IE 11 is not supported. For an optimal experience visit our site on another browser.

The software that hunts low fares

/ Source:

America West’s online site, like most of its rivals, used to ask prospective passengers when they wanted to travel and returned a price quote accordingly. Unlike most of its rivals, however, America West relies heavily on leisure travelers. So when customers said they wanted an easier way to search by price, the airline listened.

Price-conscious customers were frustrated by how long it took on traditional systems to hunt for low fares. So the airline offered a different method: Choose your fare, then browse the available dates.

“Before, when you picked a market you’d have to go back and forth, back and forth, back and forth to pick a lowest price,” says Chris Stanley, America West’s director of Internet distribution. “This essentially allows you to search by price.”

It would have been a massive project to reprogram their own internal computer system, so instead they just bypassed it.

It was a logical move, because the old systems still in use by most major airlines and travel agents are in essence an electronic update of the way travel agents checked fares by hand in the 1930s: flipping through fare tables and checking them against a list of rules.

These “global distribution systems” (GDS) are programmed to retrieve fares for a specific route and check them against airline fare rules and limitations. A specific route at a specific time costs a specific amount. For 30 years, huge mainframes have churned through those fare tables — usually updated every day — to find fares based on specific requests by customers or travel agents.

The method works fine if you have a specific flight and you want a price, but it was fundamentally designed to retrieve information, not analyze it — much as what might happen if you send your child to the supermarket to buy a specific brand of soap. You specify what brand you want, give them money, and you’re likely to get exactly what you asked for. It doesn’t matter if there’s a sale on another brand, or a 2-for-1 special, because you asked for something specific.

And that’s very much how legacy systems check airfares: with little deviation from predefined rules, in a serial pattern, one after the other. It can be effective if you want a specific item at the supermarket, but for airfares between any two cities (known as a city pair) there can be endless combinations when you consider stopovers and times of day.

“It’s not like I go to the store and there’s three kinds of soap,” says Jeremy Wertheimer, who runs ITA Software, based in Cambridge, Mass. “There’s a hundred million different ways to do this.”

New ways to search
America West turned to ITA and Wertheimer, who had studied artificial intelligence at MIT, to set up its new system. Though the site already had a low-fare finder, its new version offered shoppers dozens more options and found new fare combinations that hadn’t previously turned up.

The change may have been subtle to America West’s customers, but it is part of a profound behind-the-scenes shift in the way airlines use computers to track and sell their fares.

ITA brought two new ideas to the table. Its system accesses the same set of fare data used by airlines and travel agents, but it uses modern distributed computing techniques to search that data, operating on rows of small, Linux-based servers. That shift away from massive, mainframe-based systems is no longer revolutionary in the wired world, but it’s a leap forward in an industry that’s slow to embrace technology.

The big change, though, is how fare data gets searched. Airfares are usually computed by pulling specific figures from price tables and adding up all the flight legs a customer wants. But even if a flight can be priced, that doesn’t necessarily mean a customer can actually buy it: Attached to these price lists are rules that govern which flights actually can be purchased.

That system is complex enough as is, but when you consider how many ways you can fly between major U.S. cities, factoring in different airlines and plane changes, the options are staggering. With all those combinations, a mainframe search of every route option would take enormous computing power, so most search systems usually returned five or 10 options and then stopped. And customers grew to expect just a handful of flight options, most not realizing it was because that’s how airline managers created the systems.

“It was probably easier to go out and affect the behavior of tens of millions of customers,” Wertheimer says, “than it was to affect the behavior of 10 people in the industry who are used to doing things in a certain way.”

'Teaching the industry'
About 10 years ago, Wertheimer and his team set out to find algorithms that could effectively search every possible combination and sort them by customers’ priorities. If the mainframe systems were designed for specific searches — one question, one answer — ITA’s algorithms were built for “fuzzy” queries. When someone plans to fly from Washington, D.C., the system may know intuitively to also check fares for nearby Baltimore-Washington Airport. And because it has fare data already uploaded into its system and doesn’t have to query the mainframes, it can quickly produce a full range of possibilities.

That flexibility allows it to return far more information than a traditional airline search, which may be why it drew clients from all corners of the industry — airlines (Continental and Air Canada also use ITA), travel search engines and even GDS operators like Galileo, which are used by travel agents to access fares.

“What ITA is doing is teaching the airline industry how to price,” says Henry Harteveldt, a travel industry analyst for Forrester Research. “ITA basically said, ‘Point A and B are your starting and end points and everything else is up for grabs.’”

Web ticket shoppers are probably most familiar with ITA’s fare matrix, which has been adopted as a hallmark of one of its biggest clients, Orbitz. The Orbitz matrix shows a variety of airlines across the top, with different route options (non-stop, one stop, etc.) and fares set in rows. A single search can return hundreds of options, with the best deals highlighted. (Those deals are usually also available on competing sites, though they may not always be so prominent.) Though ITA’s algorithms were originally programmed for domestic routes, the firm is now applying them to international fares — after adding in fare rules for dozens of countries and carriers.

Popular option
ITA’s system is likely to get competition, especially as travel software firms like Sabre work to overhaul their systems with smaller, more flexible technology.

But consumers remain hooked on ITA’s price-first searches — long an unattainable goal for travel shoppers — which are available both on Orbitz and, in addition to America West’s site. Online ticket sales are growing, estimated to expand by from $12.8 billion to $15.8 billion this year, according to Forrester’s Harteveldt, and nearly six in 10 online ticket buyers believe price is more important than the airline they choose. Perhaps that’s why Stanley sees the price searches as a growth segment for his airline: “We also think for the business traveler … there’s a lot of pressure out there to cut your costs.”

Speaking of cutting costs, America West estimates the new system saves it $100,000 to $200,000 per month. Its previous supplier charged a per-transaction fee, essentially charging the airline to check its own flights. Its Web site still uses the old reservation system to actually book tickets, but gets fare data from ITA and only taps into the legacy system when a customer is ready to buy a ticket.