You are here

Supporting complex decisions

8 September, 2015 - 11:09

In an unstructured decision context, there may be numerous factors or variables that need to be considered. Often, an attempt to find the “best” decision with respect to one factor will lead to a poor solution with respect to another. Even when the situation is very complex, however, it is often possible to use information systems to help support the decision-making context.

An entire branch of management theory and practice, termed management science, has evolved to try and bring more structure to unstructured decision situations. Management science is based on the application of mathematical models, and draws fairly heavily on the use of statistical analysis techniques. Examples include the use of regression analysis (to assess possible empirical relationships), simulation (to identify potential solutions by varying certain assumptions), and optimization models (to generate a “best” solution when resource constraints exist).

When a mathematical model fits well with reality, it may be possible to create an information system to help automate the decision. When the fit is less than perfect, we need to augment the use of the model with the judgment of a human decision maker. The term decision support system is sometimes used to describe information systems that are designed to help address unstructured decision situations. Many decision support systems use management science techniques to provide decision makers with alternative options.

Decision support systems do not necessarily need to be large, complex information systems. For example, a sales manager for the chair assembly company might use spreadsheet software to develop a forecasting model that could be used to predict demand levels for a product (line of chairs). After building the model to include criteria believed to impact demand (price, success rates for promotional [marketing] campaigns, etc.) the sales manager could use it to help forecast demand and then decide what demand levels to forward to the production group.

Consider a more complex example. In the early 1990’s, American Airlines was faced with the daunting task of scheduling about 11,000 pilots and 21,000 flight attendants on close to 700 airplanes on flights to over 200 cities. In addition, they had certain constraints, such as the maximum time pilots and flight attendants can be in the air during a specific time period. This problem could generate between 10 and 12 million possible solutions (U.S. News & World Report, 1993).

The scheduling challenge facing American Airlines (and every other major airline) is very complex. When you consider overtime costs, labor contracts, federal labor mandates, fuel costs, demand for routes, and so on, it is obvious that there is no perfect solution. If a solution is derived that is “best” for one dimension (e.g., reducing overtime pay), another dimension (such as holiday preferences) will likely be compromised. To address the situation, American Airlines spent over two years working on a scheduling system that used management science techniques. The result was an information system that saves the company between $40 and $50 million U.S. dollars per year, by reducing wasted flight crew time.

The scheduling information system uses data such as flight crew availability (e.g., federal mandates concerning necessary “down time” between flights, etc.), flight crew capabilities (e.g., pilots licensed to operate certain types of airplanes), flight crew preferences (e.g., base airport, requested vacation days, etc.), airplane characteristics (seating capacity, range, etc.), and route characteristics (e.g., distance, historical demand, etc.) as input. It then uses the optimization rules and logic embedded in the software to generate possible schedules that satisfy as many of the constraints as possible. The proposed schedules could be viewed as information, which is then used by decision makers (those responsible for scheduling airplanes and flight crews) to produce final schedules.

Decisions concerning the scheduling of airplanes and flight crews to flight routes has observable outcomes (e.g., the number of times a flight is delayed because a flight crew was delayed, the number of flights crew members complete over a given time period, and so on). These outcomes can be measured and examined, leading to greater insights (knowledge) which can then be used to fine-tune the rules and logic in the scheduling information system.