Introduction: Defining the Method
In my recent post on concept modelling of work rehabilitation, I presented a series of hierarchical and graph-based visualisations synthesising different theoretical models from the vocational rehabilitation literature. The approach - extracting concepts from literature and policy, structuring them hierarchically or spatially, and rendering them as diagrams or models - represents a method I have been developing since my postgraduate studies at Brunel University around 2009-2010, and refining throughout my professional practice in interaction and service design.
This post is an attempt to articulate that method more formally. The need to do so has arisen from my current work on the ADAPT/Pathway Generator project, where I have found myself producing concept maps to try to make sense of a context that seems to defy sense-making. The project promises "machine learning" and "artificial intelligence" for vocational rehabilitation - but there is no data, no database infrastructure, no clear data pipeline. It operates within an ESF funding framework that demands outcome metrics - but where there is no coherent or structured project management. Different stakeholders appear to hold fundamentally different mental models of what the project is and what it might achieve, and it seems to move forward with its own implicitly or loosely structured momentum.
In attempting to visualise these different models, and in attempting to expose the gaps between them, I am drawing on a method I have practised for over a decade. But I have never fully articulated its theoretical foundations or critically appraised its limitations. This post attempts to do both - to situate my approach within relevant intellectual traditions, contrast it with dominant tools in service design practice, and honestly assess what it can and cannot do.
Intellectual Origins: Three Formative Influences
My interest in hierarchical conceptual modelling was first sparked by encountering Hugh Dubberly's work while studying at Brunel. Dubberly, formerly of Apple and founder of Dubberly Design Office, had been developing an approach to concept mapping that drew explicitly on Joseph Novak and D. Bob Gowin's educational research, documented in their 1984 book Learning How to Learn. I have articulated the epistemological foundations of this work - Gowin's "Vee heuristic" for knowledge construction - in a companion post.
As Dubberly (2010) explains: "A concept map is a picture of our understanding of something. It is a diagram illustrating how sets of concepts are related. Concept maps are made up of webs of terms (nodes) related by verbs (links) to other terms (nodes). The purpose of a concept map is to represent (on a single visual plane) a person's mental model of a concept". What attracted me to his work was not concept mapping as a tool for education but concept mapping as a design method - a way of making visible the conceptual structures that underpin complex systems. His concept maps of innovation, of brand, of Java technology demonstrated that concept mapping and visualisation could reveal the deep structure of domains that otherwise remained tacit and contested, particularly the complex social domains and interactions that service design as a practice claims to address.
Tergan et al. (2006), in their work on digital concept maps as "bridging technologies", make a similar argument: concept maps can function as intermediaries between different knowledge domains, making it possible to "take advantage of the remarkable capabilities of the human visual perception system" to navigate complexity. The concept map becomes not merely a representation but a tool for thinking - a way of externalising cognition so that it can be examined, shared, and revised.
A second formative influence was Jeff Johnson and Austin Henderson's work on conceptual models in interaction design. Their 2011 book Conceptual Models: Core to Good Design articulated something I had intuited but not been able to express: that the gap between how designers intend users to understand a system and how users actually understand it is the source of most usability problems. Johnson and Henderson (2011) define the conceptual model as "a high-level description of an application. It enumerates all concepts in the application that users can encounter, describes how those concepts relate to each other, and explains how those concepts fit into tasks that users perform with the application".
Crucially, they distinguish the conceptual model (what designers intend) from the user's mental model (what users actually develop through interaction). As they note: "Ideally, users' understanding of the application should match what the designers intended; otherwise users will often be baffled by what it is doing". This framing suggested that design failures often stem not from poor implementation but from misalignment between different models - between what different stakeholders believe a system to be and to do. It suggested that making these models explicit through visualisation might enable their comparison and alignment.
The third influence - and the one that most shapes my current practice - was my exposure to Ecological Interface Design (EID) and Cognitive Work Analysis (CWA) during my postgraduate studies. EID is a theoretical approach to interface design developed by Jens Rasmussen and Kim Vicente in the early 1990s, rooted in cognitive systems engineering and the analysis of complex sociotechnical systems (Vicente & Rasmussen, 1992).
Central to EID is the Abstraction Hierarchy - a framework for analysing work domains that organises information across five levels of abstraction. The hierarchy represents the system from its most abstract purposes down to its concrete physical form, connected by means-ends relationships. What a physical function is at one level becomes the means by which a generalised function is achieved at the level above. This framework is visualised below:
As Burns and Hajdukiewicz (2017) explain: "EID has adopted the Abstraction Hierarchy as a fundamental way to analyze the environment, the work domain". The hierarchy "consists of five levels of abstraction, ranging from the most abstract level of purposes to the most concrete level of physical form" (Stanton & Salmon, 2017).
The power of this framework lies in its capacity to represent the same system at multiple levels of description simultaneously. The observation that has informed my practice since is that this framework, developed for technical systems such as nuclear power plants and aviation control rooms, could be adapted for social and organisational systems. If the abstraction hierarchy could reveal the deep structure of complex technical work domains, might it also reveal the structure of complex social systems - including the services, policies, and institutions within which vocational rehabilitation takes place?
Jon Kolko and the Magic of Synthesis
Jon Kolko's work, particularly Exposing the Magic of Design (2011), provided a complementary perspective. Kolko argues that design synthesis - the process of making sense of complex, contradictory data - is the core creative act of design. It is the "magic" that transforms research findings into design insights, that enables designers to see patterns in chaos. What Kolko describes as synthesis, I understand as a form of conceptual modelling. The designer confronts a mass of observations, interview transcripts, policy documents, theoretical frameworks - and must find a way to structure them into a coherent understanding. The concept map, the affinity diagram, the framework - these are tools for performing synthesis, for externalising the cognitive work of making sense.
Kolko's emphasis on mapping as a synthesis method resonates with my own practice. As he argues, maps create "defined links between perceptions providing a framework for looking at a particular system". Through mapping, designers can visualise "the multiple, often intangible, interactions that occur between and within systems".
Multiple Mental Models in Social Systems
These intellectual influences converge on a shared proposition: that different stakeholders in a complex social system hold different mental models of that system - and that these models often operate at different levels of abstraction, employ different vocabularies, and embody different ontological assumptions about what exists and what matters.
Consider my current context: the vocational rehabilitation service ecosystem in Sweden. A caseworker at Samordningsförbundet holds a model of "what rehabilitation is" and "how it works". A data scientist at Stirling University holds a different model - focused perhaps on feature vectors and predictive algorithms. A policy administrator at the Swedish Social Insurance Agency holds yet another model - oriented to eligibility criteria and outcome metrics. The people being rehabilitated hold models grounded in lived experience of illness, hope, fear, and daily struggle.
These models are not simply different perspectives on the same underlying reality; they are different constructions of what the domain contains and how its elements relate. When a project promises "machine learning for vocational rehabilitation", each stakeholder hears something different - because each is projecting the promise onto a different conceptual structure.
This is where concept mapping becomes more than an academic exercise. By making each model explicit - by visualising the concepts, hierarchies, and relationships that different stakeholders assume - it becomes possible to see where models align and where they conflict. It becomes possible to see the gaps: the places where one model assumes something exists that another model knows to be absent.
Making Things Visible: The Design Orthodoxy
This approach sits within a broader design orthodoxy around "making things visible". Design theory consistently assumes that visualisation and materialisation enable productive outcomes. As Wastell (2011) notes, "designers make problems and ideas visible, creating frameworks to make visual sense of complex information". Bailey (2021) argues that "design works through visual and material modes to help realise ideas, to put things in the world and foster dialogue about them". Morelli (2020) states that "making solutions visible before all the information is available is a critical function of designers".
The assumption is that by externalising tacit knowledge into explicit representations, design creates conditions for shared understanding, collaborative sensemaking, and productive change. This assumption underpins several established methodologies. Soft Systems Methodology (Checkland, 1981) employs "rich pictures" - deliberately messy, hand-drawn visualisations that capture the complexity and contestation in human activity systems. As Paul and Yeates (2010) note, rich pictures "offer a free-format approach that allows analysts to document whatever is of interest or significance". Checkland's approach explicitly embraces plurality, recognising that different stakeholders see the same situation differently.
Service Blueprints (Shostack, 1984; Kingman-Brundage) map services as systems of touchpoints, activities, and supporting processes. As Haugen (2013) notes, "the service blueprint is a way to map and visually explain a service" - making visible the backstage processes that support frontstage interactions. This method too assumes that visibility enables understanding: once the system is visualised, designers and stakeholders can see how it works and identify opportunities for improvement.
Product Service Ecology (Forlizzi, 2013) takes a systems approach to design, mapping "actors, artifacts, and relationships that exist within a complex system". Forlizzi argues that "to represent a complex system, designers need to rely on visual thinking and visualization" - creating diagrams that reveal how products, services, and people interact within an ecology. What these methods share - whether working at the level of activity systems, services, or broader ecologies - is the belief that visualisation enables understanding, that making the implicit explicit creates conditions for productive change.
The Tools of Mainstream Service Design: A Critical Appraisal
Against this backdrop, it is worth examining the dominant visualisation tools in contemporary service design practice - and asking whether they are adequate to the complexity they claim to address.
Personas - archetypal descriptions of user types - have been ubiquitous since Alan Cooper popularised them in the 1990s. They are intended to build empathy and focus design decisions around human needs. Yet personas face significant critique. Turner and Turner (2010) identify "the tension between the economy of stereotyping on the one hand and the potential for bias and loss of detail on the other". Wilson and De Paoli (2018) observe that personas are subject to "the tendency towards stereotyping that seems inevitable when a large amount of varied data about people has to be compressed into one representation".
The epistemological problem is that personas present a single model of the user - an anthropomorphic representation that suggests a coherent, knowable subject. This works against the recognition that different stakeholders hold different models, that service contexts contain multiple ontologies in play, and that the "user" is always already plural and contested. As Turner and Turner (2010) note: "It is quite clear that for many designers to create a user representation is, very likely, to create a stereotype". The stereotype may be useful - it focuses attention, creates empathy - but it also flattens complexity and obscures difference.
Customer journey maps visualise the sequence of touchpoints a user encounters when interacting with a service. They have been widely adopted because they make visible the temporal unfolding of service experiences. Yet journey maps embody limiting assumptions. Linear temporality is one: journey maps assume sequential progression - awareness, consideration, decision, use. Real service experiences, particularly in contexts like healthcare or welfare, rarely follow such neat progressions. People loop, regress, drop out, and re-enter. The journey metaphor imposes a linearity that reality does not possess.
A second limitation is single-actor focus. The "journey" is typically one person's journey. Yet services involve multiple actors - clients, professionals, carers, administrators - whose journeys intersect, conflict, and co-constitute each other. A third is what might be called touchpoint atomism: by decomposing experience into discrete touchpoints, journey maps can obscure the systemic conditions that shape what happens at each point. The map shows what happens but rarely reveals why the system produces those experiences. As Mages and Neely (2023) observe, journey maps "provide a framework for authoring a service" but may not adequately capture the felt complexity of temporal experience.
The IDEO-derived design thinking methodology has become perhaps the most widely adopted framework in contemporary practice. It centres "empathy" as the foundation of human-centred design - the first stage in the process, before definition, ideation, prototyping, and testing. Yet empathy as operationalised in design thinking has been critiqued. As the Research Handbook on Design Thinking (Straker & Wrigley, 2023) notes: "Empathy, which gives its name to the first stage in the IDEO DT model, has been subject to some critique". The concern is that empathy can become a performance - a token gesture toward understanding users that does not actually challenge designers' assumptions or redistribute power.
Participatory Design, rooted in the Scandinavian democratic tradition (Ehn, 2008; Bjögvinsson et al., 2012), offers a more politically grounded alternative. As Resnick (2019) notes, "Participatory design is an approach focused on processes and procedures where all stakeholders (e.g., employees, partners, customers, citizens, end users) are actively involved in the design process". The emphasis is not on designers feeling empathy but on users having power. Yet even participatory design, in its contemporary instantiations, often relies on the same representational tools - personas, journeys, empathy maps - that mainstream design thinking employs. The political commitment to democratisation does not automatically produce epistemic tools adequate to systemic complexity.
I would argue - as have many others - that personas, empathy maps, and journey maps have acquired a totemic quality in service design practice. They have become ritual objects whose production signals "user-centredness" regardless of whether they actually inform design decisions or enable systemic understanding. The persona pinned to the wall, the journey map spanning the workshop table - these artefacts perform legitimacy as much as they produce knowledge. They say: "We are human-centred designers. Look, we have considered the user".
But the gap between what these tools claim to do (represent users, enable empathy, guide design) and what they can do (compress complexity into manageable stereotypes) becomes problematic in contexts where the stakes are high and the simplifications consequential. This gap is not a technical limitation but an epistemological one - a matter of what kinds of knowledge these tools are equipped to represent and what kinds of complexity they necessarily obscure.
Systemic Design and the Incomplete Turn
Service design has increasingly embraced systems thinking, recognising that services exist within ecosystems of actors, institutions, and interdependencies. Jones (2021) distinguishes systemic design from service or experience design "in terms of scale, social complexity and integration - it is concerned with higher order problems that shape the conditions within which services operate". Jones and Van Ael (2022) note that "systemic design adapts the human-centred design approach to complex, multi-stakeholder service systems".
This systems turn is welcome. Yet much systems-informed service design retains the representational tools of its user-centred origins - adding systems maps and ecosystem diagrams alongside, rather than instead of, personas and journeys. The result can be conceptual incoherence: a persona (which assumes a stable, knowable individual subject) sitting alongside a systems map (which reveals how that subject is constituted by and distributed across networks of relations).
What is needed, I would argue, is representational tools adequate to the systemic ontology - tools that can hold multiple levels of abstraction, multiple semantic vocabularies, and multiple stakeholder models simultaneously. This is the aim of the approach I describe below.
Concept Modelling as Method: A Definition
The concept modelling approach I have been developing attempts to address this need. It draws on the abstraction hierarchy's insight that complex domains can be represented across levels from purpose to form; concept mapping's technique of making relationships between concepts explicit and visible; and conceptual model theory's recognition that different stakeholders hold different mental models of the same system.
The method involves several distinct phases. Domain scoping comes first: identifying the domain to be modelled and the stakeholder perspectives to be included. This involves decisions about boundaries - what counts as part of the system and what falls outside it - and about whose voices and frameworks will be represented in the model.
Concept extraction comes next: reviewing literature, policy documents, professional frameworks, and empirical data to extract the concepts employed within the domain. This is not a mechanical process but an interpretive one; the concepts that emerge from a policy document depend on how one reads it, what one attends to, what one takes to be central or peripheral. This phase is necessarily partial and contested.
Hierarchical structuring follows: organising concepts into hierarchical relationships - from abstract purposes through general functions to specific capabilities and concrete forms. This phase draws directly on the abstraction hierarchy framework described earlier, using the principle of means-ends relationships to structure concepts vertically.
Cross-model synthesis comes next: producing multiple models representing different stakeholder vocabularies or theoretical frameworks, then identifying overlaps, gaps, and contradictions. This is where the approach differs most sharply from traditional concept mapping; rather than producing a single unified map, the method produces multiple maps and uses comparison to reveal what different frameworks assume, what they highlight, and what they obscure.
Finally, visual rendering produces interactive visualisations that enable navigation across levels and comparison between models. The visualisation is not merely a representation of the thinking but an active tool for further thinking - a way of making visible the relationships and gaps that might otherwise remain implicit.
In the context of the current ADAPT project, this has meant modelling vocational rehabilitation according to different theoretical frameworks - Swedish literature, international literature, ICF taxonomy. It has meant modelling the BIP assessment instrument, one of the most readily available instruments for gathering data that might help populate machine learning models, and examining its conceptual underpinnings in detail. It has meant modelling what a "machine learning system for rehabilitation" would actually require - data sources, pipelines, training processes, infrastructure. And it has meant beginning to expose the gaps between what the project assumes exists and what actually exists, what the project aspires to create and what is genuinely possible within its constraints and timeframe.
As Miller and Rusnock (2024) note in their recent work on integrating human and artificial intelligence through systems design: "A Concept Map is a diagram that shows the relationships between concepts, usually items, ideas, or information. For our context, the concept map displays the key activities, actors, and outcomes that fulfil a particular high-level goal. The connections between these concepts are linking phrases that describe the relationship between the concepts. Overall, the Concept Map(s) identifies: (1) The cognitive work required to achieve the key outcomes, (2) the outcomes that the system needs to achieve, and (3) the interactions and overlaps that occur between concept map entities".
This captures precisely what I have been attempting: to identify the cognitive work required to achieve stated outcomes, and to make visible where that cognitive work has not been done - where outcomes are promised without the means to achieve them.
Critical Appraisal: The Limits of Making Visible
It is perhaps worth reflecting on the limitations and risks of this approach, particularly in contexts where the stakes are high and stakeholder interests are divergent.
The first risk is what might be called false authority. Hierarchical concept models can appear more authoritative than they are. The visual formality - nodes and edges, layers and labels - suggests rigour and completeness. Yet the models are always partial, always reflecting the modeller's reading of sources, always embodying interpretive choices that could have been made differently. There is a risk that the model's appearance of systematic comprehensiveness obscures its status as one possible interpretation among many. The model's visual authority can mask the contingency of the choices that produced it.
A second risk is ossification. By fixing concepts into hierarchical structures, the approach may inadvertently ossify what is actually fluid and contested. Social systems are not static architectures but ongoing accomplishments - continuously reproduced, negotiated, and transformed through practice. A concept model captures a snapshot, but may be mistaken for an enduring structure. The model's fixed form can suggest a stability and coherence that the lived reality of the system does not possess.
Most significantly, I must confront the question of whether making things visible through concept modelling actually produces the change it is supposed to enable. In my current project context, the concept maps I have produced have exposed significant gaps: promises of "machine learning" without data; claims of "federated learning" without infrastructure; outcome commitments without project management capability. The visualisations have made these absences undeniable - or so I thought.
Yet the response has not been the productive dialogue and course-correction that design theory predicts. Instead, there has been absorption. The maps are acknowledged, discussed, filed. The project continues as before. The gap between imaginary and reality persists. This suggests a limitation in the design orthodoxy: the assumption that visibility automatically enables change. What if, in certain contexts, visibility can be absorbed, deflected, or rejected? What if making things visible exposes contradictions that stakeholders have a vested interest in not seeing?
These are questions I do not yet have answers to. But I suspect that the limits of concept modelling as a design method are not technical but political - not about the quality of the visualisation but about the conditions under which visualisation can produce recognition rather than denial. Making something visible does not guarantee that it will be seen, and seeing something does not guarantee that it will change the behaviour it implies.
Finally, I should acknowledge that this approach remains marginal in service design practice. The field, at least in my experience, continues to favour personas, empathy maps, and journey maps - tools that are quicker to produce, easier to explain, and more immediately engaging to non-specialist stakeholders. Concept modelling demands more time, more domain expertise, and more tolerance for abstraction. Whether these demands are justified by superior outcomes remains to be demonstrated. What I can say is that in my current context - where the need to unify disparate and apparently irrational perspectives has become acute - the approach has at least enabled me to articulate what is incoherent, even if it has not yet enabled its correction.
References
Bailey, J.A. (2021). Governmentality and power in 'design for government' in the UK.
Bjögvinsson, E., Ehn, P. & Hillgren, P.A. (2012). Design Things and Design Thinking: Contemporary Participatory Design Challenges. Design Issues, 28(3), 101-116.
Burns, C.M. & Hajdukiewicz, J.R. (2017). Ecological Interface Design. CRC Press.
Checkland, P.B. (1981). Systems Thinking, Systems Practice. Wiley.
Dubberly, H. (2010). Creating Concept Maps. Dubberly Design Office. Available at: https://www.dubberly.com/concept-maps/creating-concept-maps.html
Ehn, P. (2008). Participation in Design Things. Proceedings of the Participatory Design Conference.
Forlizzi, J. (2013). The Product Service Ecology: Using a Systems Approach in Design. Relating Systems Thinking and Design Symposium.
Haugen, M. (2013). Service Blueprints. Touchpoint, 5(2).
Johnson, J. & Henderson, A. (2011). Conceptual Models: Core to Good Design. Morgan & Claypool.
Jones, P.H. (2021). Systemic Design: Design For Complex, Social, And Sociotechnical Systems. In Handbook of Systems Sciences.
Jones, P.H. & Van Ael, K. (2022). Design Journeys Through Complex Systems. BIS Publishers.
Kolko, J. (2011). Exposing the Magic of Design: A Practitioner's Guide to the Methods and Theory of Synthesis. Oxford University Press.
Mages, M.A. & Neely, S. (2023). Mapping Temporal Experience: accounting for felt time in service design. Proceedings of the ServDes Conference.
Miller, M. & Rusnock, C. (2024). Integrating Artificial and Human Intelligence through Agent Oriented Systems Design.
Morelli, N. (2020). Service Design Capabilities.
Novak, J.D. & Gowin, D.B. (1984). Learning How to Learn. Cambridge University Press.
Paul, D. & Yeates, D. (2010). Business Analysis. BCS.
Resnick, E. (Ed.) (2019). The Social Design Reader. Bloomsbury.
Shostack, G.L. (1984). Designing Services That Deliver. Harvard Business Review.
Stanton, N.A. & Salmon, P.M. (2017). Human Factors Methods: A Practical Guide for Engineering and Design. CRC Press.
Straker, K. & Wrigley, C. (Eds.) (2023). Research Handbook on Design Thinking. Edward Elgar.
Tergan, S.O., Keller, T. & Burkhard, R.A. (2006). Integrating Knowledge and Information: Digital Concept Maps as a Bridging Technology. In Information Visualization, pp. 383-390.
Turner, P. & Turner, S. (2010). Is Stereotyping Inevitable When Designing With Personas? Design Studies, 32(1), 30-44.
Vicente, K.J. & Rasmussen, J. (1992). Ecological Interface Design: Theoretical Foundations. IEEE Transactions on Systems, Man, and Cybernetics, 22(4), 589-606.
Wastell, D. (2011). Managers as Designers in the Public Services: beyond technomagic.
Wilson, A. & De Paoli, S. (2018). Creating personas for political and social consciousness in Human-Computer Interaction design. Persona Studies, 4(2), 25-46.