In parallel with the continued development of the science of CAS, the discovery of new fundamental rules of complex systems and the interrelationships of those rules, the concepts and techniques of the application of the science to business has been evolving.
As in most new software development fields, the early attempts in CAS were “job shop” efforts, where each attempt to apply a feature of the science required a completely custom application development, with its attendant costs in time and energy. While this is still the norm, we are clearly moving toward an environment where each CAS function is an object, living in a software environment which handles the user and data interfaces. Swarm, and Dias are two examples of early environments which bring us out of the custom code arena and into a standardized environment where the costs and time commitment of development are less onerous. NSP's Pro-Mods is an example of such an architecture, custom created to serve the needs of field optimization in an upstream exploration and production company field maintenance operation.
Regardless of the level of sophistication of the application, a healthy amount of organizational change management is necessary to make any development effort successful. Any change to policy and procedure, particularly one in which some of the fundamental “rules” by which the personnel play the game may change, requires training and facilitation to be successful. The most important aspect of the success of a CAS based effort then, remains the one factor that has always been critical, regardless of the nature of the new system, that of winning the hearts and minds of the personnel who must use the new functions on a day to day basis, and their help in integrating the new capabilities into the flow of work in the organization.
Here are a few rules of the road to effective application of the science:
What are the situations in which CAS can be of value in business? When the characteristics of the business situation conform to the definition of a complex adaptive system, i.e., a system that has so many moving parts that it can't be practically solved by a mathematical calculation of all possibilities, one in which there are autonomous intelligent agents operating in a known space for their own benefit, who are capable of independent action, planning strategy, and modifying their own behavior based on the results of their efforts. Any sort of a schedule optimization, a choice between several alternative courses of action, a situation in which there are multiple factors in play, where a compromise of all factors is preferable to a solution optimizing only one factor. Look for multiple factors, frequent iteration, random disturbance of the current state, short cycle times, and many possible solutions.
Practical Expectations
CAS is a new technique which adds a tool to the optimization box, it is not a magic bullet. Practical expectations of what can be achieved should be preserved. If you are optimizing a schedule, how much room for optimization is left? We didn't just start trying to do that yesterday. Maybe a 2% improvement in the schedule is all that is left, and it's not worth going to another discipline to get it. Then again, maybe that 2% is critical to success. Know your potential business value. Optimization… of what? Can you really optimize customer satisfaction, or are there too many uncontrollable variables in that and is it really only practical to think about optimizing the scheduling of service? Decision support… to what precision and for how long? What good is stock out point to 8 decimal places? Be sure of the usefulness of the answer in the business setting. Re-calculation of the truck routing schedule when a unit unexpectedly comes off line is an excellent application, one in which CAS proves superior to previous techniques. Re-calculation of the next six months schedule calls too many variables into play.
Every CAS application starts with building a model of the system, be it a natural gas pipeline, a truck distribution operation or a commodity market. (NuTech Solutions has built a very effective model of Nasdaq, used by the exchange to simulate the effects of regulatory change on the way the market works). The effectiveness of the model, then, is connected to your understanding of the problem. The more you know about the factors that influence the system, the agents and their behavior, the better the model will represent reality and the more effective CAS techniques will be in improving performance.
My first experience in building an agent based model was for a group of natural gas marketing executives. Once they began to grasp the potential of the science they said, “great, build us a model that will predict natural gas prices at the Henry Hub”, a major pricing point for natural gas in Louisiana. Woops, sorry about that. If it worked like that the original thinkers in the field would have already built the models to predict the stock market (some of them are still working on this and have by no means given up) and retired to the islands. This is due to another of the features of a complex adaptive system, popularized as the “butterfly effect” by Edward Lorenz in a paper entitled “Does the Flap of a Butterfly's Wings in Brazil Set Off a Tornado in Texas ?” Minute variations in the input data set may lead to widely disparate results. Well, why model, then? Even though CAS can't “solve” a model for all time, it can do an excellent job of understanding the behavior of the system and its agents in a variety of circumstances, can find the stable and chaotic states of the model, and can find the most productive states of the model, which usually lie along the boundary between the stable and chaotic states, sometimes referred to as “the edge of chaos”. This deeper understanding of the behavior of the model can then be used to analyze the observed state of the real world system, and make intelligent predictions as to the probability of alternative future states. All this in reference to building agent based models to explore the nature of a system.
When the objective is to optimize a system for multiple variables, a considerable subset of the agent based modeling approach, the situation, and the results are much clearer. Here we are dealing with, say, a truck distribution system with product in certain locations, trucks and drivers with certain capabilities, a set number of alternative routes, and a known number and location of destinations. Not a new problem, of course, with a wide range of systems already available to solve the problem. The value of CAS in this instance is not so much in its ability to improve on the original schedule as it is to adapt to interruptions and unscheduled changes to the schedule during its execution.
Often the value of the CAS module is its ability to see across several business systems, and to make operational decisions with the benefit of knowledge of the economic consequence of the alternative courses of action, rather than a more traditional solution which sub optimizes on the operational efficiencies, alone.
CAS applications are voracious in their need for data. The more data available the more effective the modeling and optimization will become. Yet, because the objective of the modeling is most often to “globalize” the optimization of the entire system in an environment where the fragmentation of business processes and legacy information systems is part of the problem, a key component in the implementation will be the creation of a meta-model of the data required to run the model, and the sourcing of that data will range from computer interfaces into legacy systems of a variety of types and formats, to the manual acquisition of hard copy data. Creation of a data access engine that automatically consolidates the required information and communicates results back to the affected systems and business units is a necessary component of any application that wishes to move out of the “stop time analysis” mode and into the production stream.
Most business managers are acquainted with the systems development life cycle with which their IT department manages the development of new software. In the case of the re-development or improvement of the business processes within the company it can be relied upon to put some parameters around the investment of time and resources necessary to build the systems. It is natural for the business community to expect the same predictability from the personnel developing a CAS based application. However, because the CAS project often seeks to create a new understanding of the interrelationships of systems and processes, and because the selection of the particular tool of complexity science, agent based model, genetic algorithm, neural network, etc., the initial phase of the project must be regarded as a research and development project, not a systems development project. A business value assessment must be conducted which states the nature of the improvement envisioned, and a series of prototyping exercises must follow which either validate or invalidate the assumptions made in the assessment. Only then, when a scope of work has been defined for which a particular business result may be reasonably expected, should the traditional methods take over.
CAS holds the promise of giving you new insights into the nature of traditional problems and allowing you to formulate new solutions. When it works best, it allows you the advantage of a new point of view on the subject and can revolutionize your thinking. That is the real payoff of CAS. Case in point, in a distribution system optimization, it was discovered that a fundamental law of truck scheduling, that of not letting the truck go to Toledo until it was full at the loading dock, was a sub optimization that interrupted the “flow” of the entire system. Thinking about the optimization at a higher level, optimizing the entire net benefit of the system rather than of its component parts led to a completely different philosophy of the systems operation and a new definition of success. This, in turn, caused a revolution in the policies and procedures, as well as the information systems, which controlled the distribution system, resulting in huge reductions in the time and cost of the system. |
