Online marketers often like to think of themselves of themselves as pioneers of a radically new world with its own laws, untethered to the ways and limits of the old world they left behind. In many
ways this attitude is justified. But, as Bennett Zucker, vice president of marketing at aCerno, explains below, when it comes to segmenting consumers by predictive behaviors, online has much to learn
from the legacy of the past, particularly the paths blazed decades ago by direct mail catalog marketers.Behavioral Insider: What was the motivation for developing your predictive
modeling technologies? How do they expand the conventional understanding and practice of how behavioral data can be deployed?
There's actually a long heritage
in direct response that the best data for predicting future purchasing is past shopping behavior. We took a concept that had a long history offline and brought it online. In direct mail, postal data
cooperatives go back decades. The idea is that direct response advertisers pool mailing lists so each can enhance their own list based on rich, shared resources for understanding shopping
behavior. Because competitors pool data they can benefit from having far more access to predictive information than they would be able to generate internally on their own. We go out and sign up large
e-commerce sites. They permit us to tag their pages and we collect anonymous data. We currently have data from over 375 multichannel retailers and over 140 million shoppers.BI:
What kinds of data sets are used?
ACerno provides two tracks of information. First we ascertain what customers will buy or browse next based on what they've
shopped for previously. The anonymous data gives us visibility into what people have shopped for, what they've abandoned in their shopping carts and what they've browsed. With only a limited
universe of data, the kind most e-marketers have access to on their own, you have limited visibility into overall consumer behavioral patterns. But when you have the scale, computers can score and
rank likely future behavior based on the past. If you have enough data points it becomes predictive. Once you know a consumer's actual buying habits as opposed to just their habits in interacting
with editorial content or ads, you can serve ads that are based on more than inference.
If you look, on the other hand -- which is what conventional BT tends to do -- at
isolated data points, patterns can be deceptive. Say you know someone has read an article on Tiger Woods. You can surmise they like golf best, but there's also a chance they're just interested
in celebrities.BI: Is it relevant to branding as well as direct response?
Retail has been our core base, but we're seeing increasing traction in
auto and travel. The core business has been direct response transactions, but the methodology has a strong fit with the needs of brand marketers as well.
Which brings us to the second
track of information we're looking for: who the customers are. For brands the challenge is to link behavior together in a way that's more descriptive than predictive. Data sets of purchases
can give you visibility into who consumers are demographically or psycho-graphically. So a brand can start with a model or profile of who its best customers are, perhaps as derived from their CRM
data. Then it can apply those criteria to locate and serve ads to consumers whose purchase patterns most closely fit that profile.BI: Can you give an example of where and how
predictive modeling is being deployed? How does predictive modeling enhance behavioral targeting campaigns?
3M was introducing a new Post-It note line targeting women
working in "paper-intensive" jobs with a high fashion and design component.
How do you find consumers with that profile based on conventional targeting? You can target content
publications that seem to correlate with that demographic, but it's such a unique specific segment that's likely to be difficult. Or you could try a more hit-or-miss approach based on
people's piecemeal browsing or searches.
But what we were able to do was blend together from our shopping data women who bought fashion apparel, women who worked in home offices and who
had purchased designer materials and objects. By blending these segments together we would be able to home in on just the right subset. BI: What are your goals for the balance
One of the challenges going forward for us is to work with metrics companies who traditionally measure an audience based on the ad impressions on sites.
But that's not really an adequate way to measure what we do -- which is to collect data from retail sites that don't have ads and then run ads on other sites. If you only look at where ads
get served, that's not really an accurate indicator of the value of the audience data that we're pooling.