This is a flight booking website called Fareboom. com. And this is a fare forecasting feature.
What this thing does is predict flight prices using a sophisticated machine learning algorithm. It helps travelers make buying decisions: Should they buy their tickets now or later? But, wait… why is it necessary to predict flight prices in the first place?
And why do they constantly change? American Airlines, the largest US carrier, can change about half a million prices every day. And sometimes, the prices for the same service class, on the same flight, can rise or fall multiple times within several days.
So, what’s going on? Obviously, airlines try to maximize their profits, just like any other business does. This means that carriers have to sell as many seats as possible for a flight and - at the same time - for the maximum price.
It’s inefficient if an aircraft takes off almost empty with a couple of expensive tickets sold. And it’s also inefficient to occupy the entire aircraft, but for the lowest cost per seat. That’s why airline revenue managers engage in a balancing act between finding the highest price and filling as many seats as possible.
To reach this balance, airlines must first understand their customers. Carriers track traveler purchasing behavior assuming that they are divided into two major groups: leisure and business travelers. And it looks pretty straightforward.
Leisure travelers are sensitive to prices and plan their trips in advance, sometimes months before departure. So, you have to sell cheaper tickets to leisure travelers and do it earlier. And, this will help you fill more seats.
Business travelers, on the other hand, need more flexibility and usually book days before departure. You’ll have to increase the prices as the departure date approaches to efficiently capture the business segment of your customers. As the fare is employer funded, business travelers don’t care as much about cost.
Not a very precise approximation of the customers, right? But if you look at the general, smoothed-out trend of any air fare, you’ll see this logic in action. The fare remains almost the same for months before a departure and then it starts growing in multiple, distinct steps.
These steps are caused by advance purchase discount requirements. If you fail to purchase a flight at least, say, two weeks in advance, the minimum fare remaining will get considerably higher for the same service class, whether it’s economy, business, or first class. Usually, a flight has multiple step increases as fares rise towards the departure date.
This difference between fares helps divide leisure travelers, who spend less, and business travelers, ready to pay more. However, the prices also dynamically change regardless of discount requirements, as airlines utilize another strategy that adds more peaks and valleys to this graph. Imagine there are fifty economy class seats on the aircraft for a given flight.
Even though these seats belong to the same class, an airline doesn’t want to sell all of them at the lowest cost, nor at the highest cost. Carriers divide all those fifty seats into multiple fare groups, or buckets. For instance, there will only be five seats at the lowest fare with minimum services and smallest bag allowances.
Once these five seats are sold out, the fare bucket is closed. You can no longer purchase a seat at that bargain basement price. But forty-five seats with more services and bag allowances are still available at a higher cost in other buckets.
They start filling up and then gradually close. So, the prices go up no matter what, as travelers purchase seats from higher-fare buckets until the aircraft is fully packed, right? Nope.
Sometimes, the prices do go down. If the only price direction was up as buckets filled, this would mean a lot of lost price-sensitive customers for airlines. And there wouldn’t be the need for sophisticated machine learning algorithms to predict fares in the first place.
Travelers would just have to purchase flights as early as possible. Besides gradually increasing prices, airlines track demand. How fast do those buckets get filled?
Imagine, you’ve sold five seats from the first bucket in a week, but then only a single traveler purchased the seat from the second bucket next week. If fewer people start buying out seats after a bucket or two are closed, the lower-fare buckets may be opened again to invite more price-sensitive customers. Some of the available seats from higher-priced buckets would then move to the re-opened lower bucket.
And this works in the other direction as well. If the aircraft starts filling up too fast, the airline may close low-fare buckets to get more revenue or even to prevent some people from buying these seats. Because there must be some room available for business travelers who will purchase their flights at a much higher cost right before departure.
Advance purchase discounts and fare-bucket motion are the two main drivers that complement each other and cause most of the price changes at airlines. And because demand dictates pricing logic, it may seem opaque to the average traveler, since only the airline revenue manager knows what’s going on with current demand and how fast those buckets fill up. But there are even more factors that impact dynamically changing prices.
The fares within buckets can also change in reaction to various external conditions. If the cost of fuel increases, this may cause the entire base fare to rise as well. On top of that, airlines consider seasonal trends.
If there are more people who fly for summer vacation to some destinations, the set of fares will be adjusted to this trend. This also happens if demand is likely experience an interval increase at a specific destination. For example, Super Bowl or a rock festival will trigger revenue managers to manually increase fares as higher demand is anticipated for these dates.
And, finally, if some low-cost airline or other competitor opens a new flight, the competing flights will get cheaper, even at traditional airlines. This is how dynamic pricing at airlines works today. But things may change in the near future.
The current pricing strategies that airlines use are based on broad rules. The simple idea behind these rules is to sell cheaper tickets to leisure, price-sensitive travelers and sell more expensive tickets to business travelers that don’t care that much about prices. This straightforward logic has worked for years.
With the arrival of new technologies, the old method started aging fast. The past understanding of one traveler being price-sensitive, while the other is not is very limited. There are many more nuances, and airlines realize that.
But the distribution model in which travelers purchase tickets from travel agencies puts blinders on carriers, leaving them guessing exactly what their customer looks like. They mostly judge by demand and time of purchase. On the other hand, by directly interacting with customers, carriers can get a more granular and detailed view of the actual person looking for a flight: What other flights are they looking for?
How often do they check prices? Which links do they click on? If airlines managed to tap into this data, they could use more advanced AI systems that define fully personalized prices.
Since 2012, airlines have been slowly embarking on a new data exchange standard: new distribution capability or NDC. It will allow airlines to receive more personal and detailed data about their customers and eventually tweak existing revenue strategies. This may render such tools as Fareboom’s price predictor obsolete, as fares will be adapted to each individual traveler.
The question is, are travelers themselves ready for such a change?
