The next time you're on an airplane, ask the guy beside you how much his ticket cost; chances are the two of you paid different prices for the same uncomfortable seats. Depending on when you book your flight and how long you stay at your destination, your seat may cost hundreds of dollars more-or less-than your seatmate's.
What sounds like a disservice to full-fare passengers is actually a shrewd strategy that airlines use to maximise profits by forecasting customer demand and charging multiple fares on single flights. Airlines hold as many seats as possible for last minute, full-fare business travelers, but if they miscalculate demand they end up with empty seats.
The only way airlines can make that game of chicken profitable is by using a passenger-demand forecasting system to help decide what to charge when. Most airlines' systems work reasonably well but don't have the power necessary to mine all available passenger data. But United Airlines has found all the power it needs by using parallel processing technology to forecast demand and increase profits by as much as $US100 million a year.
Previously, United crunched passenger data on its MVS mainframe. But when the airline wanted to analyse data more thoroughly, the mainframe, which could measure only point-to-point flights, couldn't keep up. "Sometimes the old system would make the wrong decisions," says Bob Bongiorno, director of research and development at United's Information Services Division. That led to imprecise forecasts and empty seats.
In early 1997 United turned to parallel processing, which uses a large number of linked microprocessors that work together to solve a common problem. As its parallel processing platform, United chose a scalable RS/6000 SP2 system from IBM with Orchestrate middleware from Torrent Systems. The Orchestrate middleware allows United to add processors without having to rewrite core code in the system. "If we had to alter stuff down in the core code every time we wanted to add processors, it would require a lot more risk as far as creating bugs," Bongiorno says.
The new $18 million system, dubbed Orion, debuted in March 1998 and is now helping United executives make better decisions about which fare to apply to which seats. "It has made the planning folks think more along the lines of how people fly and buy," says Ajay Singh, United Airlines manager of R&D. "We're getting closer to a lot more real-world analysis." In addition, Orion allows United to look at flight "paths" rather than "legs," which means that a flight from Boston to Los Angeles, with a stopover in Chicago, is analysed as a Boston-to-Los Angeles path rather than a Boston-to-Chicago leg and a Chicago-to-Los Angeles leg. That probably doesn't mean much to a passenger, but for United it can mean the difference between an empty plane and a full one because United planners know when to lower prices to fill a plane and when to wait for full-fare travelers.
Parallel processing was uncharted territory for United, so the middleware was key in the company's decision to make the switch. Bongiorno was concerned that the new system would be difficult to maintain and expand in the future. "But Orchestrate manages all the lower-level communication for us," says Bongiorno.
The middleware also made it easy for United to scale up its system from 16 processors to 47. "We could do it almost overnight," Bongiorno says.
In the future, Orion can scale up to 512 processors and can even be linked to a second RS/6000. United doesn't need all that power right now, but knowing it's available gives the airline the option of someday using Orion for other applications like plane scheduling and fare setting.
In the meantime, United is enjoying its ability to sell airplane seats at the highest prices its customers will pay.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.