American Airlines
All firms engaged in activities as a tactical entity will, in some form or another, attempt to get a handle on expected demand for their products within a certain future time period such as a week, month, quarter or year. This is a tactical environment and, aside from any earth shattering new developments or shocks to the existing environment, forecasts for expected demand/maximum-likelihood share of market may be made with a fair degree of accuracy with little variance. There are several key points that are important to this process, such as: activities of competitors, market projections for the industry by industry insiders/analysts, and a great deal of historical data.
Competitive intelligence is a parameter which attempts to add subjective background to the environment in which demand forecasting is carried out. Information comes from a variety of sources such as secondary information gathered from written sources, direct observation, and from competitors themselves through press releases, industry gatherings and trade journals. This information provides some indication of what the competition plans to do as far as pricing, new products, promotions and distribution/sales.
This data has a dual purpose since it may also be used within model based contingency planning when management scrutinizes competition in an effort to uncover developing threats and opportunities. Experienced tactical managers have the valuable ability to incorporate this type of information, which is not easily quantifiable, as a complement to the numerical aspects of demand forecasting. However, this is not to say that there is no information system requirement for this input into the demand forecasting process simply because it is difficult to assimilate into an objective, quantifiable form. On the contrary, a database should be set up in the context of an expert system to contain information gathered on competitors. It must be readily accessible, updated and accurate in order to aid tactical management in this process.
Another input item for demand forecasting comes from aggregate market projections. These types of analyses are readily accessible, mostly in the form of secondary information found in trade journals and economic publications. Airlines and transportation in general comprise a large industrial group within the economy of the United States and, accordingly, there is a large interest in its economic future. Wall Street brokerage firms and other financial firms are resplendent with analysts, some of which are charged with the task of tracking the airline industry’s past economic performance, as well as anticipated future projections. All of this knowledge is available from many sources and, again, wise tactical managers will take the time to incorporate it. System facilities required for this type of support for demand forecasting are databases which can contain quantifiable economic information.
Since this input to demand forecasting is quantifiable, a database with analytical utilities for ranking and analyzing stored economic projections and raw data are used. This facility may also be presented to management in the guise of a dressed up expert system containing decision table constructs which will allow them to adjust many demand forecasting parameters in order to make the most accurate forecast.
Arguably the most important input into the demand forecasting process is a firm’s actual historical data from its own internal records sources. Historical sales data may be thought of as the most dependable and accurate input into demand forecasting since it is derived by the firm itself rather than arriving in a second hand fashion from sources outside of the organization. Historical sales data is helpful not only in developing a demand forecast, but is also used as a check against post production performance when the time arrives to compare actual demand to the forecast. This information will likely come from another massive record keeping database which records sales transactions from the point of sale. For American Airlines, as well as the rest of the airline industry in general, this requirement is served through a reservation system of some kind. The reservation system must be capable of handling queries, data inflows and other types of processing from thousands of nodes. Dummy terminals, which simply display data, will not be sufficient to satisfy reservation system requirements, and any implementation will involve connections and terminals designed to carry two-way traffic. Additional discussion of reservation systems, including specifically what American Airlines has installed, will follow later in this paper.
After satisfying system requirements for generating and handling inputs into the demand forecasting process, the actual forecast derivation may be viewed as somewhat mechanical. The main management decision at this point is determining which type of probabilistic instrument to use with which analytical utility to yield the most accurate results. Some tactical managers may even require an expert system that does nothing more than aid them in selecting the proper mathematical tool to address the forecasting process. There is an array of probabilistic techniques that can satisfy this management requirement including least squares regression analysis, weighted scenarios, Markov-based stochastic projections and others. Many tactical managers may use a combination of these facilities to arrive at a forecast with which they feel satisfied.
A key point to bear in mind when discussing demand forecasting for a tactical entity is that it is central to two important aspects of the firm. The demand forecast is viewed foremost as the progenitor of the firm’s production for which it is the main, direct input. However, it is also an indicator of the general trend of the firm’s revenues over time. A forecast whose extrapolation to the next period indicates a decline in revenues may be an early warning of something novel in the industry or indicative of a paradigm shift toward a new era. This aspect of troubleshooting will be discussed more at length in a later section concerning requirements for process control.
The demand forecast sets the stage for the next management task — logistical programming and its accompanying system requirements. Logistical programming is the task charged with accumulating proper amounts of the factors of production in the proper place at the proper time. The four factors of production (material, finance, equipment and manpower) have certain input requirements which determine the amounts of each factor to apply to the production process. Each of these inputs will necessitate the use of some type of information system to aid tactical managers in allocation of these factors to production. One of the first inputs into logistical programming is the supply schedule, which is the main determinant of the amount of products or services offered by a firm. For the airline industry, supply schedules manifest themselves in the form of the magnitude of flights offered to the public.
A demand forecast is the main force behind the supply schedule, but other normative microeconomic factors play an important role in its composition. One of these factors, optimal scale of plant, exerts a direct relationship against the supply schedule and, for American Airlines, consists of the optimal terminal/gate layout at its busiest hub cities. The goal of proper terminal design is to optimize the number and size of the complexes which converge on a hub terminal throughout the day. A complex consists of a group of inbound flights which land within minutes of each other and take-off within minutes of each other. This is the very heart of a hub and spoke system which allows a large number of flights due to the number of possible connections in the hub. Inbound passengers from many cities will all arrive at approximately the same time, disembark, and make connections to many outbound flights which leave within minutes of each other. This occurs many times throughout the day and the system requirement for solving this problem and optimizing the operation is available in the form of CADD design stations.
CAD/CAM design workstations may be used to solve terminal optimization problems and allow engineers to simulate the capability of the terminal to handle certain scenarios. This is, in fact, exactly what American Airlines did when it was searching for the optimum design for its $80 million expansion of its main hub in Dallas/Fort Worth in 1983. This simulation model was used by senior management to aid them in their decision on the best design to handle the desired flow of traffic in the narrow operational time constraints necessary for the hub to work. In addition to optimizing the terminal layout, the system was useful in optimizing other related areas. The system/model was used to determine dynamic gate assignments in order to minimize baggage handling costs and passenger delays. Another byproduct of the model was a useful algorithm designed to automatically program and update signs for directing passengers around the terminal. The functional facility was even used to determine the best layout for short-term parking in the face of expected increases in passenger traffic.
Though optimal scale of plant through optimal terminal design is an important aspect of American Airlines’ supply schedule determination, the most important part of the supply schedule lies in determining the number of flights to and from certain destinations. For American Airlines and most of the airline industry, flight scheduling is not a simple matter. Flight scheduling is one of the most important tasks performed by tactical airline managers because it is central to where and how the factors of production are allocated. The technical system requirements are myriad, and they must meet the daunting problem of properly scheduling thousands of flights per day between hundreds of domestic and international destinations using a fleet of over 500 aircraft. One main requirement is for a system capable of analyzing past flight offerings in search of patterns of overbookings and empty flights in order to adjust schedules to better meet forecasted demand.
Technical requirements for an airline scheduling system would include a data base structure to house the body of past and present schedules from which managers could choose when composing a new schedule. The problem is compounded since airline schedules are determined months in advance. In addition to using optimization techniques, the system requires certain expert system facilities such as decision table constructs to aid in schedule development. Diagnostic remedial aids are used in order to spot bottlenecks in the proposed schedules where patterns of frequent overbookings are occurring. In addition, remedial systems capable of offering solutions by reshuffling proposed schedules provides valuable information to flight scheduling managers. Historical data is fed into the scheduling model from the database containing past schedules and data concerning past parameters which influenced those schedules. The system takes this data and combines it with the demand forecast in order to develop a preliminary schedule. The process requires diagnostic and remedial systems to optimize the schedule so that the expected demand will be met in the most efficient manner possible.
Even with an optimal schedule in place, there will always be disruptions due to weather and shortages of planes and crews; thus forcing scheduling managers to constantly rearrange flights. Before 1991, this was a complex task for American Airlines since dispatchers had to scan data from many different mainframe databases in order to get a handle on managing daily flights. The schedule was constantly being reconfigured to meet anticipated external obstacles such as delays due to inclement weather. In 1991, however, American Airlines invested in a new system known as Smalltalk which made schedule maintenance easier and more efficient. Smalltalk uses of object-oriented programming techniques in order to keep flights running smoothly. The dispatcher simply clicks on an object representing a flight and, when he changes the flight, the system automatically updates other objects (flights) in the system in order to propagate the change throughout the entire system. In fact, it only took three programmers eight months to write the program which contained only two errors.
Once an optimal schedule has been developed through simulation and optimization techniques, the next step is to arrange the factors of production in order to generate enough products and/or services to meet prospective demand. Since manpower costs equal over one-third of all expenditures for American Airlines, it is the first factor to receive consideration. Manpower for an airline takes on many forms; however, almost all of the employees of American Airlines can be classified into one of three different broad categories. The first category represents the aircraft crew whose duty stations are on the aircraft: pilots, copilots, navigators and flight engineers, as well as the cabin crew or flight attendants. The second category is referred to as maintenance workers, and they are the people that maintain the aircraft, which includes anything from refuelers to engine mechanics. The final classification includes all of the ramp workers such as baggage handlers, ticketing personnel and office workers. By far the most difficult category to allocate within the manpower group is the aircraft crews.
Manpower requirements for airline crews are derived from the flight schedule. The main goal for crew schedulers is to develop a schedule for the entire following month which will ensure that all of the upcoming flights for the month are properly staffed. Flight crews at most airlines bid by seniority for the flights that they will fly in the next month and crew schedulers develop flight packages for them. The flight packages are known in the industry as bidlines. The bidlines in turn are composed of flight segments called trip pairings, and they customarily cover a one to three day time frame. Compounding the problem for the schedulers are FAA and union work rules designed to minimize the risk of accidents resulting from crew fatigue. Therefore, the main requirement of a generation and optimization system is that it is able to find the optimal set of bidlines (i.e. the set which yields the lowest cost) which maximize the utilization of each crew member, evenly distributes flying time among the bidlines and covers every scheduled flight.
The properties inherent in the crew scheduling dilemma require an expert system design. The first part of the system uses manpower loading algorithms, the current and previous month’s schedules (from various databases) and optimization techniques in order to develop the set of trip pairings, which would adequately cover all scheduled flights for the upcoming month within FAA and union work guidelines. The trip pairing process is made even more onerous because American Airlines operates several fleets of different aircraft and most pilots are trained to fly only one type. The following diagram illustrates the requirements for a crew assignment system.