Often, marketers employ traditional offline targeting, like geography and demographic targeting, as surrogates for identifying their target audiences. They'll target suburban moms with a credit card offer, because research or intuition suggests these are profitable prospects. However, behavioral targeting enables marketers to identify and target these prospects without ignoring customers such as suburban women without children or moms who live in cities.
Behavioral targeting technology can identify profitable customer segments without regard to the offline surrogates that correlate to profitable customers. Unlike contextual targeting, which focuses on the single page a visitor happens to be on at any given moment, behavioral targeting uses the recency and frequency of previous online experiences to identify which visitors are eligible for your campaign.
BT uses more data and can therefore be more accurate than other forms of targeting. But despite the popularity and power of it in theory, an array of companies pitching behavioral targeting solutions disappoint in practice. To understand the behavioral targeting landscape, it is worthwhile to review the four main approaches.
Retargeting is the simplest form of behavioral targeting. It enables marketers to target visitors who have performed a single activity (or a specific sequence of activities) on a marketer's website or with that marketer's creative. Since the qualifying event can be stored in a cookie, no investment in large scale data processing is required for a company to offer this approach. Accordingly, this approach is the most widespread behavioral targeting offering in the market.
The most successful uses of retargeting require access to a large advertising network. The ability to retarget requires the visitor to first interact with the marketer's brand and later be seen on the network, so those with less reach have less of a chance to see that visitor and thus deliver a campaign with any scale. These smaller networks typically buy traffic from larger networks, but since they are not embedded into the ad servers of the larger networks, their delivery is often less optimal than working directly with the large network.
The cluster approach assigns each visitor to one and only one segment. A visitor can either be a soccer mom or a gadget geek, but not both. Since all people have more than one interest and cannot be grouped into a single, predetermined clusters, this approach is simple to understand but is not likely to outperform other gross generalization targeting tactics. Database marketers such as Acxiom and Experian are best known for pioneering the cluster model, but Tacoda is the most popular behavioral targeting vendor to use this approach.
Custom business rules approach
The custom business rules approach offers marketers the maximum flexibility for defining which visitors belong in a segment. Marketers can target visitors who have done X events in Y days or Y events in X days. Unfortunately, that's the problem. How many events should a visitor perform to belong to the segment? The only way to find out is to test, and this approach requires continued testing because populations don't remain constant. Transparency of the inputs is a must for this type of targeting, because the burden is on the marketer to define the criteria that qualify each individual. Revenue Science is best known for popularizing the custom business rules approach.
Predictive attribute approach
The predictive attribute approach automates business rules to identify the ideal recency and frequency of activities that best correlate with brand and performance goals (e.g., brand recall, clicks and conversions). Thus, the system is continuously learning and adapting to changing behavior and is able to identify multiple interest attributes per visitor. Unlike the previous examples, this approach groups visitors based not on the similarity of their previous activities, but on their predicted future activities and likelihood to perform certain actions. Yahoo and ValueClick Media are the largest behavioral targeting providers to offer a predictive attribute approach.
A few other approaches are now included in the BT menagerie, but these tactics are not true behavioral targeting, since they no longer target the visitor's own behavior.
One method uses social network connections to identify prospects. When a prospect interacts with a marketer's brand, the marketer tries to communicate with all the people connected to that individual. This is similar to viral marketing in reverse.
Another approach is "look-alike modeling." Instead of targeting people who match the marketer's criteria, the marketer is offered an audience that looks like the people who match these criteria. While this enables publishers to expand their available inventory of "behavioral" targeting, it often performs poorly and it is easy to explain why.
Think of your neighbors and friends. Typically you live in the same geographic marketing area (e.g., ZIP code or city), if not on the same street, and you may even have very similar socio-demographic attributes. But among these neighbors and friends, how many drive the same model car as you do? What percent has the same stock portfolio? People are inherently different, and trying to categorize them based on gender, age or location is often no better than grouping people together by their eye color or shoe size.
Until recently, behavior-based targeting with any potential for scale was impossible, so marketers were forced to use these less accurate approaches. Now that more sophisticated forms of behavioral targeting have arrived, we should understand that not all approaches are created equal.
Joshua Koran is VP, targeting and optimization, at ValueClick.