MarketingSherpa recently published a mini case study on how wine.com has increased their average conversion to purchase rate and order size by overhauling the product recommendations found throughout the shopping process for its visitors. I led marketing for Wine.com in 2000, and we were doing similar personalizations with our email marketing program at the time. I'm surprised it took them this long to catch up on the site - especially with all of the recommendation engine providers out there. But then again, we continue to hear stories about the lack of personalization in product recommendations that even successful ecommerce brands are doing, so I guess it's not a surprise that wine.com is only now catching on? (Disclaimer: I worked for Wine.com, the former Virtual Vineyards. The current Wine.com evolved out of eVineyard.com in 2001).
Read the full text here.
I also spent a few years marketing for Smith & Hawken,and during that time we outsourced a recommendation engine for our ecommerce site - providing the product recommendations that you commonly see down the right hand side bar of product pages. The results showing in these sidebars were products that other consumers had purchased in their basket along with the item showing. It made sense to follow the community recommendation path: "Customers who bought this, also bought this". There was one main problem with this engine: Customers don't always purchase in "rational" pairs (i.e. Candleholders with a candle, two books by the same author). The results were that our sidebars would therefore show "irrational" results. Customers viewing a candleholder would be shown a $500 teak bench, or potters soil. We quickly abandoned this vendor - finding that a more manually handled, intuitive product recommendation, was more successful in upselling the customer and increasing their cart size. But even this was only a subjective recommendation - and one which didn't take into consideration the needs, likes or past shopping or browsing behaviors of the individual customer.
Wine.com is incorporating both community-led recommendations, along with product and geographical profiles for their merchandise (wine flavor attributes, wine producing regions, as well as which states the specific SKU is available to distribute into), to provide these recommendations to their consumers. The recommendations are provided in the following format:
o Customers who viewed this item also viewed...
o Customers who bought this item also bought...
o Products ultimately purchased by customers who viewed this item...
o Customers who searched for "xxxx" ultimately bought...
o Top sellers in this price range...
o Items related to search "xxxx"...
But none of these yet speak to true PERSONALIZATION. The first 4 recommendations are based on other customer behaviors. the 5th based on a price range filter, and the last may likely be the most relevant, if they are leveraging robust product attributes to create the recommendation. However, they are not identifying for the specific customers which products would be most relevant based on THEIR interests, taste preferences, purchase behaviors and needs. The opportunity lies in associating not only very robust product attributes (for wine specifically, flavor, aroma, regional, varietal, etc. identifiers), with personal interests and tastes identified through a variety of means: preferences stated, purchasing data, personal reviews given, browsing behavior and relevancy scoring. The goal is to make a product recommendation be RELEVANT to their preferences and needs, MEANINGFUL in the context of their current needs, and to their individual profile, and TIMELY, delivering a recommendation which makes sense right now.
NetFlix has developed perhaps the most successful recommendation engine on the market in terms of relevancy. In fact, they ran a contest with a $1M prize for anyone who could produce a recommendation engine that could best theirs. As far as I know, no one has won yet. However, NetFlix does not have an engine that delivers on the MEANINGFUL and TIMELY part of the equation.
Of course, experts like Amazon have perfected the product attribute equation. The Smith & Hawken products we distributed through Amazon.com were associated with literally hundreds of attributes - but primarily to ensure that customers found what they were looking for in their search algorithm. They have also mastered the community-led recommendations and product review-based recommendations. But again, do they truly deliver a RELEVANT, MEANINGFUL and TIMELY recommendation for me?
Think about it. Who's nailed it yet? Anyone? Would love to hear your thoughts.
Read the full text here.
I also spent a few years marketing for Smith & Hawken,and during that time we outsourced a recommendation engine for our ecommerce site - providing the product recommendations that you commonly see down the right hand side bar of product pages. The results showing in these sidebars were products that other consumers had purchased in their basket along with the item showing. It made sense to follow the community recommendation path: "Customers who bought this, also bought this". There was one main problem with this engine: Customers don't always purchase in "rational" pairs (i.e. Candleholders with a candle, two books by the same author). The results were that our sidebars would therefore show "irrational" results. Customers viewing a candleholder would be shown a $500 teak bench, or potters soil. We quickly abandoned this vendor - finding that a more manually handled, intuitive product recommendation, was more successful in upselling the customer and increasing their cart size. But even this was only a subjective recommendation - and one which didn't take into consideration the needs, likes or past shopping or browsing behaviors of the individual customer.
Wine.com is incorporating both community-led recommendations, along with product and geographical profiles for their merchandise (wine flavor attributes, wine producing regions, as well as which states the specific SKU is available to distribute into), to provide these recommendations to their consumers. The recommendations are provided in the following format:
o Customers who viewed this item also viewed...
o Customers who bought this item also bought...
o Products ultimately purchased by customers who viewed this item...
o Customers who searched for "xxxx" ultimately bought...
o Top sellers in this price range...
o Items related to search "xxxx"...
But none of these yet speak to true PERSONALIZATION. The first 4 recommendations are based on other customer behaviors. the 5th based on a price range filter, and the last may likely be the most relevant, if they are leveraging robust product attributes to create the recommendation. However, they are not identifying for the specific customers which products would be most relevant based on THEIR interests, taste preferences, purchase behaviors and needs. The opportunity lies in associating not only very robust product attributes (for wine specifically, flavor, aroma, regional, varietal, etc. identifiers), with personal interests and tastes identified through a variety of means: preferences stated, purchasing data, personal reviews given, browsing behavior and relevancy scoring. The goal is to make a product recommendation be RELEVANT to their preferences and needs, MEANINGFUL in the context of their current needs, and to their individual profile, and TIMELY, delivering a recommendation which makes sense right now.
NetFlix has developed perhaps the most successful recommendation engine on the market in terms of relevancy. In fact, they ran a contest with a $1M prize for anyone who could produce a recommendation engine that could best theirs. As far as I know, no one has won yet. However, NetFlix does not have an engine that delivers on the MEANINGFUL and TIMELY part of the equation.
Of course, experts like Amazon have perfected the product attribute equation. The Smith & Hawken products we distributed through Amazon.com were associated with literally hundreds of attributes - but primarily to ensure that customers found what they were looking for in their search algorithm. They have also mastered the community-led recommendations and product review-based recommendations. But again, do they truly deliver a RELEVANT, MEANINGFUL and TIMELY recommendation for me?
Think about it. Who's nailed it yet? Anyone? Would love to hear your thoughts.
RSS Feed