• by Steve Smith , Staff Writer @popeyesm, November 2, 2007

Understanding the likelihood that an incoming customer will convert, or even if that customer is legitimate, a high spender or just a deadbeat, is a sixth sense among ace retail salespeople. In the digital world, you can't tell by the look of the shoes or the tone of voice whether this user or that is worth investing marketing resources or company trust. That is where predictive scoring comes in. As Gordon Meyer, CEO of eBureau, tells us this week, it is now possible to process the known attributes of an online lead in real time against large databases of past behaviors of people just like this prospect. Within a second -- the time it takes to scope a customer's jacket and shoes at retail -- his company scores help companies determine if the user is real or fake, a prospect or a window shopper.

Behavioral Insider: What sorts of behaviors do you gather, and from whom?

 Meyer: Predictive scoring is all about using past events to predict the future. So typically we get performance data, transactional data from customers. And it can be lots of different things we are trying to predict: whether someone is a fraud or not, whether someone is going to pay a bill or not, whether someone will convert from a lead or not, whether someone will respond from an offer or not. Then basically we run it through modeling techniques that have been around for a long time to correlate those independent data elements to whatever you are trying to predict.

Behavioral Insider: What are some examples of this in action?

 Meyer: You are running correlations between literally 50,000 predictive potential variables to a thing like is this lead going to convert? So we have to understand which variables are predictive. It will determine the general traits, habits and profiles of that consumer or business and do they match and fit with this particular thing. If I am doing a lead conversion model for education, there is a certain profile of people that are more likely to convert on that lead. It really tries to bear down on several factors about that population that makes it more likely that they will convert or not. You are merging transactional experience from [clients] with this massive warehouse of information of all this predictive content. You come up with an algorithm that says, here are the variables that correlate with that lead converting.

Behavioral Insider: This is happening in real time as the customer comes in the door?

 Meyer: Real time. One second. We convert a very simple three-digit score with potentially some reason code that may explain why they scored the way they did. The customer can use it by saying above 500 is somebody I will accept as a lead, or above 700 is a group of people I really need to invest in because the likelihood that they will convert is much higher.

Behavioral Insider: What level of information are you getting from the sites about these prospects?

 Meyer: Take a co-registration lead. Someone may have gone to a site to get some free service, and along the way gets asked if they are interested in music. So they fill out the form and then those leads come to us and we assess them for validity by just assessing and comparing a lot of identity data. So you are eliminating a lot of fictional character stuff. In co-registration you can have as much as 25% that is fictional or falsified stuff. That is much higher than keyword search [lead] -- those people have a real intent when they go somewhere, so there is less falsification.

Behavioral Insider: What data is the publisher passing to you?

 Meyer: It varies dramatically, but typically we need the name and address; phone is nice, date of birth, and information like what source channel this lead comes from. If you take a bill-me company like music that ships CDs and bills the customer, we also are doing some credit screening. We determine is the consumer going to pay or not.

Take another app like keyword search from a person interested in education. On a landing page you put them through a questioning process: who are you, what kind of education are you interested in, prior education. And they come to us to assess that consumer and score him and profile him as to the likelihood that he is a good educational lead or not for a particular university.

Behavioral Insider: How do you advise publishers to maximize the value of predictive modeling by asking questions that have more predictive value?  

Meyer: By using our services in many cases they can ask fewer questions, because they don't have to rely just on the consumer. In mortgage, we have all the mortgage and property ownership data, so we can determine things of that nature so [the client] can forego those questions and get real factual data.

Behavioral Insider: Can you use cookied behavioral data on a user?

Meyer: We have not done that and probably won't. We don't want to merge offline and online traffic data. It raises some privacy issues. 

Behavioral Insider: Can this process relate to measuring ad effectiveness?

Meyer: From the publisher's side we allow [clients] to assess the media sources they are capturing the lead information from and providing more transparency about that lead. So we're kind of evil to the publisher side today because they may be getting paid for all the leads and we may be telling them and the advertisers that 20% of the leads just aren't valid and have no chance of converting and shouldn't be consumed. We look at ourselves as having a score that provides more transparency about leads and therefore makes that whole market more efficient. So we see the publishers using it to help assess their media buys. For the advertiser it is real simple. It is assessing all the various lead sources that they buy and that lead's propensity to convert. Should I buy it at all? If I do buy it, what kind of resources should I put into it from a telemarketing or a direct marketing standpoint?

Behavioral Insider: Do you see this changing the game in lead generation?

Meyer: Historically, lead pricing has been pretty static in terms of fixed price. This will probably create the opportunity for variable lead pricing, because in essence the score can really indicate its ultimate value to the advertiser.  I think it's not happening yet, but the scoring has the potential for variable lead pricing. The publishers will probably fight that for a bit, but it's probably inevitable, and a lot of them know it's coming.