TACODA Systems' Dave Morgan provides examples of effective behavioral targeting. (Part 1 of 4)

This breakout session took place at February's iMedia Brand Summit in Florida. Sean Finnegan, U.S. director of OMD Digital, moderated.

Sean Finnegan: This breakout session is on behavioral targeting. It’s an interesting subject. You can go an inch or a mile deep on it. There are a variety of partners, a variety of services and technologies. Some of you guys in here are probably extreme experts in your own company’s representation of it or your own brands’ and products’ experience with it. We’re going to take a slice out of that and go a little bit deeper with one segment of that with TACODA Systems and Revenue Science.

We’ll take you through the agenda real quick -- a quick introduction and company offerings: 30 seconds, no sales pitch because we want to get into the case studies. And then we’re going to talk a little bit about benefits and issues. And then leave it open to questions and answers in case we didn’t cover anything and you have any burning questions. Feel free to pipe in. This is sort of a random amount of information and if there’s something that you feel could benefit the group, definitely, please chime in.

After this morning you guys are probably aware of me and probably, arguably, sick of me. So I’m going to bypass my introduction and lead it off with Dave Morgan (CEO, TACODA Systems) and then Omar Tawakol (SVP product marketing, Revenue Science Inc.).  We’ll start with Dave.

Dave Morgan: Okay. Thanks Sean. TACODA is a four-year-old company based in New York. We have two service offerings. We provide targeting services to publishers. We call it our Audience Management Services. And we’ve recently launched an advertising network called the Audience Match Network for advertisers to be able to buy targeted ads directly. 

Basically what our targeting services do, they give the publisher the ability to build a database of their audience, particularly to include in this database the behaviors of the audience, what kind of content they consume, what kind of loyalty. That’s matched with registration data. And then have that available for real-time targeting on their websites. And so the kind of sites that use our services are a lot of the newspaper sites, probably 75 percent of the major local newspaper sites in the country, from the New York Times on down to Cleveland Live, some of the large vertical sites like weather.com, About.com, iVillage … and they give marketers the ability to, instead of just buying on the auto page or the travel page to target ads to auto buyers or travelers.

Okay. Omar?

Omar Tawakol: Hi, my name is Omar Tawakol and I run marketing at Revenue Science. I got my start in this field back in ’94 when I remember in graduate school a professor came to me and said your assignment is to map the web. And back then you could actually fit it on a poster. But it’s changed a lot since then, I assure you.

Revenue Science has been around for four years doing behavioral segmentation. We provide behavioral targeting through premium publishers, focused on particular verticals. So if you look at the financial services and financial news area, we work with people like The Wall Street Journal and Financial Times and The Street and Reuters and a whole set of premium publishers. And if you look at sports we work at ESPN; in the auto section people like Kelly Blue Book and Edmonds and Autobytel; and in the news area ABC News and The Washington Post and so on. 

The reason we focus on these verticals, they have extremely rich behaviors and marketers are after these behaviors. And our whole model is all based on allowing you to target people and not pages. We offer this capability through two services, our audience search capability that allows publishers to package, to find, evaluate and execute on audiences that an advertiser would buy. In particular we focus on brand advertisers. Eighty percent of the top 20 Adweek brand advertisers have bought behavioral targeting through our premium publishers. And the trend that we were seeing near the end of 2004, and now into 2005, is big momentum in terms of revenue growth month-to-month through behavioral targeting; and adoption where advertisers test that one site, they like what they’re seeing and then they try it out at other vertical sites. So we expect 2005 to be a very interesting year.

Finnegan: Okay, that’s a nice set-up. Let’s go a little bit deeper with some actual results. We actually have two travel-based case studies for you. I’ll have Dave Morgan take you through the first one. 

Morgan: Great. What I thought I’d do is try to focus in on a pretty simple model on how behavioral targeting is being used by publishers and advertisers. One of the things that we’ve learned -- and I'm sure Omar at Revenue Science has had some of the same experiences we have -- is that when you start moving into more advanced targeting you’re certainly starting to get closer to a lot of proprietary information of clients. And so, many of the case studies we use will be unnamed because that’s been one of the requirements in being able to do this. So we can’t always shout from the rooftops with some of the brands that have been doing it. 

I want to use an example of how clients are approaching this. This is a travel client working with a relatively large content website, using our technology. They’re just getting started in the notion of behavioral targeting. They know a fair amount of information from their own analytics and their own marketing about who they want to target and who their business travel audience is. But on the web it’s all been limited to contextual buys -- buying in the business section or in the travel section, let’s say. So now when you start talking about behavioral targeting -- and even in just a very simple way -- you can now start looking at people who have been in the business section when they’re in some other section, like sports. Or people who have been in the travel section when they’re in some other section like entertainment. Or, people who have been in both the business section and travel section and maybe that gets a tighter match. 

So the simple platform is to start identifying the audiences. For this publisher, the first thing they did was to try to understand “How many people do we have that are in both business and travel every month? How many people do we have that are just business? How many people do we have that are just in travel? And, how many page views can be associated with that?” So you can see numbers here; 150,000 are in both; up to 600,000 for the business audience. And then try to understand what inventory might be available. You know 50,000,000 or 60,000,000 page views of the business audience; 26,000,000 of the matched between the two; and a certain number in travel.

The objectives in the campaign, and for this advertiser, were, as they always are, drive traffic and bookings and make money. But it was also to learn: Let us start understanding what the audience looks like when they’re coming into our sites and then we can start understanding these characteristics. And so, it was identifying the consumers focused in these two vertical areas and also the ones who were overlapping. And to try to -- and I think this is one of the most important pieces and it really hasn’t gotten attention in behavioral targeting yet -- deliver audience-relevant creative. Because being able to find a very specific audience means nothing if you’re really not going to focus your creative on what you’re going to do with that audience once you have them. When you start targeting to people and not pages, you’ve got to have much more people-centric creative rather than the page-centric creative.

So what was done was the first phase of the campaign was just to start looking at these audience segments and market segments and start to profile the buyers and the non-buyers. One advantage with the behavioral targeting services, both of ours and Revenue Science’s, are that you don’t always have to target to your campaigns to know how your campaigns performed by target. In other words, to have the measurement aspect to understand how people who have been in business and are not in sports, or have been in both business and travel and are now in entertainment, perform -- and start to match those against the ideal profiles of a target audience.

After running the basic campaigns through with the first sets of creatives and trying to figure out “Who are the lookers? Who are the bookers?” then we came back in Phase 2. And this is something that’s pretty critical in behavioral targeting, is to try to learn -- much like in direct marketing -- from those first experiences, re-segment the audience, maybe find that there’s a certain pattern of the number of pages they’ve consumed, or their loyalty of visit, or exactly what out-of-context area that you find them to be “a better looker or a better booker.” And then test new creatives against those, refine the creative and start seeing what drops out the bottom.

In this case, the results -- and as I say, it’s unfortunate … I usually like to try to share things with numbers, but I thought I’d just give a simple one and we couldn’t talk about the numbers here -- but the amount of bookings that they were able to achieve out of the same media spend and media buy here was dramatically higher than what they’d had previously. I mean, exponentially higher than what they had been running before they were doing this. And the result to them, obviously, in the end was the dollar-per-booking was just dramatically lower.

Now on the perm media unit, purchase was a little bit higher, but the total campaign was the same and the cost of acquisition was dramatically lower.

Finnegan: Dave, just to talk about specifics for one second, on Campaign Phase 2, the testing of the creative and messaging and profiles -- how long did it take? You obviously got the information immediately, but then how long did it take, in this instance, for the turnaround and the optimization of that to get back into the media plan? And then also maybe a comment on … this is obviously impacting production, so you’ve got different messages, different segments, different people, different creative elements … how many? Do you have any sense of how many different ad units or components to ads that were dynamically put together?

Morgan: This one didn’t really push a lot of the dynamic pieces.

Finnegan: Right. That’s fine.

Morgan: Though that’s where it needs to go. Clearly, to take advantage of it you need to be able to change your pricing and your destinations. And you need to be able to really have dynamic creative. 

In this case I think it was probably like a grid of about four different ones against four different groups, so probably like 16 different units, I think. The thing that is critical, and that was done here, is when you bring up the turnaround time. You can analyze data forever. And one of the problems with some of the early attempts in behavioral targeting -- and when I was running Real Media we were doing some things with folks back in ’96 and ’97 -- was that it might take a month to get the lessons and then re-segment an audience and then target them again. That’s part of the problem with some of the homegrown solutions that are out there. The least turnaround was minutes. So the segments were recalculated within minutes.

And we’ve also worked on campaigns where segments are made dynamic so they recalculate on their own, on a constant basis and start adjusting to a model. So you can adjust to, let’s say a booker model, and actually have the segment recalculating itself and re-segmenting the audience into it as the campaign goes, so basically it self-optimizes.

Finnegan: Okay.

Morgan: One of the things that you hear a lot of times when people talk about behavioral targeting is, "well the audiences are too small. We can’t really make reach when we start narrowing it down that much." And so I wanted to give an example of something that people wouldn’t have normally thought about as behavioral targeting, but it’s very powerful and very mass. Many of you may be familiar with HGTV’s Dream Home Giveaway. It’s the largest annual sweepstakes on the internet. Scripps Networks, which uses TACODA’s technology, last year when they ran the competition decided to basically incorporate that database into their targeting database. So for a six-week time period … and what they wanted to do was demonstrate to their advertisers, even first just from an information basis, who are the kinds of people entering the sweepstakes because there’s a lot of money that’s being put up offline for this. And so this is a lot of information going back to the client about the kinds of people who register. This only is something that has been possible online, is to tell about what kind of behaviors these people have when they're on cooking, when they’re doing other things. 

So in this six-week time period there were 36,000,000 entries. Now you’re allowed to register multiple times. So our system did a de-duplication of it and found there were 6,000,000 unique individuals. There were 200 segments established against these 6,000,000. And just to give you a sense of the granularity here, it was not only things like geography, or how their browsing habits had been on other content that they have, but the Dream Home Sweepstakes used one of those view-around kinds of technologies where you could actually look inside the home. I think it was from iPix technology. And you could actually measure, much like you could in a rich media ad unit, all the interactions. And they were all branded: Pella Windows, Armstrong Flooring. These segments were things like, “How many of these 6,000,000 people interacted with Pella Windows? How many interacted with Armstrong Flooring?” 

And then, of course, not only was that incredibly valuable information to people that are spending millions of dollars as advertisers to HGTV proper, but the retargeting capability, now, for them to go into … and this database keeps living and growing, as you might imagine, with this year’s contest. Now they can go in and say we’re not only going to get you to people who we think have aspirations of being home buyers, we’re going to get you to people who have been interacting with Pella Windows, that have a very specific interest. And so, the numbers, the gross numbers are big. These are big media numbers when you get 6,000,000 people on a home-buying mode that you can target. But they also can be very granular in that you can get those 100,000 that expressed an interest in flooring products, for example.

Finnegan: Okay. Thank you Dave.

Tomorrow: Case studies from Revenue Science.