Computer models have been used to describe social and economic systems for about 40 years now. Today, models are routinely used to analyse supply chains, the performance of national economies the spread of epidemics, population growth, the effects of climate change, food production and the spread of pollution. At CoDynamics, we use system dynamics models (or on occasion pure mathematical or statistical models) to underpin much of the work that we undertake. Our reasoning is that most (apocryphally 70%) of change initiatives fail to meet the promises that were made when the project was appraised and approved and this failure is in large part a failure to completely understand the problem which the change initiative is intended to fix. This post elaborates on why models are a vital component of understanding and changing what is going on in an organisation.
A point on terminology: the word model can mean a number of things but in this case, we are referring to simulations – i.e. models that reproduce structure and behaviour so that inputs can be processed and turned into meaningful results.
Everyone Uses Models, Most Aren’t Formal
People use models every day. Anyone who makes a prediction or tries to work out what is going on in a particular set of circumstances is using some kind of model. Our decisions and actions are based on our mental images of the world as we perceive it. When you close your eyes and imagine the consequences of someone’s actions, you are running a model, it just happens to be in your head and not written down.
Peter Senge  popularised the term mental models which he described as:
- Deeply ingrained filters through which we interpret our experiences, understand the world and which affect how we take action.
- Images, assumptions and stories which we carry in our minds describing every aspect of the world. They are unique to the individual and they are all flawed in some way.
- Flexible, dealing with more than just numerical data and they can be modified as new information comes to light.
These kinds of models are implicit. Assumptions are hidden so ambiguities and contradictions remain undetected. Their structure and consistency are untested and usually, they are unsupported by data. Also, they are prone to being not understood by others. Interpretations differ.
Add to this the fact that people are not consistently rational. They are subject to emotions, unconscious motivations and a host of irrational urges Finally, our mental capacity falls woefully short of coping with the complexities of the real world . Psychologists have shown that we can only cope with a few factors when making decisions. In short, mental models are extremely simple and usually flawed.
Explicit Models Don’t Suffer From These Shortcomings
Computer models are explicit. Assumptions are laid out in detail for anyone to challenge and the effects of changing assumptions individually or in groups can be tested. They are comprehensive, coping with many factors simultaneously and they enable others to replicate your results.
Explicit models can incorporate the knowledge of many people, some of whom will be experts in their subject. Used well, they often become the focal point of multi-disciplinary teams that have been brought together to solve a problem.
You can calibrate and test the model with historical and current data and explicit models can be subjected to sensitivity analysis where a number of parameters can be tested to see which ones have the most effect on particular outcomes. With sensitivity analysis you can identify areas of risk and robustness and also, important thresholds. This in turn informs the debate about probabilities of (un)desirable outcomes and the trade-offs and options available to decision makers.
Which is not to claim that computer models are always perfect. Often they are poorly documented and therefore prone to being misunderstood. They can be made too complicated so there is no confidence in the correctness of their outputs and where factors are intangible and not easy to quantify, they can be subject to approximation and guesswork – how do you accurately portray the effects of poor teamwork for instance?
Uses of Models
Epstein  suggests a list of uses for models. The following is my condensed version:
Use 1: Explanation
How results occur in real life is not always obvious. Models can reveal hidden phenomena, weaknesses or imperfections in your organisation and help you understand (explain) how an event came to pass. As Epstein says “ understanding electrostatics doesn’t help you predict where the next lightning bolt will strike” – but it may persuade decision makers to install lightning rods on your buildings.
Uses 2 and 3: To Guide Data Collection and Discover New Questions
Model creation, especially when it is done by a team from within the organisation often highlights inadequacies in the available data which to be fair is usually collected for operational reporting and not troubleshooting purposes. As the model is built up and understanding increases, new theories about behaviour and relationships emerge and to test them, new questions have to be asked and new data collected. More often than not, it is the new questions which lead to breakthrough innovation.
Use 4: To illuminate Core Dynamics
As the famous statistician George Box said, “Essentially, all models are wrong, but some are useful.” and the important point here is that models should not attempt to faithfully recreate reality. A model is a simplified and incomplete approximation. A good model will capture the underlying, often hidden truth about what is happening and bring it to the surface for all to see.
Use 5: To Suggest Analogies
Many seemingly unrelated syndromes have underlying rules and principles in common. Could the take up of a new product spread in the same way as an epidemic? The nobel laureate, Paul Samuelson wrote at length about the similarities between the monopolistic firm and an entropy-maximising thermodynamic system. Identifying an underpinning law can lead to an accelerated and deeper understanding of a situation and its possible solutions.
Use 6: To Demonstrate Trade-offs and Suggest Efficiencies
Once a model has been developed to the point where a simulation can be run, then any of it’s variables can be adjusted to see what effect they have on the outcome. The results can be surprising as the obvious candidates for adjustment often turn out not to have that great an effect when tweaked in isolation. The model is used to understand what combination of adjustments is needed to give the desired effect without the risk and expense of experimenting in real life. The model will also tell you how much of a performance improvement you can expect given a specific set of changes providing a more realistic estimate of return on the investment in changing your operation.
Use 7: To Guide Policy Development
It has long been acknowledged that efforts to solve a problem often make it worse. New policies can create unanticipated side effects and poor organisational performance can often be tracked back to policy decisions which were made to address a particular set of circumstances and never altered when those circumstances changed. Models can be used to show the effect of unhelpful policies.
Use 8: Test Options Cheaply In A Risk Free Environment
Most solutions can’t be tested in the real world because of the time it takes, the expense and the possibility of reputational damage if it all goes wrong. With a simulation, you can experiment to your heart’s content, trying out a range of options and tuning the most promising ones before embarking on implementation.
Use 9: To Communicate With, Educate and Train Stakeholders
One of the causes of resistance to change is a lack of trust – of managers and of the case for change. By providing a real-time demonstration of the problem and the proposed solution, models can help to overcome lack of trust. Turning the model into a management simulator by providing a set of user controls which the audience can alter is a particularly effective way of communicating the problem and its complexities.
It can be particularly powerful if when exposing the model to people, they identify variables or flows which the model hasn’t incorporated but which may be relevant. By adding new elements, which the audience deem to be important and demonstrating the effect (which may be minimal), you are not only demonstrating that their knowledge is being used – you may well discover something of importance.
Management simulators can also be useful adjuncts to the training process when the chosen solution is being rolled out.
Use 10: Reveal the Apparently Simple To Be Complex and Vice Versa
Models can be a very effective aid to decision making. They can reveal that apparently complex problems have a simple answer or alternatively they can be used to demonstrate beyond doubt that the obvious answer to a problem is not the right one and that prevailing wisdom is not compatible with the available data.
Use 11: Test Solutions In Extreme Conditions
Many organisations work well enough in everyday conditions but often, when a crisis occurs or when extreme conditions prevail, standard policies and operating procedures do not cope. Models allow you to test extremes in a safe environment and equip you with the knowledge to create meaningful contingency plans. Models can be used to inform operational risk management strategies.
Some Words of Caution
This doesn’t mean that the creation of models should be regarded as the answer to every problem. People assume that once a model has been created it can be used for prediction purposes, rather like a crystal ball. Although prediction may be feasible within bounds if the model was built with that in mind, it should not generally be thought of as the model’s raison d’etre.
Models should have a purpose and that purpose should be to shed light on if not solve a particular problem. The practitioner who proposes to model your entire organisation should be treated with suspicion. To recreate an entire system will make the model unnecessarily complicated, more prone to error and that much more difficult to maintain. Because most systems exist in an environment that is subject to regular change, the larger and more complicated the model, the more onerous the task of keeping it in line with changes in the real world and the higher the probability of introducing mistakes. The art of model building is knowing what to leave out and the model’s purpose defines the filter.
A model is only as good as the assumptions that it contains. These might be thought of as a description of the physical system and the rules which define its behaviour and the decision making process. Although this sounds straightforward, problems can occur in a number of areas:
- Describing the decision rules: The way that decisions are made must be portrayed accurately even if it is sub-optimal. The model must reflect the actual decision making strategies used by individuals and take into account the shortcomings and limitations of those strategies. Discovering these rules is not easy requiring close observation and at times, the skills of a trained anthropologist.
- Whereas representing numerical data is straightforward, quantifying soft and intangible variables such as reputation, optimism, product quality is not. Much soft data is never recorded and is entrenched in attitudes but if the decision making process is to be accurately represented, these factors must be included. A model which does not embrace qualitative knowledge is not going to tell you very much that isn’t obvious. The difficulty is that people will disagree about the interpretation of soft variables and so the modeller needs to understand through sensitivity analysis how conclusions might change if other plausible interpretations were used.
- Choosing the model boundary is also a delicate task which needs a good deal of thought. Get it wrong and you are in danger of leaving out crucial elements or including too much unnecessary detail. The danger here is that the modeller will concentrate on those factors which she is comfortable with or which the sponsor deems important and in so doing, misses a crucial factor.
Models can’t be built in isolation, indeed they should be a means of uniting stakeholders around the problem – its definition and solution. Building the model with a cross disciplinary team and opening it up to regular inspection and criticism is a key part of making it represent the ‘real’ real world and as a side effect, if the consultation is thorough and wide ranging, it will help to minimise downstream resistance to change.
As long as the strengths and weaknesses of a model based approach are understood and accommodated, modelling a problem is hugely beneficial to the decision making process.The primary function of model based analysis is educative not predictive. Models are not meant to be a substitute for critical thought, but as a tool for improving judgement and confirming intuition, taking human fallibility out of the equation as far as possible. Indeed, the value of computer models derives from the difference between them and mental models. When the two are in conflict, the differences, when discovered will lead to an improvement in both.
This post is based on the author’s practical experience and draws on the wisdom to be found in the following publications:
 Epstein J; “Why Model?”, http://jasss.soc.surrey.ac.uk/11/4/12.html
 Senge P; 1994, “The Fifth Discipline Fieldbook”, Nicholas Brealey Pubishing
 Sterman J D; 2000 “Business Dynamics: Systems Thinking and Modelling for a Complex World”, McGraw-Hill, MIT SLoan School of Management.