Predictive Modeling For The Masses
As the scale of both online traffic and online advertising trying to connect with it expands, the challenge of transforming all that quantity of traffic into quality intensifies. Till now, however, the tools to focus on building quality traffic have been reserved for larger advertisers. Most smaller advertisers have had to be content with hunting for quantity of clicks. In the conversation below, Scott Lynn, CEO of AdKnowledge, an ad network focused on SMB advertisers, outlines how an auction system for bidding on traffic based on behaviorally predictive analysis is beginning to help rectify that mismatch.
Behavioral Insider: What kinds of behavior do you believe are the most relevant to the kind of email targeting AdKnowledge has introduced?
Scott Lynn: We feel there's a lot of ambiguity about what behavioral targeting really is -- and by extension, confusion about what its meaning and role can be. The biggest companies construe visits to pages as a behavior. But are they? Maybe, but maybe not. For example, if a user visits a financial site, it tells you a little -- but by itself, it doesn't give you gauge or measure of their level of interest in what you have to offer. But if they actually interact on financial items to obtain more information, that's a behavioral indicator that's likely to be correlated with real-world performance marketing. So we don't count passive browsing as a full-fledged behavior.
BI: Can you get more into the specifics of how it works?
Lynn: AdStation is an API (that stands for Application Protocol Interface) that enables our clients to deliver targeted ads across any channel, email being one. What happens is our publisher clients who are motivated to increase their yield on inventory transmit an encrypted "hash" of user information. From there we run that data against a database we've compiled internally of over 300 million consumer behaviors to identify what that consumer has responded to in the past, and to see what kind of wider ad response patterns there are by similar users on the database. Based on that, we can predict the likelihood of their response to a specific kind of message.
BI: What kinds of data do clients provide?
Lynn: For instance if a publisher sends us information on 'User 1234,' we look at the database and we look at what they as individuals have clicked on or interacted with, as well as what messages consumers with a similar profile to theirs have clicked on. Say it's email tag lines we're focused on. We might find that people who've interacted frequently with hunting and camping content and also done research for autos have a high likelihood of opening emails with a Ford Truck tag line.
BI: Do you provide re-targeting?
Lynn: We see re-targeting as being geared more to pursuit of relevant brand ad impressions, where you're looking to reinforce a brand message among users with particular interest patterns. However what we're looking for is more focused on direct response metrics. The challenge we have is identifying behavior profiles highly predictive of direct response to very specific offers, in the case of email to specific taglines, for instance, or one tag line versus another.
BI: And this is done primarily internally?
Lynn: Yes, it's all internally derived data. We don't use third parties.
BI: What kinds of benchmarks do you use to gauge success?
Lynn: The nature of our model is an auction process which allows the huge 99% of smaller advertisers who in the past have only used Google's AdSense to bid on a far broader array of inventory. So you find out very, very quickly how well your targeting efforts are working. If bids are going up from 60 to 70 cents they like our traffic. If they're going down, we're doing something wrong so we need to adjust. What we're finding unequivocally is that the more closely we target behavioral patterns the better the bids get.
BI: Any examples of how this is actually being deployed in real-world marketing?
Lynn: We have one client whose permission-based email marketing business had taken off - actually, too much. As it did, their number of third-party affiliate marketers had also mushroomed. The challenge was, they now had an incredible number of offers from all these people looking to market to their lists, but they had no idea which offers to target to which customers on their very huge, very demographically diverse, database. That's the kind of situation where the kind of behaviorally based profiling we're talking about makes the most sense. The technology is designed to match up each customer on the list with the offers that will likely be most relevant to them. The result has been a significant rise in CPM, delivery rates, open rates and conversions.
BI: What are your goals for both AdStation and behavioral targeting more generally over the next several months?
Lynn: The method of matching users with relevant offers based on an auction model applies across channels. So far we've targeted search, banners, email graphics and email taglines but we see the applicability as extending beyond that, especially into video messages.
At this point and going forward over the next several months we see two trends at the forefront of driving adoption to more behaviorally based approaches. The competitive drive for better traffic quality on the part of advertisers, and for yield improvement on the part of publishers, makes constant sharpening of the targeting edge imperative just to compete.