AI knows how much you’re willing to pay for flights

“We are able to determine at every price point how many people will buy a ticket,” said Roy Cohen, chief executive officer and co-founder of Fetcherr, whose directors include Alex Cruz, a former CEO of British Airways Plc. “It’s very hard to hide from a system like us.”

Brazilian carrier Azul SA last month announced the first public trial of Fetcherr’s demand prediction and pricing technology. Azul didn’t reply to emails asking for more information about the trial.

Fetcherr’s demand simulations are so accurate, according to Cohen, that fares determined by algorithms for flights six months away barely change by the time the plane takes off. “Almost spot on,” he said. “Sometimes to the cent.”

Here’s how AI can help airlines make more money:
Say someone wants to fly from New York to Boston next Thursday to attend a rock concert. An online search turns up flights at $263, $303 and $424. The customer decides $424 is too much and so grabs the seat at $263. This particular band rarely performs, so the person was willing to pay slightly more. The airline doesn’t know that. However, data-hungry algorithms might have worked out that demand for Boston-bound services next Thursday was high enough to charge $293 and still fill the plane. In this case, the airline offering seats at those pre-determined prices could have collected an extra $30.
Aviation needs all the help it can get. Travel evaporated in 2020 as governments around the world closed borders and rolled out Covid-19 restrictions. A recovery from the pandemic should drive global airline revenue to $782 billion this year, still shy of the $838 billion in 2019, according to the International Air Transport Association. Typical annual revenue growth has been in the single digits since the financial crisis more than a decade ago.

While airlines have for years used software to manage airfares, what passengers ultimately pay has been governed to some degree by seat availability in various price brackets. AI seeks to match fares far more closely to passengers’ desire to pay, something that has become tougher to pinpoint after two years of lockdowns.

“The traditional techniques are almost blunt instruments, really, to deliver certain products at certain price points to the market,” Amanda Campbell, solutions marketing director at global travel technology provider Accelya, said in an interview.

AI’s influence on aviation is in its infancy, but the information flows are already too large to sensibly grasp. Cohen reckons Fetcherr alone processes multiple petabytes of data from around the world every second as it sizes up travel demand. A single petabyte is estimated to equate to 500 billion pages of standard printed text. “The bigger we become, the better we become,” he said.

The supply of data is endless, said Conor O’Sullivan, chief product officer at Datalex Plc, a provider of real-time pricing. The Dublin-based company last year announced a trial with Aer Lingus, the Irish airline owned by IAG SA. Aer Lingus didn’t reply to an email seeking details of the tests.

Datalex still leans heavily on historical information such as airline bookings and schedules to estimate current and future flight demand, O’Sullivan said. But computers are increasingly weighing one-off events such as concerts and sports tournaments, as well as hotel reservations and airport queues. Changes in governments and policy, or even a ministerial ousting, can influence the market. It’s the algorithm’s job to determine the relative importance of each byte.

“All of these things have effects,” O’Sullivan said. “Then you get down to all sorts of behavioral psychology. If it’s raining outside, you’re more likely to book to a sunny destination than if it’s sunny.”

While giant AI-powered retailers like Amazon.com Inc. clearly show the benefits of machine learning, aviation’s in-built aversion to risk means it is likely to embrace the technology at a far slower pace. Change in the industry moves at a glacial pace, hamstrung by legacy network systems and aging ticket distribution partnerships.

“A lot of trust needs to be built up before they go full on into something like this,” O’Sullivan said. “They see this as really high potential value, but high risk as well.”

Revenue on some routes can jump as much as 8% with the benefit of AI-driven ticket pricing, though on an airline’s entire network the benefit is likely closer to 2%-3%, he said.

“If it’s raining outside, you’re more likely to book to a sunny destination”

Frequent flyers can provide useful data -- although it’s not personalized -- when they log onto airline websites to plan trips. Browsing sessions that don’t end up with a booking are sometimes as useful as those that do.

“How many searches got abandoned? You have to figure out why,” said Tim Reiz, Accelya’s chief product and technology officer. “It’s about finding the optimum price where the airline can fill aircraft to capacity.”