But the uneasy pandemic years have made all of this more complicated. Oren Etzioni is now the CEO of the Allen Institute for AI, but in the early 2000s he built—and sold to Microsoft—one of the first airfare prediction tools. Prediction algorithms are pretty good at reweighting the importance of different factors as the world changes, and, he says, “they have a shot at adjusting automatically by having the freshest available data.” But that can take some time, according to Etzioni: days, if not weeks.
Google Flights helps customers track down the least expensive tickets for their preferred routes and dates. But since spring 2020, the search engine has significantly cut down on the number of “predictive insights”—forecasts of when prices are likely to go up or drop—it offers searchers. In general, Flights aims for 90 percent prediction accuracy, says Eric Zimmerman, the director of travel products at Google. “With the increased volatility in airfares, it has become more difficult to reach that high level of confidence,” he says. The pandemic and its effects on air travel also pushed the company to halt an experiment launched in summer 2019, in which it guaranteed fares for some specific itineraries and would send flyers refunds if the price dipped before takeoff. It could bring the project back soon, Zimmerman says, as the industry starts to stabilize.
Giorgos Zacharia, president of online travel agency and search engine Kayak, says he has a team of MIT PhDs who spend their working lives tending to the website’s price-prediction tool. While the prediction algorithm, first launched in 2013, usually needs adjusting every few years, he says, the past two have seen “serious retraining” every few months, and sometimes every few weeks. He says that the accuracy of the prediction tools, which is generally around 85 percent, may have periodically dipped in the last few years—maybe closer to 83 percent. That means that, at some low points, waiting or buying when the website told you to was less likely to have led to the lowest possible price—and could have led, instead, to some light fist-shaking toward the sky.
“Machine learning likes to learn from old and past repeatable patterns, and make predictions based on the likelihood of those patterns working again,” Zacharia says. “So the pandemic, which brings a lot of unexpected outlier events, also affects the input data of models like this and makes it a more challenging environment.”
Hayley Berg, the lead economist at Hopper, says the company’s predictive tool is trained on 75 trillion itineraries and eight years of historical price data. But today the algorithm more heavily weights what it’s seen in the past three years, which has helped the tool maintain 95 percent accuracy throughout the pandemic, according to the company. Even in the first few days of Covid-related shutdowns, she says, Hopper got its airfare price predictions right 90 percent of the time. Still, customers shouldn’t be shocked by price volatility—Hopper has found that the average domestic flight changes price 17 times in two days, and 12 times if it’s international.
All those changes lead to plenty of conspiracy theories among ticket buyers, even those who don’t bother with price-prediction platforms. No, executives say, airlines aren’t tracking cookies and hiking prices if they see you’re interested in a certain route. (Zacharia, the Kayak president, does say that fares are occasionally higher or lower depending on your location when you’re searching, because the systems do take “point of sale” into account.) No, there’s no reason why flights would be cheaper on a Tuesday than any other day, a persistent rumor among bargain hunters. “The best time to book will depend on your trip, specifically the origin, destination, departure, and return,” says Berg. “And it can be wildly different depending on where you’re going.”
Today, though, it doesn’t always take a sophisticated machine learning algorithm to pick the best time to buy—there is no good time. Prices are so high, says Victoria Hart, a spokesperson for Kayak, that there aren’t “many ‘wait’ indicators these days.”