilisation – (pre-project and week 1), the team reviewed benchmarking cases and decomposed the project into two sub-tasks. In the second stage (weeks 2–4), based on the project scoping the team mapped the interacting behaviours and technologies involved in the project. In the third stage (weeks 5–9), the team developed behavioural interventions and selected three which might be effective. In the fourth stage (weeks 10–15), the team designed experiments by transforming behavioural interventions into designed artefacts and conducted field experiments. In the fifth stage (post-project) the final report was delivered, and the team shared learnings. Throughout this process the team navigated at least three major complexity related challenges. First, they managed significant complexity in translating broad psychological concepts (e.g. goal priming) into multiple linked design interventions. Second, behavioural mechanisms manifested in multiple design forms. For example, one such mechanism (empathy gap) could be triggered via teaser text, page structure, or multimedia pacing. Third, there were systemic issues related to aligning interventions with the layered structure of the product ecosystem. This required integrating the interventions with bundles composed of hardware, services, and platform features. Selecting three overall interventions allowed the team to explore a representative subset of design possibilities across different behavioural domains while maintaining experimental clarity. The interventions served as focal points through which the broader design system could be evaluated.
Second, to draw out process dynamics we organised all data across the project timeline, before picking out the individual progression of each team member. Team member processes were established based on the first-hand concurrent notes, conversations, observations, and project logs of the second author coupled with the retrospective interview data. Here, due to differences in involvement across the project timeline some team members were primarily represented in the early or late stages, with two members working across the full five stages. This process data formed the basis for our first order codes (Gibbert et al., 2008; Miles and Huberman, 1984), we then compared and contrasted processes to elicit more abstract insights, resulting in second-order codes and aggregate themes.
Third, we performed an inductive/deductive analysis by comparing our abstracted insights with existing theory from complex design (Introduction), moving iteratively between data and literature to ensure conceptual validity and insight (Miles and Huberman, 1984). We based this initially on systems design and then expanded our theoretical scope to include other process elements as outlined in Introduction. This enabled us to bring out the nuances in our data and elaborate relevant theoretical contributions (Langley 1999).
Findings
We first outline the overall progression of the project to contextualise our insights (Overall progression of the project), before exploring how complexity was handled in the design process more specifically (Exploring design of complex behavioural interventions). To ensure analytical rigour, we distinguish between the five project stages, which reflect the terms used by the Samsung team, and the three abstraction levels and other analytical constructs, derived by the research team via our analysis.
Overall progression of the project
The project progressed through five main stages. In stage 1 (mobilisation: pre-project, week 1), the team worked to scope the overall objective of the project. Their aim was to help users make informed, goal-aligned decisions about the purchase and use of smart products. Initially, they reviewed existing cases such as Nike+. However, this approach proved inadequate due to the range of goals users could pursue as well as the often undefined or unarticulated behavioural purpose. As a result, the team realized that the benchmarking approach could not address the behavioural and systemic complexity of the project and would not provide an actionable design direction. To accommodate this, they decomposed the overall process into two main sub-tasks (stages 2 and 3).
In stage 2 (first sub-task: mapping, weeks 2–4) the team aimed to map the complexity of the project by adopting Goal Systems Theory (Hwang et al., 2022; Kruglanski et al., 2018). This helped to structure understanding of the multiple involved behaviours by organising goals and means hierarchically, with each goal relating to multiple sub-goals and means. In developing this understanding, the team specified target users, their top-level goals, and how each goal and associated sub-goals could be met via services connecting multiple products (i.e. object complexity associated with the behaviours and technologies in context, see Introduction). For example, one user group was parents with children ranging from 0 to 9 years old. Here, the team specified a top-level goal as childcare and sub-goals as air care, sleeping care, and safety care, to name a few. Then, they organised ‘bundles’ by matching each sub-goal with a service that leveraged multiple existing products. For example, the air care service comprised an air conditioner, an air monitor, and an air purifier. Hence, this stage resulted in the mapping of a complex system with behavioural and technical dimensions, from which points for behavioural intervention could be derived.
In stage 3 (second sub-task: development, weeks 5–9) the team worked to encourage users to make appropriate decisions about smart technologies, such that they could achieve their desired top-level goals. This was complicated by the cost implications of purchasing multiple technologies. Specifically, users face a trade-off between purchasing as a bundle versus individual items that cannot achieve the user’s goals. For instance, purchasing a smart air conditioner alone was suboptimal because users paid a premium for smart functionality but could not leverage its full potential without complementary products such as an air purifier or sensor. In response, the team focused on aligning user goals with appropriate bundles and enabling more effective, informed decisions. To do this, they mapped out the full behavioural sequence involved in the user journey of the smart product decision-making from key steps and tasks to barriers and solutions (e.g. behavioural interventions) as illustrated in Figure 2. This process generated a broad range of potential behavioural interventions tied to technical elements in the system (i.e. object complexity associated with combining multiple techniques and interventions, see Introduction). Steps, tasks, barriers, and solutions of a user journey drawn from the team’s weekly meeting materials.
After intensive cross-functional discussion, the team selected three overall interventions for experimental testing: goal priming, empathy gap, and mental accounting. Each was paired with a specific smart product bundle and mapped to a key moment in the user journey. To implement these interventions, the team used validated techniques from behavioural research. Goal priming relied on short, targeted sentences. Empathy gap was evoked through pacing and framing. Mental accounting was activated by adjusting how pricing was presented. All interventions were designed to be subtle, using minimal changes in text, layout, or sequence.
In stage 4 (Testing, weeks 10–15), the team translated conceptual behavioural interventions into experimentally testable interventions. They selected three product bundles, each designed for a specific parenting context such as improving home air quality, supporting children’s sleep, and facilitating early learning. Each bundle served as the basis for one behavioural experiment, aligning respectively with three behavioural interventions: goal priming, empathy gap, and mental accounting (i.e. object complexity associated with combining multiple techniques and interventions as well as their evaluation in context, see Introduction). The team then designed and conducted three randomized controlled experiments to test the effectiveness of these interventions, which revealed mixed effectiveness.
In stage 5 (Reporting, week 16 and beyond), the team synthesized experimental findings and translated them into actionable design principles for Samsung Electronics. The reporting process emphasized how to manage behavioural complexity by specifying the intended user (such as parents of young children), defining psychological goals (such as safety, sleep, or learning), selecting and structuring bundles accordingly, and aligning these elements with appropriate behavioural interventions and design artefacts. Three main recommendations emerged. First, it was essential to identify realistic and meaningful goals based on user needs. Second, focussing on service-led value help clarify the focus of interventions. Third, tailoring user journeys supported more effective design strategies, accounting for perceptual and cognitive biases.
Immediately post-project, Samsung Electronics applied these insights by revising its bundle pages. New versions of the interface presented specific user goals, framed them in emotionally resonant ways, and introduced products sequentially or grouped them strategically. These changes mirrored the experimental logic developed during the project. A published example of this application can be found in Hwang et al. (2022), which documents how goal-based bundling principles were implemented in real product displays. Roughly 2 years later, further changes appeared on Samsung’s SmartThings website. Products that were previously presented in generic lists began to be grouped by user-centred goals, such as exercising home or taking care of pets. These groupings reflected behavioural insights like goal priming and were accompanied by design features intended to trigger behaviour. Several of the design patterns developed during the project appeared in this new interface.
These results serve as important context for our subsequent insights in two main ways. First, they confirm the overall framing of the project as behavioural design. Specifically, despite containing elements of UX and other design and marketing approaches, the central focus of the project and its outcomes was on behaviour change grounded in behavioural theory (Introduction). Second, despite similarities to typical behavioural design processes, several significant – complexity driven – process elements emerged that challenge our initial conceptual framework (Figure 1). Thus, these results provide a solid basis for further theory development in relation to our RQ.
Exploring design of complex behavioural interventions
Our analysis revealed three major themes that capture how the design team managed complexity during the project.
Development at multiple abstraction levels
Immediately evident in analysing the focus and activities of the individual team members was that development comprised a process with three main abstraction levels: high dealing with vision and goals; mid dealing with behavioural and technological systems; and low dealing with specific interventions and artefacts.
High-level: This involved development directed towards ensuring that the project would lead to desirable post-project outcomes. Initially, these outcomes were encapsulated in a general vision, and the goals were unclear, requiring substantial interpretation and alignment. As Designer 4 explained: ‘I thought what Samsung wanted was vague. So we kept trying to go into various directions—this and that—and I think we ended up getting it right … but because what they wanted was ambiguous, we had to pinpoint the problem they wanted so that to get the results from that problem’. Initially the corporate vision focused on sales, but as development progressed, the team – through research and testing – transformed the vision from company-driven to user-driven, emphasising the goals of individual users. Manager 1 described how the team shifted the project: ‘[Samsung] should think backwards about what users want to buy, instead of trying to package and send the products [Samsung] want to sell’. Critical to this progression was the evolution of behavioural theory from an external tool to a central organising logic. This enabled the team to structure problems and solutions around the desired long-term outcomes for users and mediate the sales-focused demands of the wider organisational context. Manager 1 and Designer 2 illustrated the importance of this development ‘I always say take a user-centric approach, but they [Samsung] can't help getting stuck in their own business’ (Manager 1) but ‘Knowing these theories plays a very big role in presenting the rationale for design… thinking about the long-term strategic goal’ (Designer 2). Several members of the team, particularly Managers 1, 2, and 3 spent significant time working at this level. Hence, development at this level was essential to defining what constituted desirable long-term, post-project outcomes.
Mid-level: This involved development of the system architecture in behavioural and technical terms as well as combining these into a cohesive whole. Behavioural theory played a central role, helping the team members break down overall goals into a structured behavioural and technical system. For example, Manager 3 stated that ‘we don't break down the user journey from providing information to payment. We can't see it in practice… but by tearing it all up and showing it, [we realised that] these processes are not one’, and Manager 4 that ‘our content changed from Samsung.com and SmartThings to bundling’. In this way the team was able to iteratively constitute behavioural and technical systems organised through Goal Systems Theory. This led to a gradual refinement of the system-level framing through which more specific theories and artefacts could be coordinated, as illustrated by Manager 4 ‘At the very beginning, I felt like I had to choose something out of so many options. Later, I considered Goal Systems Theory and picked up behavioural economics mechanisms related to it’. The importance of the systems perspective was also highlighted by Designer 5 ‘In terms of behavioural economics techniques, people point out problems one by one … but [we] think about the overall experience of the whole’. Hence, development at this level provided a structured way of translating overall vision and goals into manageable sub-tasks and coordinating individual behavioural and technical contributions to the system.
Low-level: This involved development of multiple distinct behavioural interventions and associated artefacts. Design at this level involved multiple parallel processes each following a typical behavioural design structure in terms of identifying a specific behavioural problem, using this to define relevant behaviour change techniques, and then designing and testing artefacts that embody these. Work involved detailed design of individual elements of the overall system. For example, goal priming, empathy gap, and mental accounting were designed and tested by additional mobile pages. The relationship between this low-level work and the higher levels is explained by Designer 3, who also emphasises the significance of this way of thinking in comparison to typical design work: ‘Eventually I redesigned the app and website based on the strategic direction that came out of this project… It was the most impactful project for me that year’. Designer 2 further elaborates this ‘the results of this project were a little different from what we first thought because it started from the configuration strategy that we set’. Here each separate sub-design involved specific behavioural theory: ‘Through the experiment, the results got more reliable … The parts that came out differently from the initial intentions did change my expectations’ (Designer 3). Several members of the team, especially Designers 1, 2, and 3, worked substantially at this level, focussing on detailing artefacts that could express behavioural interventions in context. Hence, development at this level was characterised by the parallel design and testing of individual system elements prior to their integration in the overall system.
Together these findings illustrate a process of development including at least three distinct levels of abstraction, with parallel but related design work on: vision/goals, behavioural/technical system, and behavioural interventions/artefacts.
Reciprocal interaction across levels
Following from this multi-level development was the emergence of distinct reciprocal interactions between levels. This served to connect work across levels from both top-down and bottom-up perspectives as well as foster non-linearity in the overall development activity via iteration on one level based on insights gained at another level. Two major modes of interaction were observed in the data, a bottom-up ‘M’ and a top-down ‘W’, where M and W denote reciprocal linkages between levels across the duration of the project.
First, the bottom-up M captures a general shift in attention as follows: low – high – low – high – low (see Development at multiple abstraction levels for more about the levels). This strategy was evident in, for example, the work of the designers who initially approached the project from an artefact-focused perspective (Designers 1, 2, and 3). Typically, this type of interaction fostered a focus on aligning the details of the project, ensuring that low-level decisions about individual theoretical lenses and intervention features were aligned with the overall goal. For example, Designer 2 highlights how they gradually aligned their work with the bigger picture outcomes of the project: ‘my main focus was on the kind of services we could provide using our devices, how to plan experiences, and how to design UX for them. I hadn't [initially] thought about these experiences leading to sales’. This linkage also allowed for insights garnered by specialists in design and behavioural science from low-level work to be pushed up to higher levels and influence the overall direction of the project. For example, Designer 3 highlights how ‘we were originally focusing more on how UX should be designed … but as the project progressed, it became a much larger scope that could touch everything before the user, both in the problem area and the solution area’ and Manager 2 that ‘I think [the design work] solves the problem of which products we should we use to test behavioural economies and mechanisms, so it reduces the scope of problems we want to solve’. Beyond just alignment of detailed design decisions and overall vision this process also substantially influenced the way the team members thought about and measured success. Instead of focussing on company goals like selling more products, they thought about how customers might behave differently. This was evident in secondary data and observations from the project where there was a shift from typical project management and corporate measures to ones informed by behavioural theory and empirical analysis, and geared towards pro-user perspectives. Ultimately the bottom-up interaction was an important element in keeping the detailed design work aligned with the overall project vision and goals whilst still allowing for in-depth behavioural design activities.
In Figure 3 we provide an annotated illustration of those members of the team that predominantly displayed this kind of interaction in their design process. This is based on secondary data, concurrent field notes and observations, as well as the interviews. While not all team members followed the exact same pattern the reciprocal, and iterative bottom-up push of information and then alignment of direction are common features denoted by the overlayed trend. Bottom-up ‘M’-type interactions in design activity across the project timeline with illustrative quotes.
Note, the curves in Figures 3–5 represent a qualitative mapping of the design process. The trajectory is a smoothed line between discrete data points (key quotes and observations) categorized by project stage and abstraction level and connected by the narrative provided by interviewees in the interviews themselves, the documentation from the secondary data, and the observations of the research team. Further, the y-axis reflects the conceptual depth of the activity rather than a quantitative count of codes. Top-down ‘W’-type interactions in design activity across the project timeline with illustrative quotes. Annotated illustration of mid-level design coordination linking high- and low-level development, and behavioural and technical perspectives.
Second, the top-down W captures a general shift in attention as follows: high – low – high – low – high (see Development at multiple abstraction levels for more about the levels). This interaction was evident in, for example, the work of the managers who initially approached the project from a high, corporate perspective (Managers 1, 2, and 3). Typically, top-down W fostered a focus on achieving the overall goals and vision for the project and ensuring that decisions taken at lower abstraction levels contributed to this and did not become too narrowly fixated on detail issues. This was key because the complexity at the low-level otherwise created problems in understanding and leadership. For example, Manger 2 explained how the scope of the project was gradually refined through interaction across levels: ‘First of all, the scope of the problem that the designers brought was too wide and too unclear… but as we went on, we got into a very specific problem to solve such a problem’, which is further elaborated by Manager 4 ‘So later, it has been narrowed down to focusing on the goal … it got a little easier to research after focusing on Goal Systems Theory because we had to work with mechanisms that fit that [system perspective]’. At the same the upward shifts allowed team leaders to better understand and respond to the theoretical and detailed insights being generated at the mid and low levels, adjusting approach to ensure the overall system was cohesive and aligned across its myriad elements. For example, Designer 4 explained how: ‘I realized through this project that there is always a goal for the users when they do something or behave in a way … I remember constantly tracking why people aim at this behaviour’. Similarly, Designer 5 explained how different elements were aligned via mid-level use of Goal Systems Theory: ‘with this theory, I realized that it is the way to solve the project while understanding the problem in the beginning … we started with this very big category, but eventually we narrowed it down, adding other papers and theories’. This interaction also fostered deeper understanding of the detailed issues associated with behavioural design that were not part of the team leadership’s typical expertise area and was hence also important for upward and outward communication between the team and other stakeholders. For example, Designer 2 highlighted that ‘When we design a service, there’s not just a UI, there are all the stakeholders and interests in the background and it’s not just to design the products and services but there’s a need to connect everything to the business model’. Ultimately the top-down interaction was an essential element in the reciprocal coordination of work across levels and team members with different capabilities. This was central to coordinating the multiple theories and interventions as a cohesive whole in the final system.
In Figure 4 we provide an annotated illustration of those members of the team that predominantly displayed this kind of interaction in their design process. This was again based on multiple, concurrent, and retrospective data sources. While not all team members followed the exact same pattern the reciprocal, and iterative top-down alignment of work formed a common pattern denoted by the overlayed trend.
The importance of these reciprocal interactions is further supported by those team members who were less prone to shifting their focus. This resulted in their work becoming narrow in its focus and highly domain dependant. This hindered their interactions with the wider team and held back their design work. Manager 3, a businessperson, exemplified this challenge. They reflected on the communication gap between designers and marketers, noting that ‘designers interpret it from the designer’s point of view, and marketers interpret it from the marketer’s perspective, so I don’t think this way is going to be very communicative’. In contrast, other members who moved between high-level and low-level developed a deeper understanding of the project’s purpose and contributed more effectively. For example, Designer 2 expressed how shifting perspectives changed his view of design itself: ‘I used to have times when design itself was my goal. But as I worked through this project, I realized that design shouldn’t be the goal. Design is a methodology for doing something’, and further that this needed to include logic from other aspects of the project: ‘you have to consider the market… since the management has a view close to the market, it will be very helpful when communicating with them’. In addition, interactions intensified at critical junctures in the project. For example, multiple interactions were evident when the team was aligning the system with relevant theoretical frameworks. Similarly, in the early stages (1 and 2), shifting between levels helped connect the project’s vision with its behavioural logic and design structure. By the time the team entered the experimental stage (3), however, those alignments were largely settled and hence interactions lessened. The focus then turned to interpreting the results of each intervention and ensuring they made sense in relation to the project’s overall objectives (stages 4 and 5).
Together these findings draw a strong link between reciprocal interaction across levels and productive developmental progress. Further, the two modes of interaction (bottom-up M and top-down W) served to both differentiate and connect team members with different roles, capabilities, and major foci in the project. This made M/W interactions integral to effectively progressing design work across the levels of vision/goals, behavioural/technical system, behavioural interventions/artefacts.
Mid-level design coordination
Key to mediating the above noted interaction was mid-level design coordination, focused on alignment of behavioural and technical systems. This served as a bridge, allowing lower-level artefacts to be mapped to system-level effects and high-level goals and vice versa. Further, this emphasised coordination and alignment of interaction between lower-level behavioural theory, behavioural interventions and change techniques, and embodied technical artefacts.
This coordination served to anchor both interaction modes (M and W), while also contributing to the design process through the application of a behavioural systems lens at the mid-level. In particular, the use of Goal Systems Theory at this level helped coordinate the multiple, more specific theories and behavioural change techniques implemented across interventions at the low-level. For example, Goal Systems Theory helped structure the logic of the whole user journey, while behavioural economics mechanisms such as goal priming, empathy gap, and mental accounting were selectively applied across different artefacts. Manager 2 (with specialist behavioural understanding) explained that within the project Goal Systems Theory structured low-level interventions as follows: ‘as long as we understand the intentions correctly, these behavioural economics mechanisms can be applied later when reducing the gap between action’. This structuring role was also very strongly evident in secondary data and field notes, where maps of goal systems were used alongside similarly structured technical mappings, such as objectives trees and function/means hierarchies; bridging the design of technical and behavioural systems and coordinating both aspects of low-level design work. This played a critical role in determining how specific artefacts would work together as a coherent system. Rather than evaluating individual artefacts in isolation, mid-level coordination focused on whether and how they contributed to the overall behavioural goals and system effects. Importantly, this perspective shifted the criteria for design success – from artefact-level behaviour focused, results to holistic, systemic outcomes.
Mid-level design coordination was also essential to translation between high-level market and design concerns and more specific low-level theory, and behaviour change techniques. This was a key element in coordination between more managerial roles within the team, focused on overall project vision, and more designerly roles focused on the details of behavioural intervention development. For example, Manager 4 explained how mid-level theory and coordination helped structure interventions: ‘I considered Goal Systems Theory and picked up behavioural economics mechanisms related to it. … We thought about what mechanisms were there to act as facilitators to help trigger things’. In contrast, Designer 4 noted how mid-level coordination prompted the team to rethink how they were using specific behavioural economics interventions: ‘SmartThings itself has devices that can create more synergy if products are sold [together]. So I think that’s why we proposed the configuration strategy in the categorization based on that goal’. Beyond this mid-level design coordination was also salient in linking specific intervention evaluations, to whole system evaluation, and overall interpretation of project success. Specifically, mid-level theory was essential to the assembly and interpretation of data from the multiple experiments carried out on individual interventions, into an approximation of the whole system’s likely impact. This is illustrated by example of data aggregation and interpretation via behavioural systems and was noted by Designer 3 as providing ‘a great tool that enables designers to conduct various experiments with confidence, providing the basis for designers’ intuition with existing research or theory’. This was strongly evident in secondary data and field notes where results were compiled and presented within and then beyond the team.
Ultimately, these findings draw out mid-level design coordination as fundamental to structuring the multiple specific interventions and artefacts towards a cohesive vision, as well as aligning the behavioural and technical components of the final system. Figure 5 provides an annotated illustration of key points where mid-level coordination helped connect high- and low-level development as well as behavioural and technical perspectives.
Discussion
Our findings serve to distinguish behavioural design for ‘complex’ interventions from current, ‘typical’ representations of behavioural design processes (Figure 1) in three main ways: (i) the progression of design at multiple abstraction levels requiring different design capabilities (high dealing with vision and goals; mid dealing with the linked behavioural and technical system; and low dealing with specific interventions and artefacts), (ii) the presence of both top-down and bottom-up reciprocal interactions across levels, and (iii) the need for mid-level design coordination in order to realise system outcomes and leverage the diverse capabilities within the team.
Insights and implications
Hence, we answer our RQ (How can complexity be addressed in a behavioural design process) via these insights, which we qualitatively visualise in the context of a CBD process in Figure 6. These have specific implications for CBD – and point to propositions that can form the basis for future work – as well as connecting to wider discussions in the design literature. Stylized visualisation of a Complex Behavioural Design (CBD) process across three levels of abstraction, with ‘M’ bottom-up and ‘W’ top-down iteration mediated by mid-level design coordination.
Multi-level design
First, we find that to deliver a complex intervention, design needs to progress in terms of – at least – three main abstraction levels. At the high-level the focus was on effective (re)framing of the project towards desired post-project and longer-term outcomes considering the ‘big picture’ vision for what is to be achieved. At the mid-level the focus was on alignment of the behavioural and technical systems and the coordination of individual interventions to achieve synergistic system level effects. At the low-level the focus was more aligned with typical behavioural design work dealing with interventions, change techniques, and specific artefacts. These parallel design foci and approaches are in distinct contrast to prior behavioural design literature that has typically either treated these foci sequentially or focused predominantly on progression at the interaction/artefact level (Introduction) (what we observed at the low-level). This insight thus highlights the significance of continuing high- and mid-level design work in explicitly guiding the development and emergence of systemic outcomes as well as coordinating behavioural and technical systems, which have previously been treated implicitly within behavioural design frameworks (Bay Brix Nielsen et al., 2024; Schmidt, 2022). Hence, we suggest that: Proposition 1a: For Complex Behavioural Design (CBD) to be effective, low-level intervention design needs to be aligned with on-going mid-level coordination of interventions and design of the behavioural/technical system as well as high-level project design and framing.
Further, these levels leverage distinct capabilities within the team and demand differing perspectives on the process and practices needed. At the high-level team members leveraged strategic design, reframing, and managerial capabilities and were essential to facilitating communication within and beyond the team as well as working towards longer-term, post-project outcomes. At the mid-level team members leveraged systems design and design coordination capabilities and were essential to combining multiple interventions, change techniques, and artefacts into a cohesive whole. At the low-level team members leveraged both behavioural science and design skills and were essential to ensuring explicability and efficacy of individual interventions. While the observed low-level capabilities closely align with prior work on behavioural design (Cash et al., 2022; Schmidt, 2020; Voorheis et al., 2022) the high- and mid-level capabilities have not been widely discussed. Authors such as Aunger and Curtis (2016) and Wendel (2013) have noted the need for skills such as team management, decision-making, and creativity, but have not previously explored their role in the process itself. Our findings thus point to the need to significantly diversify the capabilities considered relevant for CBD, as well as the need to distinguish their roles, foci, and interactions across the duration of the design process. Hence, we suggest that: Proposition 1b: In Complex Behavioural Design (CBD), each level of abstraction requires different design capabilities that need to be aligned through mid-level coordination to progress the overall process effectively.
Cross-level reciprocity
Second, the observed design process was reliant on reciprocal interactions between levels. These connected bottom-up and top-down perspectives in the design team as work progressed in the design process. This helped to scaffold effective iteration by aligning high-level vision and thinking about intended post-project effectiveness and outcomes with low-level detail and within-project tests of efficacy. This is an important development from prior behavioural design literature because it explicitly differentiates team roles and the need for a system’s logic in the organisation of the design team and process itself. This is something that is central to, for example, systems focused design processes (Haberfellner et al., 2019; Tromp and Hekkert, 2018; Ulrich and Eppinger, 2015), but not previously incorporated into behavioural design (Introduction). Further, these findings highlight how the different capabilities discussed in Multi-level design need to be composed and coordinated for the team to function. This significantly changes the more typical, implicit focus on behavioural science capabilities or the evident stage-based ‘hand-over’ logic of ‘involving creatives’ only in the final stages of form-giving design work (see e.g. Aunger and Curtis (2016) or Michie et al. (2015)). Most notably this highlights the need to integrate team capabilities and role considerations into behavioural design process and theory, which currently focuses mainly on the object itself (e.g. the selection of change techniques) or on the basic tasks to be carried out. Task organisation and process dynamics cannot be considered in simple, linear terms in complex contexts. Hence, we suggest that: Proposition 2: In Complex Behavioural Design (CBD), there is need for a layered processes model incorporating reciprocal top-down and bottom-up linkages to support the increased complexity of team roles, development work, and outputs.
Mid-level systems configuration and coordination
Third, we find that to make this type of multi-level design process work, mid-level design coordination is needed to realise system outcomes and leverage the diverse capabilities within the team. This reflects a combination of coordinating overview of behavioural and technical systems, capabilities of the team members, and structure of the process. As such, mid-level coordination connects both work across levels (high and low) and across technical and behavioural perspectives, to ensure the design progresses towards an aligned and cohesive system. The salience of type of coordination and use of mid-level systems-oriented behavioural and technical theory has only recently been hinted at by authors such as Schmidt (2022), but has not, to date, been elaborated in the process context. We find that this coordination is key to linking specific interventions, change techniques, and artefacts to overall outcomes. This constitutes a systems focused role and capability within the team, which is distinct from typical descriptions of behavioural design teams (that tend to focus on behaviour change techniques and interventions (Introduction). Further, the importance of this role in mediating more strategic design and reframing (at the high-level) versus more detailed intervention/artefact development (at the low-level) suggests a potential means towards resolving tensions between design and behavioural science perspectives in this context (Bay Brix Nielsen et al., 2024; Schmidt, 2020; Voorheis et al., 2022). This also highlights how, in CBD, process and team facilitation is needed alongside and in conjunction with the design of the overall system. Hence, we suggest that: Proposition 3a: In Complex Behavioural Design (CBD), mid-level design coordination is essential to realising high-level outcomes through low-level intervention development by synergistically coordinating and aligning design work around a cohesive system addressing both behavioural and technical perspectives. Proposition 3b: In Complex Behavioural Design (CBD), mid-level design coordination capabilities are needed to mediate the relationship between outcomes focused, high-level framing and strategic design and efficacy focused, low-level intervention development and detailed design.
Limitations
This research has two main limitations. First, due to the formative nature of research in this area our study built on a single in-depth case. The depth of our data provides a strong basis for analytical theory development and elicitation of propositions (hence the adopted paper structure) but is also limited by only providing a single successful account. While aspects of the observed case are aligned with some prior empirical accounts of behavioural design and meet relevant sampling criteria additional purposively sampled cases could reveal further insights. Further work is needed to evaluate the developed propositions from this study. Notably, it could be valuable to contrast cases with more and less effective design processes in this context as well as cases dealing with other application areas.
Second, we acknowledge that single case analysis can lead to bias, especially in our focus on CBD. Within this specific context we have worked to de-bias our study at each stage by using sampling criteria, clear analytical protocols, and a theory engaged approach. Further, by formulating concrete propositions we provide a foundation for testing of our analytical conclusions in future work. However, beyond the context of CBD it is less clear to what degree our findings might translate to other complex or systems focused design contexts. Hence, future work is needed to examine other independent cases, as well as non-behavioural design cases, which could foster further extension and generalisation.
Conclusions
Behavioural design has emerged as a distinctive approach to shaping positive, ethical behaviour change. Realising such change often necessitates complex interventions with multiple behaviour change techniques and designed artefacts. Yet, it is unclear how current practices could be adapted to or implemented in such CBD processes. Hence, we set out to answer the RQ: How can complexity be addressed in a behavioural design process?
We adopted an explorative, qualitative, theory building approach, supported by a single in-depth case study. Through this we develop several significant contributions. Specifically, we theorise CBD in systemic terms and differentiate this from typical behavioural design processes in three main ways: (i) the progression of design at multiple abstraction levels requiring different design capabilities, (ii) the presence of both top-down and bottom-up reciprocal interactions across levels, and (iii) the need for mid-level design coordination in order to realise system outcomes and leverage the diverse capabilities within the team. These significantly extend current discussions in behavioural design and point to propositional directions for future research.
Footnotes
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.