5.7. Modelling for Learning and Soft Systems
Modelling is not a cookbook procedure – it is fundamentally creative. At the same time, it is a disciplined, scientific and rigorous process that involves observing dynamic phenomena in the real world, surfacing and testing assumptions, gathering data and revising the model to improve understanding. In the absence of strategic modelling, people run business and society, with varied degrees of success, relying on judgement, experience and gut feel. The outer layer of Figure 5.29 shows this normal trial-and-error approach. Based on their mental models of organisations and industries, people devise strategy, structure and decision rules. They take strategic decisions (requiring organisational experiments) whose full implications are never clear at the outset. Then they observe what happens in the real world. The learning cycle is complete when people adjust their mental models on the basis of outcome feedback, in other words by comparing what was achieved with what was intended. The model building process described in this chapter complements the normal but fallible learning cycle. Modelling helps people to share, clarify and improve their mental models. It also enables them to test and refine strategic decisions and organisational experiments through simulation (before trying them out in the real world). The overall process we call 'Modelling for Learning' (Morecroft & Sterman, 1994).
5.7.1. A Second Pause for Reflection: System Dynamics and Soft Systems
Mention of the term 'learning' brings me to a philosophical point about models and modelling of social systems. To what extent should models represent something tangible 'out there' in the real world as opposed to perceptions in the minds of those who must take action in the real world? The distinction may seem of academic interest only, but it is of practical importance to all modellers – important enough that a group of academics and practitioners, funded by the UK Engineering and Physical Sciences Research Council (EPSRC), was convened to discuss the whole topic of hard and soft models (Pidd, 2004).[] The group included representatives from Shell International, BT Exact Technologies, the UK Defence Science and Technology Laboratory (Dstl), the UK Inland Revenue and several academic institutions.[] A useful picture used to stimulate the group's discussion is shown in Figure 5.30 (Checkland, 2004). Here, a distinction is drawn between two approaches to business and social modelling. At the top is the approach of the 'hard' modeller, who observes systems in the real world that can be engineered and made to work better. Below is the approach of the 'soft' modeller, who observes complexity and confusion in the real world, but can organise exploration of it as a learning system. There is a big step from the hard outlook to the soft outlook.
[] The group called itself the 'Interdisciplinary Network on Complementarity in Systems Modelling' (IN-CISM) and its main interest was the combined use of hard and soft approaches in systems modelling.
[] The academic institutions involved in the INCISM network were Lancaster University, Department of Management Science; the University of Strathclyde, Department of Management Science; Cranfield University/Royal Military College of Science; London Business School and the Open University.
Figure 5.29. Modelling for learning
Source: Sterman, J.D., Business Dynamics: Systems Thinking and Modeling for a Complex World, © 2000, Irwin McGraw-Hill, Boston, MA. Reproduced with permission of the McGraw-Hill Companies.
Figure 5.30. Two approaches to business and social modelling
Source: Checkland, P. (2004) Proceedings of the 2004 International System Dynamics Conference, Oxford
Where does system dynamics fit in this scheme (Lane, 1994)? Many people automatically assume that because the approach involves equations and simulations it must therefore be a type of hard system modelling. Equally, other people assume that soft or qualitative modelling is vague and woolly, lacking in discipline. These stereotypes are misleading. More important is whether problematical situations in business and society are thought to arise from a system in the real world or from multiple interacting perceptions of 'reality' in the minds of stakeholders. I will explore this abstract territory here as it relates to the practice of system dynamics, drawing on the factory model to illustrate my arguments. What could be more tangible in the business world than a factory?
Advice frequently given to beginning modellers in system dynamics is NOT to model the system. Setting out to replicate reality is considered to be a futile exercise that prevents people from exercising judgement about what to include and what to exclude from the model. So, to use the terminology of Figure 5.30, system dynamics modellers do not spy systems. Rather they spy dynamics in the real world and they organise modelling as a learning process, with the project team, to discover the feedback structure that lies behind the dynamics. Figure 5.31 is a reminder of this process from the factory model. We began with puzzling and undesirable cyclicality in factory production and employment. This phenomenon in the real world suggested to the modeller that, somewhere in the complexity and confusion of factory operations, there is a balancing loop with delay, which is capable of explaining the observed cyclicality. The search was then on for such a balancing loop – a search conducted in close consultation with factory managers to understand how they go about planning and coordinating production and workforce. This investigation probed many aspects of manufacturing operations and policy from forecasting and inventory control to production scheduling and workforce hiring, and uncovered the feedback loops shown in the top right of Figure 5.31 (which summarise the main closed-loop connections in Figures 5.8, 5.17 and 5.18). There are two interacting balancing loops, one for hiring (that ensures there are enough workers) and the other for inventory control (that ensures there is enough finished inventory – neither too much nor too little). Crucially the inventory control loop is a balancing loop with delay, where the delay arises in adjusting the size of the workforce. This loop is mainly responsible for the observed cyclicality in production and employment.
Figure 5.31. I Spy dynamics in the complexity of everyday life and can discover underlying feedback structure
The factory model is one particular case, but it is representative of broader practice. In system dynamics we take the view that, behind every dynamical problem, there is an enduring feedback structure waiting to be discovered. The structure may be well-hidden in the flux of everyday life, but nevertheless it is there and can be teased out in the ways illustrated with the factory model. The end result is a hard system model in the sense that the loops identified (in this case two balancing loops) are presumed to be there in the real world, stemming from much simplified, but nevertheless realistic, aspects of operations and policy. System dynamics delivers hard system models from a soft and interpretive modelling process (Lane, 2000).
Where in these models are the multiple interacting perceptions of reality that are the hallmark of soft systems? Such perceptions are to be found in the contrast between a whole system or global view (comprising several sectors and interlocking feedback loops) and a local view (comprising a single sector and just a few causal links).[] Feedback loops that are responsible for puzzling dynamics nearly always weave their way across organisational boundaries (Wolstenholme, 1993). Consider the troublesome inventory control loop in the factory model that crosses the boundary between production control and workforce management. For that very reason, the loop is invisible at the functional level. A system dynamics project invariably expands the boundary of people's thinking, requiring them to see beyond the narrow confines of a single function or department to take proper account of organisational interdependencies. This broader view challenges narrow, departmental perspectives and helps to remedy the problems that arise from such myopia. For example, the factory model makes clear that the production department is not fully in control of its own destiny, no matter how much it would like to be. Production is not the same as desired production. If it were, then the two concepts would be directly connected in Figure 5.31 and the factory's feedback structure would reduce to a simple balancing loop without a delay and without cyclicality (as the simulations of an 'ideal' factory in Figure 5.9 showed).
[] For more examples of the contrast between global and local perspectives, see the causal loop diagram about road congestion in Chapter 2, and the many causal loop diagrams based on practical situations in Coyle 1996, Senge 1990, Sherwood 2002 and Wolstenholme 1990.
The reality is that more production requires more workers and vice versa. If people in the production department (wishing to be responsive to demand) think the factory has the freedom to boost or cut production on the spur of the moment then the factory's cyclicality problems will persist. The volatile schedule resulting from this desire for responsiveness ripples into workforce planning and eventually feeds back to production. A feedback systems view makes the production and workforce planning departments aware of their mutual dependence and its knock-on consequences.
5.7.2. A Link to Soft Systems Methodology
A model that raises people's awareness of interdependence (by requiring them to 'think outside the box') is a particular example of a more general learning activity that takes place when using Soft Systems Methodology (SSM), as shown in Figure 5.32 (Checkland, 2004; Checkland & Poulter, 2006). Underpinning this picture is the recognition that all problematical situations contain people who are trying to act purposefully or with intention. In any human organisation, many intentions are in play simultaneously and it is the interaction of these intentions that generates both progress and problems. Here, a perceived problem situation is deconstructed into alternative models of 'purposeful activity' within a declared worldview. These models are not necessarily descriptions of the world as it is, with its current real-world actions. The models are in people's minds and reflect an ideal or desired world consistent with a declared worldview, which may or may not correspond to hard reality. For example, as Checkland 2004 explains, a 'prison' can have a set of relevant models: it can be modelled as a punishment system, a rehabilitation system, a system to protect society or a system to train criminals. Any actual prison is a changing mix of all these perceptions and others. These alternative models of purposeful activity are sketched and then compared with each other and the problem situation in the real world. The comparison, facilitated by the modeller(s), leads to structured debate among the problem owners (often managers) that seeks the accommodations which allow action to be taken to improve the problem situation. And so a learning cycle is established.
Soft Systems Methodology then deals with models-in-the-mind and applies to a very broad class of problematic situations where action to improve is impeded because of conflicts in worldviews. By making clear the models of purposeful activity that belong with these worldviews, and seeking accommodations, the situation can be improved. The approach does not ignore the real world but sees reality as an ever-changing amalgam of idealised models that can co-exist either fruitfully (the goal of a soft system study) or destructively (as found in chronic problematic situations in which overt or hidden conflict is rife).
Figure 5.32. The learning system in soft systems methodology involves a comparison of alternative models of purposeful activity
Source: Checkland, P. (2004) Proceedings of the 2004 International System Dynamics Conference, Oxford.
System dynamics tackles an important subset of these problematic situations where the 'problem' can be expressed dynamically and where dynamics arise from partial and idealised models that coincide with organisational, functional or political responsibilities.[] When combined, such partial models reveal feedback loops in the real world.[] While there is undoubtedly overlap between the two approaches, it is important to realise that the learning cycle in system dynamics seeks to discover enduring feedback structure as a hidden characteristic of the real world. This is hard system modelling dressed in soft clothing.
[] Strictly speaking this 'subset' of problematic situations handled by system dynamics is partly outside the boundary of situations handled by SSM, since it is difficult to reliably infer dynamics without the use of simulation.
[] It is relevant to note that when system dynamics modellers conduct partial model tests they are effectively constructing imaginary models that are deliberately lacking real-world feedback loops, as in the 'ideal factory' simulations presented earlier in the chapter. These ideal worlds are conceptually similar to SSM's purposeful activity models though their stylised content and appearance is very different.
The overlaps between the approaches bear consideration. Among them is the central proposition that organisations contain people and groups who act purposefully. Purposeful action is at the heart of the goal-seeking behaviour generated by balancing loops. Moreover, problematic situations arise from interactions among purposeful actions motivated by conflicting worldviews. However, in system dynamics differences of purpose are distributed across organisations (to be found within functions or departments), whereas in SSM differences of purpose can exist within and between departments and even for the enterprise as a whole. These differences of purpose are deeply engrained in people's taken-as-given worldviews that SSM seeks to uncover.
The distinction I am drawing here is abstract and best illustrated with a practical example. The example uses system dynamics modelling symbols to sketch the kind of alternative models that characterise a soft systems approach. I should stress that this example does not adhere to the conventions and principles of SSM, and neither does it convey the rigour, discipline or richness of the approach. It does, however, illustrate the notion of soft models derived from a declared worldview. Readers who wish to learn the skills of soft system modelling are referred to Checkland and Poulter 2006 and to Checkland and Scholes 1990.
5.7.3. Alternative Views of A Radio Broadcaster
Imagine you are an executive from a successful commercial radio station in the USA and have been assigned to London to set up a rival to Classic FM, a very successful radio station that broadcasts classical music (with advertising and a sprinkling of news and weather) to a region 200 miles around the capital. You want to find out more about broadcasting in the UK and have been invited to sit-in on a budget and strategy meeting of another successful radio station in London (not Classic FM). You know nothing about this broadcaster (other than it too is successful), so you listen. Also, you recall your MBA strategic modelling class a decade ago and wonder if the concepts you learned back then will help you interpret the situation.
You soon realise that the organisation is keen to win listeners while remaining cost competitive with rivals. Those goals are no surprise. The available budget depends on the number of listeners, but how is this budget deployed? Members of the management team make clear that for them it is vitally important to have good staff and well-equipped studios. Also, to win and retain listeners requires the right mix of programmes as well as transmitters to reach the audience. So far, so good. Then you hear some views that surprise you. The team insists that a large fraction of the capital budget is spent on high-cost short-wave transmitters – the kind that can broadcast from a remote location in the UK to almost any part of the globe. Spare capital budget buys new FM transmitters to improve listening quality in selected cities. This transmission strategy is certainly not what you had in mind for your rival to Classic FM, but you continue to listen. The next two items in the discussion really surprise you. There is a strong and passionate case made for broadcasting in at least 40 different languages and maintaining a cadre of specialist correspondents in political hotspots such as the Balkans and Baghdad. You know such experts are really expensive and are puzzled how this organisation can be so financially successful. The final items make much more sense – the team agrees that high-quality programming and an excellent impartial editorial reputation is vital to continued success. Recalling your MBA modelling class you realise there is no mention at all of dynamics in this discussion, it has been all about management's priorities. Nevertheless, you muse on the list of asset stocks that would fit this description – if ever a dynamic model were to be created. The picture at the top of Figure 5.33 comes to mind. Here are 10 asset stocks that management believe are essential to attract the particular audience they have in mind.
As you reflect on the conversation, you note that if you were running this organisation you would cut the language portfolio, redeploy the correspondents, change the programme mix, scrap the expensive short-wave transmitters and invest in a network of good local FM transmitters. The point here is that the model of the organisation you have in mind depends on the purpose you wish it to fulfil.
Figure 5.33. Alternative models of a radio broadcaster
Source: Morecroft, J.D.W. (2004) Proceedings of the 2004 International System Dynamics Conference, Oxford
Your mission is to establish a new radio station that brings classical music to a large domestic audience, and does so profitably. A list of asset stocks that fits this purpose is shown in the bottom of Figure 5.33. There are fewer stocks than before and, of those remaining, the ones in the grey oval would be much different than at present. This second picture is a soft model in the sense that it shows the assets you believe are appropriate for a commercial broadcaster of classical music.
You subsequently discover the organisation you visited was BBC World Service, whose purpose is to be the world's best known and most respected voice in global radio broadcasting, and first choice among the international politically minded elite for authoritative and impartial news. In this example, the two different pictures of asset stocks are a bit like models of purposeful activity with different declared worldviews - one worldview coincides with the current reality and purpose of BBC World Service and the other worldview matches the purpose of a commercial radio station like Classic FM. If these two conflicting worldviews were to co-exist among the management team of a real radio station, then the team would be facing a problematical situation requiring significant accommodations on either side for action to be taken.
Note that neither of these pictures is a satisfactory basis for a conventional system dynamics model because there is no dynamical problem or issue guiding the selection of asset stocks or pointing to feedback loops that might explain dynamics.[] More important, if a system dynamics modeller were working with a broadcaster like World Service, they would not normally build a model containing only the asset stocks in the lower picture (even if a member of the management team had in mind a new World Service like Classic FM) since these stocks contradict what is known about the current organisation.[] However, if the modeller were helping a management team to investigate the start-up of a completely new commercial radio station then they may well develop a model with the slimmed-down assets shown. In that case, attention would focus on growth dynamics and the search would begin for the combination of reinforcing and balancing loops (containing the selected asset stocks) that determine growth. These growth structures are the focus of Chapter 6 and Chapter 7.
[] A system dynamics model of BBC World Service was in fact built to investigate alternative government funding scenarios over a 10-year planning horizon. The problem situation was framed dynamically in terms of future trajectories for the number of listeners and the cost per listener. The project is described in Delauzun and Mollona 1999.
[] In principle a dynamic hypothesis for World Service may exist that involves only the six asset stocks shown, but this is unlikely in practice since (for example) the language portfolio is so important in determining the number of listeners who tune in to an international news broadcaster.