From Personalization To Targeting: Finding The Right Ads For The Right People
The recommendation and personalization engines that have been working to make e-commerce an increasingly efficient channel over the years are moving into the ad ecosystem data game. This is a natural extension of what they do, of course. Many of these companies have so much richer, deeper data about a user than just what content they last visited that the theoretical value of these profiles could be enormous. Amazon, for instance, just announced a deal with demand side platform Triggit to leverage its customer data in off-site ad targeting. After dealing with Amazon for over a decade in hundreds of transactions, I imagine these guys know me better than my mother in a lot of respects. Their product recommendations are exceptional and have prompted me to buy scores of items I otherwise didn't even know existed. If they can't target a relevant ad to me elsewhere on the Web, I don't know who can.
Amazon is not the only one trying to turn the personalization model into an ad model. In classic behavioral targeting , he advertiser knows what kind of segment of people he is looking for based on what a group of users has done recently. "In a personalization setting, you know the user," says Eric Bosco, COO, Choicestream, which powers personalization and recommendations for AT&T, MTV, Zappos and Tesco, among others. "You have observed what they are doing on a site, and it is a finite list of things you want to present to them is content offers and picking the best thing for that users. Bosco says that by leveraging personalization data and that perspective that originates with the individual user, not the segment, they better understand not only what interests a person but also what engages them. "We are instrumented much more deeply than standard behavioral targeting. We know what people are putting in shopping carts, the items they use on and offline from recommendation partners."
Bosco, who worked for years on first-generation behavioral targeting engines at Ad.com where he was SVP of Operations, says this model turns the old one on its head, in that it is more about looking for the right ads for a person. ChoiceStream has just launched its Crunch targeting platform. Its design is not just to target segments but to anticipate how people will respond or engage with ads. The idea is to understand -- mainly through personalization, in combination with third-party data -- what audiences are most likely to deliver the desired marketer results. "In the Crunch model, you know the advertiser and the result the advertiser wants, and you know a whole set of users you have information about, so you are trying to pick the users most able to respond to the ad," he says. "It picks people for the ad."
Bosco says this is not just a distinction without a difference. Along with data partners like Exelate and AlmondNet, the system creates rich profiles of users it has assembled from the many recommendation engine clients who have agreed to share data. For the time being their efforts are focused on mid-funnel marketing goals involving customer acquisition.
For Dish Network, for instance, the client brings to a campaign the specific goal of getting people to the online shopping cart. TV services and all of their subscription options can be a complex sell that often requires a phone assist. But Dish does know that among those who do shop with them online, a high percentage convert once into the shopping cart. "We instrumented their page so that action are fed back to Crunch and we started learning about audiences and showed ad impressions to people," says Bosco. "You can start with historically [strong] points like 35-year-old females or southern regions that convert well. We start with that as an input, but the next step for Crunch is to go into learning phase and sample to see what types of people are going into the shopping cart and what is their multi-segment membership." The system will go through innumerable variables to understand the make-up of people most likely to behave the way the campaign wants. Does a person who falls into attributes A and B convert as well as someone who simply falls into one or the other, for instance. "It comes upon an initial profile of people that are being led into the shopping cart and starts looking for more of these people," Bosco says.
ChoiceStream has been running campaigns on the platform in beta since January for about a dozen advertisers in 20 deployments. For one client, whose aim it was to move people into shopping carts, the Crunch process delivered 3,100 new customers in 60 days. One of the promises of leveraging profiles rather than segments is that the user who gets delivered is simply more thoroughly pre-qualified to be receptive. In one campaign that drove new registrations, ChoiceStream claims these new acquisitions spent 35% more on follow-up purchases than average.
Bosco says that the dynamic learning and optimization of the model is especially important because unexpected changes in the consumer environment can alter the effectiveness of initial targeting strategies. "When we started one campaign, folks who had the Spanish language attribute were converting really well, but that dropped off, and folks with other properties like health and fitness started converting well." Techies stayed constant throughout, however. Crunch adjusted for the evolution in consumer environment.
The connection between recommendation/personalization engines and ad targeting is likely to get stronger in coming months. Amazon's entry into the market raises the profile of the model. In fact at the OMMA Behavioral show in San Francisco on July 20 we will be looking at the expanded use of these kinds of data. Executives from StumbleUpon, CNET, Pandora, ChoiceStream and Organic will be part of a panel "You Might Like this...The New Age of Discovery in Recommendation Nation."