Category Archives: Methodology

Dealing with AI

Dealing with AI

The UK Government publication of its AI Opportunities Action Plan (Clifford & Department for Science Innovation and Technology, 2025) sets out an agenda primarily focussed on the growth opportunities enabled by AI.

Many of the recommendations require translation from the problematic context to implementations using specific AI technologies, such as generative pre-trained transformers (GPT), and are therefore all examples of where problematisation is inevitable. As Callon (1980) observes 

“Problematisation culminates in configurations characterised by their relative singularity. There is not one single way of defining problems, identifying and organising what is certain, repressing what cannot be analysed.”

How then are these decisions to be enacted, the processes of deciding, in the operationalisation of this action plan? This question is difficult to answer as the action plan is just a set of recommendations, a “roadmap for government to capture the opportunities of AI to enhance growth and productivity and create tangible benefits for UK citizens.” 

Roadmaps, like actions and solutions, present static nominalisations of what should be a dynamic process. The actual intervening in problematic areas such as health care, education and provision of services etc. will emerge from the normal processes of deliberating over any technology, not just AI. Restating Callon’s perspective, there will be an “abundance of problematisations” resulting from these deliberating processes leading to options where specific choices will need to be made. It is in the choosing between options, in the different ways in which interventions with AI technologies can be problematised, that the inevitable conflicts between different stakeholder needs, differing policy objectives, will be made clear. It is here that problem structuring as a deliberative process needs to operate. 

“The usefulness of Callon’s work on problematisation is that it draws attention to the choices faced by an operational researcher when investigating a problem, and that there is not a single right answer to what is problematised. Whilst this supports the claims of Churchman, Ackoff and Checkland that these choices exist and that operational researchers are making conscious decisions about what to work on (and therefore what to ignore), it does not provide an answer to the problem of OR becoming practitioner-free, i.e. the problem of dealing with situational logics. The answer is found in the emergence of problem structuring methods (PSMs).” (Yearworth, 2025, pp. 13-14)

I was writing for Operational Researchers in my book, but the point is valid for any analyst or decision maker and their profession. Choices exist and conscious decisions need to be made about what to work on and what to ignore and these choices need to be made visible1 and debatable through formal processes – such as through the use of PSMs.

The situational logic that sits at the heart of the AI opportunities action plan is that choosing AI leads to (economic) growth. However, if we abrogate on our moral responsibility to make ethical choices and fall back on the simple rule-following of situational logics then we may as well hand the deliberation and implementation of an AI action plan to an AI itself2 and wash our hands of the consequences. 

I recognise that the AI opportunities action plan makes specific reference to “[g]lobal leadership on AI safety and governance via the AI Safety Institute, and a proportionate, flexible regulatory approach” and reflects the fact that choices need to be made by overloading the use of “proportionate.” Deliberating and deciding over a flexible regulatory approach will require hard work, these will be (and should be) difficult choices. Given the scale of the challenges and opportunities of AI, apportioning sole agency for this deliberating and deciding to the AI Safety Institute (AISI) just narrows the location and scope of debate around problematisations to, in effect, informing decisions about the boundaries of regulation that are broadly pro-innovation. Deflecting focus away from this concentration of decision making by talking about assurance tools in an AI assurance ecosystem just sounds like marketing i.e., our attention on the situational logic in operation here is being misdirected by the AISI.

For almost all the recommendations in the action plan, problematising should be a diffuse activity across a very broad range of actors, problem contexts, stakeholders, and technologies – putting choice into the hands of people best able to decide for themselves the scope of adoption of AI. By all means give organisations, and individuals, the processes that would enable them to make informed decisions, but these are not imposed ‘flexible regulation’ and ‘assurance tools’ that ultimately disempower.      

  • Callon, M. (1980). Struggles and Negotiations to Define What is Problematic and What is Not. In K. D. Knorr, R. Krohn, & R. Whitley (Eds.), The Social Process of Scientific Investigation (pp. 197-219). Springer Netherlands: Dordrecht. https://doi.org/10.1007/978-94-009-9109-5_8

  1. In effect the models/maps of the structured problem. ↩︎
  2. Checkland introduced the idea of the trap of situational logics in OR practice. However it was Rosenhead, writing in Rational Analysis for a Problematic World, who used the analogy of the sausage machine, which is more than apt here. ↩︎

Not getting lost in process

Not getting lost in process

Debates about political process and endless delays in making decisions threaten to weaken trust in our democratic institutions especially with regard to pressing matters like the provision of adult social care. Politics has become a “prisoner of process” (Bagehot, 2025). 

Bagehot makes a good point and cites Blair in support of the observation that process, rather than being the means to the end, has become an end in itself. Bagehot draws on Stafford Beer’s (not attributed) POSIWID heuristic – the purpose of a system is what it does – to suggest that the system’s purpose has become an endless cycle of debate without action, although, in passing, observing that deliberation is necessary to “ensure that decisions are simply not made” (my emphasis). This is all good, but I think there are a number of conceptual errors that unhelpfully muddy the argument.

Starting with Beer’s POSIWID, it is a simple observation that system is being interpreted here in a narrow sense. We would hope that any system of governance has feedback mechanisms in it. Rather than a simple linear sequence of steps we would expect something like deliberation → action → observation (of effects of actions) → comparison → deliberation → …, where the comparison step derives an error signal based on the difference between what was intended and what happened. This system should operate in a continual cycle of feedback – it is both unlikely that our actions achieve the desired effect and the world keeps changing anyway. While we might conclude that the evident purpose of the system is to endlessly deliberate i.e., deliberation → deliberation → …, we could go a bit further and observe where the system is broken – the action element is missing and therefore the feedback loop is not operating. I think we would both agree that the system needs to be repaired. 

POSIWID is useful and the elicitation of feedback loops, at any desirable level of detail, provides a powerful analytical tool; but I believe there is another way of looking at this problem and the use of a process approach offers some benefits, rather than being consigned in the narrow sense to a trap of deliberation. The key can be found in the way in which we use language in our analyses. I have previously railed against the use of language like ‘solution’ and ‘fix’ in the context of complex problems, but in the analysis of the feedback loop above I rather consciously used ‘deliberation’, ‘action’ and ‘observation’ to emphasise the linearisation of what should be a system and that this arises from the nominalization of elements that should be thought of as verbs.

Getting stuck in a process of deliberation, or an endless sequence of deliberations, is likely when all the actors, including analysts and commentators (expert and otherwise), are constrained by their nominalizations. A better conceptualising of process thinking is to think of governing as a process and that for it to perform it must consist of further processes such as deliberating, acting (or effecting change, or intervening) and observing (or measuring). These processes are all necessary for governing but none are sufficient, by themselves, for properly enacting the process of governing. Note the use of the gerund form of the verb to convey a sense of continual ongoingness of the process. We can decompose this schema (or model) to any level of detail that is required using conditions of necessity and sufficiency as a test on whether a process is required in the model. 

Coming back to Bagehot’s analysis, we can clearly agree that the process of acting is not working well, but it cannot be reduced to a simple intervention that is yet to happen and that will somehow ‘fix’ the problem. The process of deliberating is obviously not working well either, it is clearly not sufficient by itself to enable the process of governing and our measure of its performance should of necessity include its commissioning of useful planning to enable acting. Rather than being prisoners of process, we would be better served by realising that processes are all there are, both in the world and in our ways of intervening in the world. To do this, amongst other things, requires a change in our language, away from nominalizations, especially ones like ‘action’ and ‘solution’, and recognise that acting or intervening is a continuing and ongoing process and may be enacted at any level of scale (socially, temporally, spatially,…).

In the case of adult social care there is clearly a whole lot of process detail that is completely missing between deliberating and intervening and nobody seems to be talking about it. We are left with unedifying analyses and useless solutionist traps. 

Modelling HPM as a PSM, using HPM

Modelling HPM as a PSM, using HPM

I conclude my book on Problem Structuring with some comments on a processual turn in Operational Research (Yearworth, 2025) and specifically comment

“In modelling the world processually we can also model our interventions within the same model … Or put more simply, problematic situations are processes as are the means of intervention.” (p. 270). 

Modelling our interventions specifically requires the possibility of representing our use of a problem structuring method as a model. Checkland and Poulter described the process of using Soft Systems Methodology (SSM) in the activity system ‘language’ of SSM itself (Checkland & Poulter, 2006, p. 194; Yearworth, 2025, p. 78). Checkland and Scholes went further and modelled the system to use SSM in the same purposeful activity system language i.e., the ongoing reflective practice of using SSM in client engagements (Checkland & Scholes, 1990, p. 294).  

The same approach has been used for describing the use of Hierarchical Process Models (HPM) for problem structuring. This was first manifest in the STEEP Project as means of self-evaluation of how well the methodology was performing, making use of the Italian Flag as a means of capturing judgement of process performance (Yearworth et al., 2015, p. 9). In the Healthy Resilient Cities project (Yearworth, 2015), we started to model the process of using the PSM within the model of the problematic situation itself (Yearworth, 2025, p. 173) i.e., the process <Improving the resilience of healthcare provisioning in Bristol…> contained within it the process <Using problem structuring>. Further work on exploiting the Italian Flag for capturing judgements of process performance in the use of a PSM was explored in depth by Lowe, Espinosa and Yearworth (2020).

The development of HPM as a PSM is described fully in Chapters 9 and 10 of my book. However, the work of modelling HPM as a PSM using HPM itself that was started in the STEEP project is still ongoing. The following shows its current incarnation in Strategyfinder.

HPM of using HPM as a PSM

Note that methodological learning, an essential element of using a PSM, is reflected in the model at Process #26 <Evaluating the engagement and improving our understanding …> and the processes it contains. The model also references other models that could be incorporated as enhancements, for example using the framework for improving facilitation developed by Ackermann (1996, p. 95), which has been interpreted in my book as another process model (Yearworth, 2025, p. 108). This illustrates the property that all HPM are composable according to their necessity and sufficiency for the success of the process that acts as the anchor for incorporation.

Ackermann, F. (1996). Participants’ perceptions on the role of facilitators using group decision support systems. Group Decision and Negotiation, 5(1), 93-112. https://doi.org/10.1007/BF02404178

Checkland, P., & Poulter, J. (2006). Learning for action : a short definitive account of soft systems methodology, and its use for practitioner, teachers and students. John Wiley & Sons: Chichester. 

Checkland, P., & Scholes, J. (1990). Soft systems methodology in action. John Wiley & Sons: Chichester. 

Lowe, D., Espinosa, A., & Yearworth, M. (2020). Constitutive rules for guiding the use of the viable system model: Reflections on practice. European Journal of Operational Research, 287(3), 1014-1035. https://doi.org/10.1016/j.ejor.2020.05.030

Yearworth, M. (2015). Healthy Resilient Cities: Building a Business Case for Adaption (NERC NE/N007638/1)[Grant]. Bristol. http://gotw.nerc.ac.uk/list_full.asp?pcode=NE%2FN007638%2F1

Yearworth, M. (2025). Problem Structuring: Methodology in Practice (1st ed.). John Wiley & Sons, Inc.: Hoboken. https://doi.org/10.1002/9781119744856

Yearworth, M., Schien, D., Burger, K., Shabajee, P., & Freeman, R. (2015). STEEP Project Deliverable D2.1(R2) – Energy Master Plan Process Modelling. STEEP PROJECT (314277) – Systems Thinking for Comprehensive City Efficient Energy Planning, pp78. Retrieved 26th January 2023, from https://www.grounded.systems/wp-content/uploads/2023/01/01_STEEP_D2.1_Energy_Master_Plan_process_model_update_M24_DEF_sent.pdf

Augmented Qualitative Analysis (AQA) and Large Language Models (LLMs)

Augmented Qualitative Analysis (AQA) and Large Language Models (LLMs)

Conducting experiments in the automated labelling of topics generated using the Augmented Qualitative Analysis (AQA) process outlined in an earlier post has resulted in some observations that have some bearing on the use of Large Language Models (LLMs) in Soft OR/PSM practice.

The starting point for AQA was the partial automation of some of the elements of qualitative analysis, which has resulted in the use of probabilistic topic models to code text in a corpus at the paragraph level and to produce maps of the interrelationship of concepts (qua topics). These maps of interrelationships can only be put forward for consideration as potentially causal links after the topics have been labelled. A map of topic X being linked to topic Y n times is only of statistical interest. We need meaning to be attached to the topics – ideally a process of consideration by a group that ‘owns’ the raw data – before we can produce a putative causal loop diagram (CLD).

To ground this technique in traditions of qualitative analysis would require the labelling of topics to proceed through an inductive process of inspecting the term lists and inspecting the text coded by the topics to build-up an understanding of what the topic means to the stakeholders (e.g., see Croidieu and Kim (2017)). This is a back-and-forth process that continues until all the topics have been labelled. The fact that the map can be updated with these topic labels in parallel provides an additional perspective on the understanding of the meaning of the topics. 

With the advent of LLMs it is possible to feed the term lists and the example text coded by a topic – and even the map of interrelationships – into a tool like ChatGPT with the purpose of generating topic labels. However, experiments in doing this have produced disappointing results, despite extensive efforts in refining prompts. From the perspective of a qualitative researcher, the coding seems to be too much in the text, too in-vivo. Despite attempts to get the LLM to draw on the breadth of its training data there seemed little evidence of the sort of theorising from the data that is a key feature of qualitative analysis (Hannigan et al., 2019).

This is clearly a new area and other researchers have conducted experiments on precisely this point of prompt engineering e.g., see Barua, Widmer and Hitzler (2024). However, there is still the sense that a LLM is operating as nothing more than a ‘stochastic parrot’ (Bender et al., 2021). Further, coupling the outputs from a probabilistic topic model to a LLM are unlikely to generate the sort of management insight that is discussed by Hannigan et al. (2019); although the putative causal maps are likely to make sense to the participants in a group, and are statistically justified. Ultimately, the use of LLMs in a process of problem structuring is only ever going to be limited. Problematising is a human activity, it requires a felt-sense of a situation being problematic for an intent to intervene to emerge. Asking a LLM to feel something is a wayward expectation.

The recommendation here, for any group working with a large and potentially growing corpus of documents and in need of a technique that supports rapid problematisation, is to work with two Group Support Systems (GSS). The first presents an interactive means of exploring the probabilistic topic model (e.g., using pyLDAvis the topic model for the 2012 Olympics data set discussed in a previous post can be explored here) combined with a means of investigating the text as coded by the topic model i.e., selecting text that is coded by a specific topic and inductively generating a topic label that has meaning to the group. In effect, replicating some of the features of a Computer Aided Qualitative Data Analysis Software (CAQDAS). The fully labelled model can then be taken into a strategy-making workshop supported by the second GSS, in this case Strategyfinder. 

The prospects of these two GSS merging into a single Problem Structuring Platform is the subject of my upcoming talk at OR66, see my previous post on AQA.

Barua, A., Widmer, C., & Hitzler, P. (2024). Concept Induction using LLMs: a user experiment for assessment. https://doi.org/10.48550/arXiv.2404.11875

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? FAccT 2021 – Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, https://doi.org/10.1145/3442188.3445922

Croidieu, G., & Kim, P. H. (2017). Labor of Love: Amateurs and Lay-expertise Legitimation in the Early U.S. Radio Field. Administrative Science Quarterly, 63(1), 1-42. https://doi.org/10.1177/0001839216686531

Hannigan, T. R., Haan, R. F. J., Vakili, K., Tchalian, H., Glaser, V. L., Wang, M. S., Kaplan, S., & Jennings, P. D. (2019). Topic modeling in management research: Rendering new theory from textual data. Academy of Management Annals, 13(2), 586-632. https://doi.org/10.5465/annals.2017.0099

Augmented Qualitative Analysis (AQA)

Augmented Qualitative Analysis (AQA)

I first started working on the semi-automated construction of Causal Loop Diagrams (CLDs) as part of a process of qualitative analysis for my MBA project. This work was developed into a full paper and published in the European Journal of Operational Research in 2013.

Yearworth, M., & White, L. (2013). The Uses of Qualitative Data in Multimethodology: Developing Causal Loop Diagrams During the Coding Process. European Journal of Operational Research, 231(1), 151-161. https://doi.org/10.1016/j.ejor.2013.05.002 [post-print version]

Judging by the citations, the methodology described in this paper has been used to develop CLDs across a wide range of application areas.

At the time of writing, the methodology was supported by the use of conventional Computer Aided Qualitative Data Analysis Software (CAQDAS) and, specifically, the use of matrix queries to compute the number of times pairs of concepts (codes) are related by the fact that they co-code paragraphs of text in the sources. The resulting adjacency matrix could be interpreted as a preliminary CLD and an input to further analysis.

Since then, the emergence of probabilistic topic modelling based on the Latent Dirichlet allocation (LDA) means that it is now possible to automate the coding process for very large collections of documents (hundreds to thousands), comprising millions to hundreds of millions of words. The co-coding technique described in Yearworth and White (2013) can be similarly applied by feeding the corpus into the topic model and counting and thresholding the resultant classification of paragraphs by topic(s). This process still requires an inductive1 bridge – in deciding i) a meaningful number of topics (k) to be used for the topic model, ii) the labelling of the topics based on an exploration of the term lists and a re-reading of the text coded by each topic, and iii) a decision about the meaning of the links – can they be interpreted as causal? In practice the last two processes are bound together and interactive once a topic model of particular size has been chosen.

An example graph is shown below based on a corpus of documents assembled from archival material about the 2012 London Olympic Games. The corpus is relatively small for a machine learning technique, 170 documents and just over a million words. However, this is of the sort of volume of data that is starting to get beyond the abilities of a single qualitative researcher to analyse. I’ve called this territory a hinterland for qualitative analysis and hence the motivation for an augmented approach. The graph presented here has been coloured according to betweenness centrality.

The graph was automatically translated into the JSON format supported by the Strategyfinder platform and is shown as a potential causal map below. Once imported into Strategyfinder it is then a further process to discuss the meaning of the relationship between statements i.e., the nature of the links, whether they represent a causal relationship, and their directionality. An export filter to create MDL files suitable for import into Vensim also exists.


I discuss this technique further in Chapters 14 and 15 of my forthcoming book Problem Structuring : Methodology in Practice in the context of a deployment of the technique via a problem structuring platform as a type of Group Support System (GSS). This is also the subject of my upcoming talk at OR66.

Annual OR Conference OR66
Details of the my talk at OR66
  1. I’ve labelled this an inductive step because at this stage in a process of problem structuring this is what it feels like you’re doing. However, situated inside a wider problem structuring process loop that includes modelling and taking action then it could be considered as abductive reasoning. ↩︎

Problem Structuring : Methodology in Practice

Problem Structuring : Methodology in Practice

My new book is now available!

Current perspectives on approaches to problem structuring in operational research and engineering and prospects for problem structuring methods applicable to a wide range of practice.

Despite the myriad successes of Operational Research (OR) in government and industry, critique of its continued relevance to complex, wicked problems led to the emergence and evolution of Soft OR as a more humanist orientation of the discipline centred on a methodological framing of techniques known as Problem Structuring Methods (PSMs). These have enabled OR practitioners to broaden the scope of OR to address complex problem contexts that require transforming, planning and strategising interventions for their clients. The original core PSMs of Soft Systems Methodology (SSM), Strategic Options Development and Analysis (SODA) and the Strategic Choice Approach (SCA) are presented using a new analytical framework based on constitutive rules, epistemologies, and affordances of the modelling approach. Practical considerations in PSM based interventions are discussed emphasising trust-building, stakeholder identification, facilitation and ethical practice. A wide range of PSM applications are surveyed demonstrating clear intersections with communities of practice grounded in the applied social sciences. The development of a new PSM based on Hierarchical Process Modelling (HPM) of purpose arising from a processual turn in engineering practice offers additional insights for the practice of Soft OR. New developments in PSM practice built on use of Group Support Systems (GSS) and exploiting developments in machine learning are presented. Prospects for bringing the Soft OR project back into better alignment with mainstream OR are discussed in the context of new education programs and a possible processual turn in OR.

Problem Structuring: Methodology in Practice contains four linked sections that cover:

  1. Problem formulation when dealing with wicked problems, justification for a methodological approach, the emergence of soft OR, the relevance of pragmatic philosophy to OR practice.  
  2. Traces debates and issues in OR leading to the emergence of soft OR, comparative analysis of PSMs leading to a generic framework for soft OR practice, addressing practical considerations in delivering PSM interventions.
  3. Charts the emergence of a problem structuring sensibility in engineering practice, introduces a new PSM based on hierarchical process modelling (HPM) supported by teaching and case studies, makes the case for a processual turn in engineering practice supported by HPM with relevance to OR practice.
  4. Evaluation of PSM interventions, survey of applications, use of group support systems, new developments supported by machine learning, re-contextualising soft OR practice.

Problem Structuring: Methodology in Practice is a thought-provoking and highly valuable resource relevant to all “students of problems.” It is suitable for any UK Level 7 (or equivalent) programme in OR, engineering, or applied social science where a reflective, methodological approach to dealing with wicked problems is an essential requirement for practice.

Note that some institutions may have access via their usual eBook provider (e.g. ProQuest, Perlego), if not by default on the subscription then by a separate order for this specific title.

Wicked problems and category mistakes

Wicked problems and category mistakes

This is a brief introduction to the notion of a wicked problem. It is based on the highly-cited paper by Rittel and Webber (1973). The following characterise wicked problems:

  1. There is no definitive formulation. In a sense, formulating a wicked problem is the problem
  2. There are no stopping rules. The process of intervening is also the same as understanding the nature of the problem – the intervention is “good enough” or the best that can be achieved within other limitations (e.g. of time, budget…)
  3. Interventions are not right or wrong, they can only be viewed as making things better or worse for certain interests i.e. the intervention has made things both better and worse depending on who you ask
  4. There is no immediate or ultimate test of an intervention. Interventions will generate “waves of consequences” over a period of time
  5. Interventions are “one-shot operations”, experiments are difficult to conduct, every intervention counts significantly, they are essentially unique in nature
  6. No enumerable, exhaustively describable, set of possible interventions
  7. Every wicked problem is essentially unique. “Essentially” implies that aspects may be common, but to think in terms of categories or “classes” of wicked problems with common “solutions” is misleading
  8. Wicked problems can be considered as symptoms of other problems i.e. there is inherent systemicity in the world
  9. Can be contested at the level of explanation, there is likely to be conflicting evidence or data

The corollary of this definition is that certain statements about problems are likely to be rendered false or meaningless if it can be shown that the problem is actually wicked, in effect the statement is demonstrating that a category mistake is being made. The following is not an exhaustive list:

  1. ‘Solving’ or ‘curing’ a wicked problem is a contradiction; there are no ‘solutions’, ‘cures’…
  2. Words that suggest an objective point of view used in the context of the problem at the very least need to be debated e.g. words like optimal, best, right, smart, correct, … all suggest the question – for whom? Alternatively, no decision taken should ever be considered wrong.
  3. Any statement of measurable quantity that supports an argument for the problem getting better or worse without acknowledging the dynamic complexity that systemicity implies i.e. “…worse then better…” is a more believable statement given dynamic complexity
  4. Statements that appear to deny the systemic nature of the problem e.g. ignoring requisite variety
  5. Containing irrefutable assertions of fact e.g. “…this proves conclusively that…”
  6. Use of binary choices, any mention of “silver bullets”
  7. Misrepresenting or ignoring plurality e.g. “The public…”
  8. Emphasis on producing plans rather planning as a process

If any of these corollaries are contested e.g. if someone claims to have a solution to a wicked problem, then they are likely to be making a claim about only an aspect of the problem, or only from a certain viewpoint; or their formulation is not that of a wicked problem i.e. they are talking about something ‘tame’. Statements that contain phrases like “…optimal solution…” or “…this proves conclusively that if we do this we will have the best outcome…” in the context of a wicked problem definitely signal a likely category mistake.

Category mistakes are a warning sign – be sceptical of claims being made. They suggest either misunderstanding or partiality.

It’s worth reading the Rittel and Webber paper. Despite its age, it still does an exceptionally good job of reminding us of the characteristics of wicked problems that’s just as relevant today.

The first steps towards a coherent approach to problem formulation can be found in Rosenhead’s (1996) introduction to Problem Structuring Methods.

Rittel, H. W. J., & Webber, M. M. (1973). Dilemmas in a general theory of planning. Policy Sciences, 4(2), 155-169. doi:10.1007/BF01405730
Rosenhead, J. (1996). What’s the problem? An introduction to problem structuring methods. Interfaces, 26(6), 117-131. doi:10.1287/inte.26.6.117

A short guide to System Dynamics

A short guide to System Dynamics

This guide was produced to help explain System Dynamics modelling to a group of interested stakeholders for a modelling workshop. However, if you have the time I would really recommend reading John Sterman’s textbook for a definitive account:

  • Sterman, J.D. (2000). Business dynamics : systems thinking and modeling for a complex world. Boston: Irwin McGraw-Hill.

Otherwise this short paper summarises the key points

  • Sterman, J.D. (2001). System dynamics modeling: Tools for learning in a complex world. California Management Review, 43(4), pp. 8-25. doi: 10.2307/41166098

And if neither are available, or time is really short, then try this SD-Introduction-MY-20200527.pdf from me.