As the world’s most interactive medium of information exchange, the Web generates enough forensic evidence on individual online “behavior” to make a CSI detective team positively giddy. That is, of course, if you know where to look. Sure, the Internet lacks “Caller ID” and there are elaborate lengths hackers can go to disguise their identity. But there are also enough observable click tracks, look ups, sign ups, page views and other online behaviors to make a privacy proponent blush.
In its early days, the company formerly known as DigiMine was in the business of mining the depths of this Web user data from its theoretical and analytical quarry. Then a miner’s headlamp clicked on. DigiMine recognized there was an offering it could make that just might save advertising as we know it. A new name, Revenue Science, and a new mission, was born.
“We decided to focus completely on behavioral targeting because we felt it would become a multi-billion dollar opportunity over the next three years and one of the top three forms of advertising online,” gushes Omar Tawakol, senior vice president of marketing, in this interview. “It was the evolution that energized everyone.”
To understand how the concept of behavioral targeting will radically change the online revenue landscape, it’s helpful to start not at the beginning, but at the end — with the ultimate benefit that publishers will derive from it.
Typically, the way Revenue Science will start out with a client — such as ESPN, The Wall Street Journal, Reuters, CBS MarketWatch or the Dow Jones properties — is by identifying low-hanging fruit. Which sections of a publisher’s Website are already sold out? In essence, that is a signal. The market of ad buyers is saying, “I really, really want the audience that is defined by that place.” Now, imagine if, instead of saying, “Sorry, I’ve sold out of that advertising section” and losing money, you are able to say, “Wait, I’m going to get you the most loyal of the same customers you are seeking, and you can get them anywhere on the site!”
“In essence, we’ve just created more reach,” says Tawakol. “Revenue Science can go to a publisher and ask, ‘What are the sections that are sold out today, and we’ll instantly create audience segments around that.’ Or, we can say, ‘Give me your top advertiser categories that you want to do more business in, and we’ll package an audience built for that!’
“The way I like to think about this at the highest level,” says Omar who has done some very potent thinking at Stanford University’s Logic Group, HP Software Labs and MIT, “is that every page is an opportunity for the publisher to make money. You can assign a value to the page itself or you can assign the value to the person looking at the page. If I were a publisher, I would see which has the higher value. We’ve added a new way to monetize and that ends up radically impacting the revenue.”
Here’s our conversation with Omar on the science and art of Revenue Science.
Seattle24x7: How did DigiMine turn the corner into Revenue Science?
Omar: DigiMine was doing analytics and behavioral segmentation for leading publishers and advertisers. We looked at the market and recognized that the behavioral targeting opportunities were just huge. There was more to it than just applying a horizontal solution in terms of figuring out how to make it work end-to-end. We decided to focus completely on behavioral targeting knowing that it would become a multi-billion dollar opportunity over the next three years and one of the top three forms of advertising online.
Seattle24x7: DigiMine worked with mining data on an enormous scale. Do you have the same data management requirements today?
Omar: In the search world, you have to index every document. In the behavioral world, every click. With our Audience Search capabilities, we’ve actually combined both concepts. So I can know every word read by every person on every click. The dimension within this space is vast. Lots of attributes. The number of users is large. The volume is large. We’ve been able to attract some pretty good minds to throw at the problem by taking an approach to scalability that’s similiar to Google’s.
Seattle24x7: What amount of actual visitor data typically gets collected in order to form the basis for behavioral targeting?
Omar: We’re capturing every click and categorizing every click, every demographic and every word of the page and we generally target those attributes within a rolling three month window. Let’s take an example. If you click on a Loan Calculator, that event, in and of itself, is an indicative event that you’re interested in a loan. If you’ve got a single page that has the word “Loan” on it, that is much less of an indicator. However, if you’ve read five different articles on Loans and the average person has only read one, then that tells you something.
If you read one word about “Loans,” that’s interesting, that’s what the Google’s and Overture’s of the world would do in the contextual model, but we wouldn’t target that. What we would do is look at how much information you encountered that was about the Loan topic. We’d then compare that to the standard population. If our statistical confidence is high, then we’d say you’re targetable.
We’re able to correlate things like how frequently you look at content, how recently, and how relevant that content is and how differentiated you are because of your behavior. If you look at AdPlan in Nielsen, everything they do is indexed to 100 which means Web-wide. If you score 150 on 5+ business travel, it means that you’re 50% more likely than the rest of the population to be a 5+ business traveler. We’re doing the same thing but with behavior. So if you say you’re interested in travel and loyalty plans, we say you’re two times as interested in travel and loyalty plans than the rest of the people at the site. That’s the Behavior Relevance Score.
When we say behavior, we mean both what you say, which is your registration data, and what you do, which are the categories or words you read, and then how you do it, which would be geography, IP attributes and so on. We put it all together and it allows us to create segments in any or all of those attributes. You can get people who are Fortune 5000 C-Level decision makers who read about business or technology.
Seattle24x7: Are you targeting behavior at the individual or group level, and how many sites are being tracked?
Omar: The targeting is done at the segment level. The calculation about whether you’re in the segment is done at the individual level. As far as where data is collected, the quick answer is that we do it one business at a time. So, for example, we’d do it across Dow Jones properties. We currently don’t do Dow Jones and CBS. Those are kept separate.
But the question you’re asking is really one of the most fundamental questions in targeting and in advertising. And that question is, how do I trade off reach for relevance?
I can reach everybody and have low relevance. Or I can reach a very small set of very high relevance. We actually create a visual paradigm that’s an audience search engine. You type in the kind of attributes you’re looking for like “South Asian Business Travelers,” or “People who read about South Asian business and who are travelers,” and then, for example, we allow Singapore Airlines to target them.
Seattle24x7:What other behavioral characteristics are you looking at?
Omar: We have a notion of the life of the user session, we look at recency and frequency at a page level, not just a word level. We have a concept of an event, where an event is anything that the publisher is interested in. Did you click in the stock ticker? Did you use the loan calculator and actually do a compare? We call these events. Whether you visited a category multiple times. And we can look at words on the page. Words on the page are adding additional granularity. You can also do things at less granularity and that could still be very good.
Seattle24x7: What happens on the other end of the tunnel. Are we witnessing the arrival of truly segmented messaging in terms of the creative product. Are you testing creative versions?
Omar: This is an area where we are going to be putting a lot more focus. Where we’ve started is with the publisher. The first level of use is before you customize a message, to just figure out how you do the campaign buy, when you’re getting the right audience. Even assuming you just had one message, getting the message to the right audience is the first order of business for the agency. But we’re starting to do work here. For instance, how you relate to a loyal customer who you’ve had in the past vs. a new customer — the difference in the messages.
Seattle24x7: What is the system demonstrating in terms of results?
Omar: There are two ways that we encourage people to measure the value of behavioral targeting. The first one has to do with campaigns that are built around awareness and persuasion. The second one is basically direct response. Where the media, especialy media that is shifting from TV to the Internet, really needs help is in doing awareness and persuasion campaigns online that aren’t purely direct response.
You really have to think about this in the sense that of all the money going into TV, they’re very aware that you’re not going to click on your TV and buy things; it may be next week until you buy. But they know they’re investing the dollars in the right way. They’re not limiting the value to the second that you see the ad. Unfortunately in online, there tends to be a lot of measurement of how does this effect the moment and undervalues the actual branding capabilities and persuasive capabilities of advertising online. We’re seeing 100 to 1000 percent lift across different types of behavioral targeted segments. For example, if you bought a run-of-site campaign, maybe 10% of the population was actually in the advertiser’s target group. In the behaviorally targeted campaign, it was 70% within the targeted group. We’ve been doing a lot of testing on how to really increase the composition. If you think about the goal of targeting it is to get you to the right person.
Seattle24x7: What Revenue Science is making possible is really run-of-site advertising, but by audience profile not by page content?
Omar: Every page opportunity is an opportunity for you, the publisher, to make money. You can assign a value to the page itself or you can assign the value to the person looking at the page. If I were a publisher, I would do that and see which has the higher value. If it’s the Technology page and it’s selling at $80 CPM, that’s fine. If I’m at the News page, and I have a technology decision maker looking at it, why price the News page for less to target the same user? We’ve added a new way to monetize and that ends up radically impacting the revenue. [24×7]
Larry Sivitz is the Managing Editor of Seattle24x7.
