A Blog by Jonathan Low


May 21, 2021

How Covid Disrupted Air Fare Prediction Algorithms And Other Behavioral Models

Covid upended many assumptions about human behavior - including the algorithmic models on which many organizations have relied to predict human behavior. 

From failures to anticipate ecommerce demand for everything from groceries to home workout equipment to the increase in the popularity of delivery and on to voting patterns, commercial and social science algorithms had to be adjusted. Perhaps none have required more recalibration than those used to predict airfare pricing. As travel fell and then, belatedly, has begun revive, traditional models have failed to accurately capture changing patterns. As economies reopen, it is likely that algorithmic predictions will regain their utility, but those using them have been reminded that historical data are not always predictive of future behavior, especially during times of great change. JL

Jon Sindreu reports in the Wall Street Journal:

Airlines use the pricing department to set prices for each trip and cabin type. Demand falls to the yield revenue-management department. 10% improvement in the accuracy of forecasts increases sales 1%. This is managed using algorithms. Based on historical data, airlines can anticipate demand. This went out of the window during the pandemic. With historical patterns unreliable and live data muddied by cancellations, algorithms published ludicrous prices. Booking visibility is still close to zero because international and business travel is yet to return. E-commerce-like simulations could boost revenues, but in the real world, too much noisy data urged caution. Humans had to take over.

Summer vacations are coming back. So are the airline algorithms that know how much you are prepared to pay for them.

In the leisure market, domestic air travel is normalizing in terms of both bookings and fares. Crucially, vaccinations are making people start to book further out: Online travel agency Skyscanner estimates an average booking horizon of around 75 days in March, compared with a low of 55 last July.

While most businesses charge the same price for the same product, airlines’ secret sauce is so-called price discrimination: selling equivalent seats at different rates to different people. As a result, they don’t just need demand to be stronger, but also somewhat predictable.

Because the summer “seasonality” is starting to look more normal, “we will be able to hold our yield-management strategies, especially when we talk about the weekends of the peak periods,” Spirit Airlines Chief Commercial Officer Matt Klein recently told analysts.

Traditionally, airlines use two different teams to manage fares. The pricing department sets a range of prices for each trip and cabin type. Responding to demand then mostly falls to the yield- or revenue-management department, which chooses how many tickets of each class to make available. As seats in one fare category fill up, buyers are bumped to the next.

This process is managed using a complex set of algorithms. Based on what has happened before, airlines can anticipate how strong demand will be on a particular day and time, or exactly when people will fly to visit family before a holiday. Much revolves around corporate travel, which is a big chunk of airline profits: Business fliers avoid Tuesdays and Wednesdays, prefer short trips to weeklong ones and book late. They are the reason why carriers hold back seats even at the risk of not filling planes and block cheap fare classes close to departure.

All this went out of the window during the pandemic. With historical patterns suddenly unreliable and even live data being muddied by cancellations, algorithms published ludicrous prices. Humans had to take over.

Recovering these lost efficiencies should be an extra boon for carriers as travel returns. As a rule of thumb, a 10% improvement in the accuracy of demand forecasts increases sales by 1%, said Benjamin Cany, head of airline offer optimization at Amadeus, which builds revenue-management software. This recovery is still mostly an American domestic story. On Monday, European budget leader Ryanair said that, while sales picked up a bit in April, booking visibility is still close to zero. Even in the U.S., historical data alone won’t do, because international and business travel is yet to return.

But the pandemic has served to stress-test useful innovations, such as putting greater weight on recent booking figures and using online searches to predict when and where demand will pop up, in the vein of e-commerce specialists like Amazon.com. Simulations had long suggested that “dynamic” or “continuous” pricing—fully varying the fare charged to each passenger based on live data—could boost airline revenues by up to 7%. In the real world, the danger of introducing too much noisy information urged caution.

“Before, we weren’t at the stage of using this technology massively; now it has been accelerated,” Mr. Cany said.

Carriers will need to learn the right blend of historical and live data. Also, upping prices on consumers in real-time could provoke a backlash. Yet Covid-19, having almost killed airlines’ pricing bots, could end up making them even stronger.


Post a Comment