Dynamic systems can't have static solutions
I recently had an interesting conversation on cost of quality (COQ), where a client wanted a 'simple equation to model COQ'. I am sure it does not require a mathematical genius to come up with an equation to calculate COQ, given the values of all the factors that contribute to it are known. We had a long debate on the reason why the formula to calculate COQ cannot be a model to predict it.
The discussion was focused on using the formula and a linear model for predicting the values of individual contributing factors to predict COQ. All seemed so easy and straight forward - we have COQ which is the sum of several cost components and we know the values of those factors on a time series. All we need to do is use some techniques like a curve fitting to project the trend and predict the future values, add them all to predict the future COQ. It took us a while before we could agree on one thing - we need to look at the problem differently!
One of the key revelations from the brainstorming that followed was that many of the 'variables' that we considered varied at varying rates. Of course, variables are expected to vary, but we assume that they follow a fixed relationship (a static model). Usually we use techniques like curve fitting to arrive at a trend and use it to predict the future, which may not be appropriate. For example, a schedule pressure will increase when there is more pending work. This will push productivity up to certain levels and quality down. We found a few more of such dependent relationships involving skill level, process maturity, quality of inputs etc.; finally, we concurred that principles of System Dynamics are a better fit than the conventional models in such situations.
System dynamics is a perspective and set of conceptual tools that enable us to understand the structure and dynamics of complex systems. It allows us to model complex relationships between system components and model their behaviour based on the different mutual relationships and feedback loops. This makes the model simulate results that are realistic as statistics such as rates at which the variables change, the impact of feedback loops on the behaviour of the system and system's behaviour at each instant are taken into consideration.
We at Infosys are experimenting to model a system that can represent the COQ. For this, we are building the various causal loops and a set of components that contributes to or can influence the COQ. The intent is to use this model to predict the COQ in a given context. It will take a while before we complete the exercise. However, the way modelling of the system is shaping up, we are becoming confident about the way we approached the problem.