Farecast

» see farecast.com as it lived, on the wayback machine

~ the obit ~

It answered one question and it answered it with a number. You put in a route and a date, and Farecast told you whether the fare was likely to go up or down over the next week, with a confidence figure attached so you could decide how much to trust it. Buy now, or wait. Nothing else on a travel site in 2006 was willing to tell you that, because telling you to wait means telling you not to buy anything today.

Oren Etzioni started the company in Seattle in 2003, out of his own work as a computer science professor at the University of Washington. The idea came from watching the person next to him on a flight pay less for the same seat. The method was to collect airfare observations at scale over years and find the patterns in how a given route's prices moved, which is unglamorous and turned out to work.

The public version launched on 15 May 2007 and Microsoft bought the company in April 2008 for about $115 million, folding it into Live Search and then into Bing Travel when Bing arrived in June 2009. For a stretch it was the single most distinctive thing Bing had. Google could match Microsoft on results and could not match this, because this was a proprietary model trained on data Microsoft had paid to accumulate.

The distance in this one is not about access. Farecast sat on the open web and loading it from Ankara took the same three seconds it took anywhere else. It simply had nothing to say to me. A prediction model is a property of the market it was trained on, and the market here was decades of published United States domestic fare data on routes flown by half a dozen competing carriers. Turkish domestic aviation only opened to real competition after 2003 and no comparable history existed anywhere for a model to learn from. The product was never unavailable to me. It was untrained on every question I had.

~ the rap sheet ~

~ where the numbers came from ~

Farecast is one of the earliest consumer products that would now be described as machine learning, built and shipped several years before anybody was selling that phrase. GeekWire's line about it is the accurate one: it was among the first understandable consumer applications of large-scale data analysis. Understandable is the operative word. The output was two options and a percentage, and no part of using it required you to know or care what was underneath.

Etzioni's own defense of the approach at the time was that computers are comfortable sifting through large volumes of numbers to find patterns, which was a thing people still needed to be told. The travel industry's objection was less technical than commercial. An airline does not want a website telling your customers to come back in nine days, and the agencies whose revenue came from bookings had no reason to build one.

~ one hundred and fifteen million ~

Microsoft's purchase in April 2008 was not a bad decision. Search was the fight of that decade and travel was one of the highest-value verticals inside it, so $115 million for the only differentiated travel feature on the market was a reasonable price from a company spending far more than that on the same fight elsewhere.

What happened next is the part Skift wrote up in 2014 under a headline about squandering. Google bought ITA Software in 2010, the flight data platform most of the industry's fare search ran through, and cleared antitrust in 2011. That put the supply of the raw material behind Microsoft's own feature into a competitor's hands. Etzioni's account is blunt about the consequence: Bing had been outflanked, and Bing did not want to pay Google for the data required to keep Farecast running.

~ two dates for one ending ~

There are two shutdown dates in the record and both of them are defensible. Wikipedia dates the removal of the fare prediction from Bing Travel to January 2014. The press caught up on 8 April 2014, when GeekWire ran the story and Microsoft gave a statement about focusing resources where it saw the greatest opportunity to serve travel needs, which is the sentence a company produces when there is no story it wants to tell.

The gap between the two is itself the interesting fact. A flagship feature of a major search product was switched off and it took roughly three months for anyone outside the company to write about it. April went into this grave because that is when Microsoft confirmed it, and the January date sits here for anybody who wants the earlier one.

~ what took its place ~

Fare prediction did not die, which makes Farecast a rare case of a good product being killed while the idea behind it went on to win. Google Flights carries a price graph and a tracking alert today, and Hopper built an entire company on the same premise and made it work on a phone. Both of them arrived after Microsoft had already decided the thing was not worth the data bill.

Etzioni went on to run the Allen Institute for Artificial Intelligence, and told a reporter in 2014 that he had switched to using Google. Bing Travel survived without the feature that had been its reason to exist, MSN Travel took over the branding in 2015, and the domain that started it all is a parked lander today. Nobody remembers Bing Travel. A surprising number of people still remember being told to wait.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »