For years, behavioral marketers have been engaged in relentless, often obsessive pursuit of data about individuals -- where they've been, and what they've done online. They want to reach that mythic goal of "one-to-one" targeting. All that data, however, argues Mike Svatek, marketing director and product manager of Baynote, may be blinding marketers to a more fundamental fact: It is not behavior in isolation, but rather the charting of similarities and affinities of intent and interest, that ultimately matters.
BI: The development of your company's technology has received a lot of attention for pioneering the notion of predictively modeling behavior based on the "wisdom of crowds". How has this approach been deployed and applied in the real world so far? What are its applications to advertising?
Svatek: The goal is to avoid advertising if you define advertising as just trying to sell people things they don't want. The pitfall of behavioral targeting has been that it's tried to impose interpretations on individual's behavior by employing artificial rules that don't adequately encompass the questions of interest and intent. If Web browser A goes to content B, it means they will be likely to do this or that and thus, you as a marketer must automatically do C. Sometimes, you may be right. But if that's as deep as your model goes, you're fated to fall into the e-commerce cliché of eternally delivering jewelry ads to the male who bought a necklace for his niece's graduation but is really a gadget freak. Or to be hopelessly confused by the mom who shops on Monday for her teenage sons, on Tuesday for her husband and college-age daughter and on Friday for herself. The model just short circuits.
BI: What's the relationship between what you call content guidance and marketing?
Svatek: Our core belief is in the efficacy of high-fidelity listening because the goal of content guidance is to reflect what people are actually interested in. So we make a distinction between recording things that are done, behavior in the abstract, and really looking at how people, as individuals and as groups, interact with content. We call it finding behavioral fingerprints. We try to pay close attention to the way people interact with the pages they visit. We look at how quickly or slowly they scroll across particular pages, what terms they highlight or right click to get more information about a given item, their link density. Other areas like click path and time spent that elucidate engagement are also increasingly important to us.
BI: Can you give some examples of how this works in customer segmentation?
Svatek: If you look at similarities in what and how people are interacting with specific content areas, you can begin to connect individuals and their behaviors to their communities. Once a Web site owner or an advertiser begins to look at site "traffic" that way, they no longer see a hundred thousand random individuals but discrete groups. Instead of sports site browsers, they'll see a community interested in golfing apparel, a community interested in tennis, a community of runners, a community of outdoor fanatics. They then look at the level of each community's presence on the site at any given time. They can see it growing or constricting at different times of the day or week, or in relation to events in the outer world, such as something a celebrity has done.
Svatek: Tracking group behavior on a Web site in this way allows a publisher or advertiser to change their product mix easily and proactively. It also gives them the intelligence to make more relevant recommendations, based not only on how a specific consumer's behavior evolves, but on how what they do relates to their affinity communities, of which they may belong to several.
BI: Any impact studies on ROI?
Svatek: Combining community-based targeting with recommendations has given several retail clients a sales conversion lift of well over 20%. What's amazing about this is we're talking about clients such as e-tailers, which have spent enormous time optimizing their sites. Given that, we were figuring on 3% to 4% as an ambitious goal.
BI: How long does it take to generate usable profiles?
Svatek: It doesn't take a lot of time in most cases to have the system finding deep patterns of interaction. If you look at Ebay where there are literally billions of data points processed in a few hours, it's like drinking from a fire hose. Even very small sites which have visitors in the tens of thousands can establish what we call heuristic communities in a few days.
BI: Where do you see the biggest challenges for enhancing targeting going forward?
Svatek: We see the next iteration as addressing a challenge that's getting more and more daunting to marketers: How to integrate all the incredible quantities of different consumer-related data they're being bombarded with and bombarding themselves with. How do you bring all that data to bear and start interpreting what it means? The industry started out trying to learn everything we could about our customers, and that's good. But the next challenge, one that's barely begun, is finding out how to know what we know.