August 08, 2010
According to Mashable; How Online Retailers Can Leverage Facebook’s Open Graph
Formerly a Principal Engineer at Amazon, Darren Vengroff is Chief Scientist at RichRelevance where he helps retailers like Overstock and Sears create a more personal shopping experience for consumers. You can read more from Darren on the RichRelevance blog.
Amazon () and Facebook () are making headlines with the launch of a new application that allows shoppers to receive product recommendations based on Facebook preferences. Once users enable this app, Amazon is able to monitor their activity on Facebook, including what pages they like, and use that information to recommend products they may be interested in purchasing. Combining accounts with an application like this, whether specific to Amazon or other merchants, has the potential to be a compelling hybrid of social networking and shopping that creates value for shoppers and merchants.
While Amazon’s move made headlines because of their market position, the fact is that any merchant can build an app to allow Facebook users to share their interests. Collecting this data is the easy part. Leveraging it appropriately is where the real challenge lies. Ultimately, the success of the recommendations driven by these apps will be predicated on how relevance is extracted — particularly from the social graph — and how recommendations are presented to shoppers.
If You “Like” It, You Might Want to Buy It
Once data has been collected via the Facebook app, the first thing a merchant’s recommendation system has to do is identify the relationship between a page that someone has “Liked” and one or more products. This can either be done purely via behavior, or via some combination of behavior and page and product attributes. For example, in analyzing the likes and purchase patterns of shoppers, we might find that people who like the Facebook page of restaurateur Danny Meyer had a propensity to purchase his book Setting the Table. As a result, the system can now identify a relationship between the page ID of Danny Meyer’s Facebook page and the ISBN of his book on the merchant’s site.
This seemingly small piece of information is invaluable. Not only does it suggest there is some value in recommending the book to people who have “Liked” the page, it has implications for our ability to construct future recommendations. In this case, because the Facebook page is specifically about Danny Meyer, who also happens to be the author of the purchased book, we now have strong evidence that when he publishes another book, a great marketing tactic upon release would be to recommend it to fans of his Facebook Page.
Tracking a purchase against a “Liked” page is just one type of relationship that can be identified. Similarly, other product attributes like stars (for films), age range (for toys), or brand (for apparel or consumer electronics) can help guide the system from a “Liked” page to related products. When carefully managed, combinations of these types of product attributes, along with raw user behavior, are likely to generate the best recommendations.
“Liking” Doesn’t Always Lead to Buying
The second key part of a successful recommendation system built on top of a social network like Facebook is a keen understanding of the way in which certain shopping behaviors — from searching, to browsing items, to adding them to a cart — are correlated with becoming a fan or “Liking” a particular page. I hinted at this above with the example of looking at the purchase behavior of people who liked Danny Meyer’s Facebook page. Recommendation systems are commonly built to understand the relationship between a previous purchase and a likely purchase, or browsing and purchasing.
The dynamics of “Liking” a Facebook page are quite a bit different, and less costly, than purchases. While for some pages, “Likes” indicate purchase propensity, there are many other pages for which “Likes” tend not to indicate purchase propensity. For example, a “Like” can be purely aspirational, as in the case of a teenager who “Likes” the Facebook Page for the Porsche 911. Attempting to sell him a custom made cover for a 911 is probably not a great idea. Recommending a Porsche Logo T-Shirt, on the other hand, might be more relevant.
The reality is that the implications of a “Like” are widely varied, even if we’ve identified the specific product associated with the Facebook page and have a fair bit of additional metadata — car make and model, in the case of the sports car example.
Spanning the Social Network
If we move from Pages () and individual “Likes” to Pages his or her friends “Like,” things can get even more complicated. A twenty-something female Facebook user might have hundreds of friends who have “Liked” hundreds of films between them. But, if she were really looking for advice on a film to purchase, she would probably most heavily trust only a few close friends. She might be close to her mother and boyfriend and yet have no interest in films they like, even though she might be interested in their feedback on other products ranging from shoes to microwave ovens to sunscreen.
So in addition to seeing how likes of a group of friends affect shopping behavior, we have to be very careful and take into account the different kinds of influence various individuals have.
Bringing it All Together
All of the recommendation opportunities and all of the associated challenges outlined above are available to any merchant who chooses to build a Facebook app and encourages their customers to use it. As with recommendations based solely on shopping behavior, merchants can and should present a multiplicity of different recommendation strategies to the shopper based on both their shopping and their social network behavior. Ideally, this should happen both when shoppers are browsing a merchant’s website and when they are using the merchant’s app on Facebook.
The final piece of the puzzle is then a real-time optimization system that monitors how various recommendation strategies are performing in different contexts — whether on a merchant’s site, or in various locations on Facebook — and chooses the most relevant content for each and every shopper at every moment of their experience.