FILLING PLANES with passengers is only a small part of an airline's worries. After all, the passengers are more than just variables in some super-complex equation -- they're customers. With deregulation and increased competition, switching carriers because of a bad experience with an airline is all too easy to do. And the shortest path to disgruntled passengers is by delaying them before they've ever reached the plane. So how many customer service agents does an airline really need for ticketing? Check-in? How many baggage handlers? And what's really causing the bottlenecks in getting customers from the front door of the terminal to the plane? Would it help if self-service points were added along the route? If so, where?

Rather than finding out the answers by lengthy analysis and then trial-and-error "experiments" where real customers would have to bear the brunt of any "errors," why not simulate the entire process?

AIR CANADA wanted to reduce waiting time for its passengers. At the same time, the airline knew that greater economic efficiencies could be obtained by adding new technologies such as self-service kiosks and electronic ticketing. But it didn't want to invest heavily in new technology and disrupt established processes before knowing precisely which changes would yield the best results. The airline had suspicions, but not definite answers. The solution offered by an IBM team drawn from Research and the Travel and Transportation ISU (known as the IBM Journey Management Library™) provided some interesting results.

First, the team began gathering domain knowledge, building a baseline "as-is" model -- a model of current processes at the airline's Toronto airport based on actual operational data from ticketing, passenger check-in, special assistance and special services, and gate control and baggage handling. (The validated "as-is" model can then be used as a basis for "what-if" studies.) The team ran simulations of entire airport processes using forecast data as input. The simulation process also collected performance measures, including peak and average wait times, peak and average numbers of passengers waiting in line, and resource utilization. It then automatically took a snapshot of system conditions whenever a simulated passenger experienced service that did not meet Air Canada's standards.

Some of the results surprised Air Canada. For instance, while intuition had favored a one-stop self-service kiosk for both check-in and baggage drop-off, the simulation showed that faster service could be obtained with a separate, staff-assisted baggage drop-off. Air Canada is now piloting IBM kiosks at their Ottawa airport facilities, after modeling and simulating them using the IBM technology.

This ability to simulate can help a business like Air Canada manage growth, control costs, and accurately plan for the future. The real beauty, of course, is that this crucial insight can be obtained before actually investing in new technology. And in Air Canada's case, perhaps even more importantly, before finding out from irate customers that a little staff assistance in the right place could have helped speed them to their destination.

FUTURE APPLICATIONS: The field of simulation is ripe for Deep Computing applications, especially since we now can simulate things never before possible. Think of IBM's recent role in providing the necessary processing power for the ASCI Blue project, where the safety of nuclear stockpiles will soon be verified through simulation, meaning nuclear explosions need not be physically carried out. Increasingly, digital things will be able to replace physical things that would otherwise be too costly, dangerous, or time-consuming to test and analyze.

For example, in pharmaceutical research, instead of actually mixing chemicals in a beaker, testing them on animals, and then (depending on the test results) running trials on human candidates, imagine being able to understand the efficacy of a drug digitally...on a computer. Less time, less money, and a higher degree of certainty than the animal to human leap.

This kind of powerful simulation and modeling can also be extended to areas such as genomics (efforts to map and understand genetic codes) and weather forecasting. But the real promise -- and power -- of Deep Computing simulations lie not so much in the simulations themselves, but in what can be done based on them. For instance, knowing weather patterns, temperatures, and major weather events in advance has far more important implications than simply knowing whether or not to bring an umbrella to work. Utility companies could use that sort of knowledge to plan energy production, distribution, and pricing. Transportation companies could alter traffic plans and contingencies. Apparel manufacturers could avoid waste and match production to actual demand; there would be no need to flood the market with down-insulated coats when an unusually mild winter approaches.