Kayak adds price forecasts to US and UK fare search, saying it’s better than Bing Travel
Today US metasearch site Kayak is introducing a price forecast tool for its US and UK users.
Kayak’s price predictor is similar to the airfare prediction service pioneered in 2005 by Farecast, which was later acquired by Microsoft and re-branded by Bing Travel as the Bing Travel Price Predictor.
Like Bing’s tool, Kayak’s price forecast tool is integrated into its main search interface. When users run a search, a box pops up with a prediction of whether airfares for the query will rise or fall in the next week.
But in an improvement Bing’s tool, Kayak offers the service for both its US and UK users, with plans to roll out the service to users of its German and other European sites soon.
Like Bing’s tool, Kayak’s airfare forecast includes a prediction of whether the prices for your search query will rise or fall in the next week. The seven-day cap on predictions is a starting point, says Giorgos Zacharia, and a broader and adjustable prediction period may be in the works.
In an improvement on Bing’s tool, Kayak’s price prediction service can produce a forecast for a search between any two city-pairs worldwide when it has enough data. In contrast, Bing limits its predictions to primarily major US domestic routes and doesn’t offer forecasts for, say, itineraries between two non-US airports.
Better than Bing
Kayak’s price predictor is based on more than one billion queries performed on Kayak websites, which is “an enormous volume of search queries,” says chief marketing officer Rob Birge.
“The volume of data used by Kayak to calculate the predictions is dramatically greater than what other prediction tools are using out there in the market.”
Interestingly, Farecast used to claim it used “a historical database of more than 175 billion airfare prices from most of the major airlines” for its analysis.
Zacharia scoffed at the relevance of much of that data. He says there’s little relevance in historical airfares, such as ones before 2010, in Kayak’s statistical model, which draws on an on-going stream of real-time and cached search queries.
Tnooz asked Kayak about its data set. In October, the company had revealed on its blog that it had analyzed a data set compiled from an analysis of all searches for airfare between January 2011 and December 2011.
That may have been the core data Kayak used to create predictions and test their validity, though the company says that its predictions are based on an ever-changing dynamic data set and that it plans to improve its calculations constantly to help make its advice more accurate.
Big Data on the march
In an improvement on Bing’s tool, Kayak’s price forecast provides an indication of how much faith it has in any given forecast. See example:
The original Farecast predictions came with a similar statement that the company’s predictions had been independently audited by consulting firm Navigant as being accurate 75% of the time on average.
Kayak has not had its predictions independently audited. It doesn’t reveal details about its prediction algorithm, such as its statistical confidence levels, and it would take some serious computing power to independently verify the company’s claims of accuracy.
When Kayak webpages describe their prediction in the rise/drop in fares in terms with a percentage “confidence,” it’s not referring to the proper statistical statistical measure confidence interval as a basis. It’s marketing-speak.
One would assume the brand would have faith in its predictions not upsetting users or else it wouldn’t have rolled the feature out. On the other hand, Bing Travel seems to have let the quality of its predictions slide and hasn’t had any independent audits and no longer offers users an opportunity to buy insurance on its price changes based on its predictions, as Farecast once offered.
Predictions sometimes appear in Kayak searches and sometimes don’t. It depends on whether or not the data is “deep” enough that Kayak’s algorithm feels it can give sound advice.
It recently beta tested the service with randomly selected users.
Google Flights already offers bar-chart histories of average fares for particular routes. It would presumably not take much additional effort to add price forecasts.
Are hotels next?
One thing that Farecast experimented with and that Bing Travel stayed away from is rate predictions for major properties in major US cities.
Rather than predict rates, Farecast told users if the rate is a good value when compared with the hotel’s rate history and other factors. Google has picked up this concept in its Hotel Finder, which has a similar feature.
Would anyone like to make a forecast about whether Kayak will expand its prediction service to hotel rates?