• by Steve Smith , Staff Writer, November 13, 2009

When I go to Amazon.com -- oh, about five times a week (yes, I am that addicted) -- the site's famously effective recommendation engine has ample opportunity to monitor my browsing habits. The result is a personalized experience that is almost eerie in its ability to anticipate the items I would like to buy. But how do other retailers who see an online visitor once, or perhaps just once a season, create a similarly intimate experience? You crowd-source it. But in the process, if you look carefully enough at the range of interactions people make with retailer sites, you will uncover some unexpected patterns.

 

At outdoor apparel and equipment retailer Sun & Ski Sports, even the loyal customers visit the site one to three times a year. Generally skiers or campers buy their needed seasonal equipment all at once, says Director of E-Commerce Scott Blair. "It takes a long time to build a profile on that."

Instead of using a behavioral tracking system that targeted an individual's previous online habits, the company partnered with recommendation engine provider Baynote to aggregate the browsing habits of all visitors at the site to predict the tastes and needs of individuals entering the store. The recommendations can be fully automated or also manually tweaked. A huge closeout promotion on an item like a snowboard can be pinned to the recommendations so it shows up more frequently in the list. But eventually as the promotion works and people browse or buy the item, it floats into the natural recommendations all by itself.

By using what Baynote likes to call the "wisdom of the crowd" to extrapolate behaviors for all visitors, the sales lifts can be dramatic. "In the last months, customers using the recommendation engine had a 49% increase in conversion rates," says Blair, and these customers have a 20% increase in average order value. With about half of the people coming to the site using the recommendations, "that tells us that half of our customers haven't made up their minds, so the recommendations are helping convert them to sales."

By tying the recommendation engine to onsite and inbound search terms, Sun & Ski can also leverage searching behaviors to manage inventory. "We can track searches into the Web site and bounce rates, and from those we can see what [products] we don't have and can take those to our buyers," says Blair. "We can see what sales we are missing, and we have been able to hit a couple of home runs with that."

Another twist on recommendation data and search terms is the predictive data it produces. Blair says that the recommendation engine lets him see items that start popping up early in the season that people may be considering but not buying yet, and he can order more inventory against that emerging trend.

According to Jack Jia, CEO Baynote, predictive analysis is one of the surprise side benefits of crowd-sourced recommendations. Watching people browse specific inventory items even when they aren't buying them can lead to counterintuitive but accurate predictions.

He recounts running the engine with a major appliances retailer. Traditionally people choose white washing machines, but the recommendations engine, which was tracking many different behaviors (time spent, mouse movements, scrolling, etc.) was driving red washers into the top of the recommendations. "They weren't selling many of them, but this was the color many people were engaged with," says Jia. Lo and behold, three months later red washing machines started selling.

Interactivity is a subtle thing. Jia likens it to trying on clothes rather than buying, but in this case the interaction anticipates larger group shifts, not just one person doing a lot of window shopping. Even in aggregate, we toy with ideas and possibilities long before we commit. Jia notes, "You have to follow the crowd and dynamically change product offerings even before the sales come."