Under the Hood: Improving Flight Search at Expedia

By John Kim and Aman Bhutani

This content originally appeared on Expedia’s Science of Travel Blog.

A closer inspection of our travel engines will reveal the millions of calculations and computations being made every millisecond so travelers everywhere can quickly and easily book a great vacation. It is here that we’ll share more about how we are developing smarter solutions and recommendations to guide our customers to the trip they want to book. Welcome.

Most people begin planning their trips with flights. We’ll start there too.

A typical customer searches for a flight up to an average of 48 times before booking. Here at Expedia, we strive to bring that number down by surfacing the most relevant results that can include alternative surrounding airports, day, or time of travel.

For a given round-trip flight search between two destinations, there can be billions of flight combinations, even when only looking at a single departure and return date. Expedia whittles that down, returning the best sixteen hundred results in the form of 40 departing and 40 return flights.

Yet as every traveler knows, each search begs more questions than it answers. Is the best flight for you flying out of different airports? Would adjusting the length of your stay or departure day save you any money?

Take a basic flight search: Seattle to Miami for a five to seven-day trip. To find the cheapest round-trip fare in a 30-day period, you could conduct 3 searches (a search for a 5-, 6- or 7-day trip) for each day in that period, which would quickly get you to 90 searches.

Now, imagine you had the inside knowledge that traveling to Fort Lauderdale is a great alternative to Miami, suddenly you find yourself having to do another 90 searches to find the best prices. Now combine those airports and you have 180 combinations. Adding other departure or arrival airports such as Portland or Palm Beach pushes the number over 500.

With an average search taking at least a minute, doing all this research would take you hours. With thousands of price updates taking place every hour, you’ll have to start again by the time you finish since the prices and availability will have changed. It’s no wonder why people end up searching 48 times!

Meet Expedia’s New Flight Recommendations

By leveraging the over three billion flight searches performed worldwide on Expedia, Flight Recommendations looks for patterns to help travelers with their next search.

In fact, we see our users follow very distinct habits. To start, travelers look to depart and return on specific days, such as leaving on Sunday and returning on Friday. They will then search leaving Saturday or returning on Thursday, all in hopes of finding a more convenient or cheaper fare. Combine this with searches for alternate airports and suddenly travelers find themselves overwhelmed with all the possibilities.

By following the patterns of how our travelers’ search, then booking and finding alternative airports/dates in those searches, Flight Recommendations surfaces the best options. We look for similar deltas for departure and arrival days and surface them as search recommendations. Our goal is to make it easy to see all the best options at a glance so travelers can choose what’s right for them whether it’s price or convenience.

Recommending alternate flights to travelers isn’t new. You’ve probably seen this in the form of fare calendars. However, this system doesn’t help travelers with anything beyond a snapshot of a saved price. Our team is looking at real purchasing patterns, of real travelers at scale so we can help you find desirable, bookable flights for less.

With each recommendation, we track the effectiveness, value, and accuracy of the prediction against real booking information and adjust them to best benefit our customers. These adjustments could take the form of different data, adjusted thresholds, or additional factors. The algorithm is always evolving and is the backbone of the recommendations feature.

Going on vacation is fun. Planning a vacation, however, is a multivariate optimization problem that some of the smartest engineering minds are working on. Together, we can advance the science of travel to help consumers save time and money. The opportunities we have to identify patterns with our data are endless and we are just barely scratching the surface. Stay tuned!