Wine.com

  • RETAIL SEGMENT: Food & Beverage
  • SOLUTION: Recommend™
  • CHALLENGE: Wine.com wanted to quickly develop, test and measure innovative new recommendation strategies that leveraged realtime access to customer data— without having to build out the I.T. architecture required to do so.
  • RESULTS: Using RichRelevance Recommend™, Wine.com tested a “similar products” strategy that drove $5 per click— becoming one of their best strategies in terms of revenue per click.

«We now have the ability to come up with algorithms on our own, create new placements and implement strategies immediately.» — Cam Fortin Sr. Director of Product Development Wine.com

Headquartered in San Francisco, between California’s wine country and Silicon Valley, Wine.com’s mission is to promote the wine lifestyle through innovation— using technology to bring the world of wine to its shoppers’ fingertips.

As the Sr. Director of Product Development, Cam Fortin is tasked with transforming the shopping experience for a million-bottle online wine shop through the most relevant information, tools and expert advice that a wine connoisseur might seek.

Having partnered with RichRelevance since 2010, Wine.com was well-versed with how behavioral recommendations could aid consumer research and inform purchases. Many of Wine.com’s behavioral recommendation strategies are product-focused, but a few are attribute-based—a critical differentiator when considering the complex product categories associated with wine.

The multi-attribute world of wine

“Wine is different. People buy the same wine multiple times. Or, if they’re interested in one wine, they’re frequently interested in other wines that are similar, so attribute-based strategies work very well,” say Fortin.

Obvious attributes such as “French,” “red,” and “2010” are easy to deploy in a recommendation strategy such as “People who purchased a French Bordeaux also viewed.” But Fortin knew there was a massive opportunity to be leveraged when considering the universe of attributes that applies to wine

“We’ve always wanted to explore recommending similar products based on the number of attributes they have in common,” he says. Wines may share up to 30 different attributes (big red, smoky, tannic, etc.) in common. Exploring how this subset of common attributes could be exploited for personalization—and capitalized upon—was a compelling challenge. “Being able to run SQL queries and have a ridiculously huge machine to access information from our site directly was something we were very interested in.”

From concept to deployment to production in hours

Through RichRelevance Recommend™, Wine.com was able to test its theory on a custom strategy that leveraged the intersection of multiple attributes. Specifically, Wine.com could access all elements of its customer “map” (browse and purchase history, loyalty, preferences, etc.) and build the algorithm it wished to test, utilizing RichRelevance’s Hadoop instance. Further, Wine.com could measure how its specific algorithm competed against the existing set of pre-built RichRelevance algorithms.

The speed with which Fortin moved from concept to deployment to production was unprecedented. “Instead of requesting a change in placement or a tweak in a strategy, we now had the ability to come up with algorithms on our own, create new placements and implement strategies immediately,” says Fortin.

Similarity spells success

The new “similar products” recommendation strategy—which ranked product recommendations based on the number of attributes they have in common—became one of the best-performing strategies in terms of revenue per click, generating about $5 per click. “We were excited that it worked right off the bat, even though it was just a rough algorithm with little polish to it. Next, we want to weight attributes differently and continue refining this strategy,” says Fortin.

For Fortin, the larger success is related to the ease of use and the speed associated with developing and testing algorithms. The ability for any of his SQL programmers to develop and test several algorithms a month with minimal IT investment was a huge win made possible through Recommend.

DOWNLOAD
Share :