Predictive Modeling Across Channels
Lead generation and email marketing are often cursorily dismissed as "old school" by aficionados of the ever-new. Yet when it comes to the new next and best in behavioral platforms, so called "predictive modeling," Matt Wise, CEO of Q Interactive explains below, lead generation has actually blazed trails well ahead of the pack.
Behavioral Insider: Q Interactive has made its main reputation so far in lead generation, where most of the action in behavioral targeting has been associated with display ads. What role has behavioral targeting played in your evolution, and where are you with it now?
Matt Wise: It's true we haven't until now been in display advertising. Most people see us as a lead generation firm, which is something that's conventionally put in a separate bucket from behavioral targeting. But what's interesting is that our history actually gives us a unique take on how behavioral data can be leveraged. Our history goes back to the mid-‘90s. A major focus for us for a long time has been on how to generate revenue from post-registration leads. The key was that we used online registration data, online behavior and tied it to offline data sources to predict the likelihood of response to particular offers. So we were working in the area now called ‘predictive targeting' before it had a name.
BI: So in a way you've jumped past the so-called retargeting stage?
Wise: The use of multiple data points marks a major departure from behavioral targeting as most advertisers still understand it. Retargeting, the hot topic of the past two or three years, for instance, is based on one-dimension -- Web surfing -- which in and of itself, is a relatively weak indicator of future behavior.
I think of retargeting as Behavioral Targeting 1.1 (with basic contextual targeting being 1.0). The premise is, ‘I see someone is doing a lot of browsing in auto. But hey, the top two or three auto sites are all sold out of ad inventory. So I'll deliver the same ad somewhere else that that consumer goes to.' That's really just an extension of contextual advertising. When you think of it, it's not really fundamentally different from what an ad man would have done in the 1950s. It's like waking up and saying, ‘I need to sell Pampers. I think mothers with kids read Family Circle, so I'll buy an ad there.' That's the best they could do.
We can do much better than that by taking in all the information derived from registration data and matching it against offline data sources. Then it's no longer guessing. If you know how to connect all the data at your disposal you can look at a woman, 39, with family income between $75,000 and $125,000, and predict with high accuracy which offers to serve. It's no longer a guess.
BI: This is all anonymous data?
Wise: We don't target based on personally identifiable information, so we don't target to Ms. Jane Smith. Instead we target to the prospect as a member of the segment based on particular data points. The key again is not what an individual has done with one particular past behavior, but associating all available data points. When we look at data points related to a consumer segment, they exist across a wide continuum. The most basic and rudimentary level is contextual, just having information about the site they're on. But where possible, that information can be related to IP information, which yields geographic data, from which demographic clustering profiles can be derived. On top of that are layers of transactional data, online transactions and purchases in the offline world. We tap into as many pieces of data within that wide continuum as we can. The more that can be converged, the richer the segment.
BI: How is the registration data generated and deployed, and how does that relate to privacy issues?
Wise: All of our registration and post-registration data is based on an opt-in model.
The consumer is informed about the processes and how their information may be used. Most privacy concerns that are now being discussed on a government level are somewhat reactionary. The issue should be presented, and hopefully will be by the industry, in terms of economic efficiency. Companies waste millions of dollars delivering a glut of irrelevant ads to consumers who aren't interested and therefore waste consumers' time. What behavioral targeting is about is eliminating the waste.
One concern I have is that behavioral targeting has flipped by many practitioners in the direction of excessive caution. That is, only targeting known behaviors. By paying attention only to known intentions such as keyword search, they're already focused far down the funnel. This neglects the whole area of awareness building and branding and how predictive targeting can be leveraged there.
An example of how efficiency can be enhanced on a direct response level can be seen with magazine subscribers. Magazine subscription marketing has been notoriously hard to do online. Offline consumers just get an insert card and X percent pop it in the mailbox and magazine publishers have a good idea of what percent of them will actually pay. Online, on the other hand, everybody fills out a form and nobody pays. What we can do is use consumer behavioral segments and predict with great precision who will pay.
On a branding level, a company may say ‘hey we want to figure out how to reach people who influence moms with children who live, say, on the East Coast.' We can leverage multiple data points and blend them together and see who may not be clicking on ads, but who looks at ads for new products for mothers and forwards them.
Our goal is that in a few years people will look back at the early days of online advertising and think, ‘wow, how could so much money have been spent on random ads and how could advertisers have lived with such low expectations of ROI?'