advertisement Offers $1 Million For Improved Recommendations

The online retailer is pulling a Netflix, dangling the promise of a rich reward–not to mention some serious bragging rights–to the team that increases customer purchases.


advertisement is about to pull a Netflix. The online retailer known for offering high-quality goods at discount prices will hand out $1 million early next year to the team that can improve its recommendation engine by 10 percent or more.

That probably sounds familiar. Netflix held a similar competition two years ago, offering the same amount of cash to the team that could improve the predictions of how a particular user would rate a particular film, by 10% accuracy or more. Not one, but two teams were able to pull off that feat in the closely followed competition. Overstock is hoping for similar results, given the increasing importance recommendations–and personalization in general–are playing in the world of online commerce.

“[Consumers today] understand that a website isn’t just a dead screen blasting stuff at them,” Overstock CEO Patrick Byrn tells Fast Company. “[They know] it’s learning about them, their likes and dislikes, and optimizing to them as an individual.”

“The absence of a good recommendation system is a liability,” he adds. “People have come to expect it of a professional site.”

As with the Netflix Prize, teams working on the Overstock competition will be given a dataset–in this case, synthetic–to use for refining their algorithms. But unlike with the previous event, Overstock will judge entrants by running their solutions against live data–actual customers visiting the Overstock site–and the algorithms will be evaluated by how much lift they create in real-world purchases.

To run the competition, Overstock is teaming up with RichRevelance, the e-commerce recommendation vendor which was cofounded by former Amazon rock stars who worked in this space there (and one of whom also once worked for Overstock).


The competition starts today and is open to anyone or any team that wants to compete. Algorithms must be submitted by Dec. 1. Finalists will be announced next year on February 12. And the winning team will be announced March 27.

In a nod to the importance that academia plays in helping to fostering innovative thinking, if the winning team is affiliated with an academic institution, the competition will award additional prize money, up to $250,000, to that institution.

If you think all the recommendation geniuses out there burnt out on the Netflix Prize, think again, says RichRelevance Chief Scientist Darren Vengroff. He tells Fast Company that the thing researchers always ask him about at conferences and other such gatherings is how they can get their hands on real-world data. Running their theories against made-up data is ultimately unsatisfying. They can come up with all sorts of ideas about what might work, but they never are able to find out–for sure–if they’re right, unless they can set their algorithms loose on actual user data.

With the Overstock competition–officially called the RecLab Prize, named for the data lab RichRelevance set up for researchers–competitors still won’t get their ownhands on actual data. There are too many privacy issues involved–even with anonymized data, which can sometimes be reverse-engineered to identify actual people–to turn over actual data.

But the competition will give researchers the next best thing–and in some ways, perhaps, an even better thing. The algorithms they submit will be run against actual, real-time shopping–in a system controlled by Overstock and RichRelevance–and they’ll be judged on how much the boost actual purchases. For researchers, that’s like heading straight to the Super Bowl.

Overstock’s Byrne has high hopes for the competiton. “I’m a big believer in the wisdom of crowds,” he says. “A lot of good ideas are out at the periphery.”


About the author

E.B. Boyd (@ebboyd) has holed up in conference rooms with pioneers in Silicon Valley and hunkered down in bunkers with soldiers in Afghanistan