• by Steve Smith , Staff Writer @popeyesm, March 23, 2007

As we saw recently with ActiveAthlete, content verticals can access a depth of profiling information that typically eludes a larger BT network. Once users are registered, or their passion is well-documented, you can start targeting against their future actions in very precise ways. SilverCarrot is an eight-year-old performance marketing company that uses robust content at lifestyle sites like Recipe4Living.com and 40 e-letters to capture 50,000 to 70,000 new registrants a day in the 25- to 50-year-old female segment. But as CEO Allan Levy explains, once you get a member into the system, your database engine can learn and respond to a stunning range of behaviors and responses to different stimuli. As Levy outlines here, SilverCarrot developed a performance advertising engine of its own based on behaviors that are both specific to the individual users and generalized across a history of use.

Behavioral Insider: What happens when someone responds to one of your initial offers to come to the site?

Allan Levy: When they sign up we have about 40 different newsletter topics we market to -- health, crafting, recipes, etc.. The newsletters themselves are ad-supported. We do look at the data and target based on opens, clicks, response rates to various offers. We have a rules-based engine we built ourselves. It's got a very deep set of algorithms and it's all based off of the 'link,' what we call the links that we use to acquire the traffic.

On each media source, each creative we run, there is a different link regardless of media source. So if I'm running a banner on iVillage and I've got six different acquisition banners for Recipe4Living, each link would be different. Maybe I am running a promotion to win a kitchen makeover when you join, or something for Williams-Sonoma. Our system looks at the information the consumer comes in on and says, OK, this consumer came in from an iVillage link or this promotion for Recipe4Living -- and historically, consumers who came in off of that point took this action.

And we have about 1,000 different offers and advertisers we work with. We see they have responded to this type of offer. So if there is Suzy Jones, she's 62, she's coming in from the Midwest, we take certain actions based on where the link is and where she is coming from. We take certain demos and geo information. We compare that to the information that we have in our database of other consumers who are similar, and then we show specific promotions.

BI: So women coming from specific places online, and from specific places in the country and responding to certain appeals, are known to act in different ways?

Levy: So we find that women coming in on the fiction promotion link from iVillage who are also from the Midwest and in their 40s have responded better to work-from-home offers. And the next person who comes in matches women from the South who have responded better to online education offers. We cookie them so that if they do come in again, we can clearly see that they have said yes to take an education offer but never filled it out. The best thing to do there is to show them another educational offer. Or if they said yes and took an education offer, we suppress those offers from being shown.

BI: It sounds like an intricate decision tree that is based both on the choices the user is making and the historical behaviors of people just like her.

Levy: We have our own team... that comes up with at least ten new creatives. The decision tree looks at the different media sources that we're buying media from, Google, banners or newsletters. The system looks at every single link individually and at historical performance. It will rotate in every offer into every position to test where that offer best performs within the flow of offers. It rotates the offer in. So it's pretty much taking whatever it knows from the system and takes a step back and says how people of the same gender, age, zip code or region have historically performed for these types of offers.

BI: Over time, has the nature of the appeal had to change -- or what it takes to move people into the system?

Levy: Yes. They definitely have evolved. The users have gotten smarter and we used to get much higher open and click-through rates. A couple of years ago we were not as concerned about the percentage of clicks from the opens, because we just wanted to get as many people in the door. But today if people keep getting unsatisfied, the unsubscribe rate jumps up. That has evolved substantially, that we are much more conscious of that open-to-click rate.

BI: What have you learned about registration and getting people to opt in?

Levy: We look very carefully at the correlation between opens and click-throughs. We take a sub-segment every week and we test the various creatives against that list. We'll test opens against subject lines, click vs. creative, etc. We will only go out and acquire members based on successful creative. Some of the metrics we look for are not just opens and clicks but the combination of both opens and clicks. You may have a broad subject line which opens very well, but then you basically frustrated the user because they didn't get enough information in the subject line to qualify them, and they're not clicking on the campaign. The opens piqued their interest. If 25% or more clicked through, then we know that subject line related enough to the actual content that they wanted to go further. If it is below that 25% then you didn't relate enough. You're ultimately frustrating the user -- and that is not good for us or the publisher where we buy the media.