Thought Leaders in Big Data: Interview with Dr. Saed Syad, Chief Data Scientist at AdTheorent (Part 4)

Sramana Mitra: One common experience I find myself having often is that I go to a website that I use to keep track of theaters, for example. Then I buy tickets to a particular show. But there is some sort of a re-targeting algorithm that has picked up that I was interested in this show and it constantly shows me those ads everywhere I go. It is a complete waste of advertising dollars because I already bought the ticket. Can you comment on why we don’t have the input from actual transactions yet?

Saed Syad: The existing targeting algorithms they use are simple and unintelligent. If you visit some places, they just go after you everywhere you go. RTLM is going to change that behavior. The way we do it is not just based on this simple visiting of a page. We get the various information of the user from the website and also from third parties, instead of just looking at the data and touching some of those variables. This is a predictive model in real-time about the behavior of the user. We put many variables in the system, and we detect its interactions. The model is taking care of many aspects of the user’s behavior before bidding on that request or even sending the creative to the website. The reasons we don’t have this at this stage of the situation is because they can not handle the amount of data they have. They use very simple models, and that is the reason why the behavior of those systems is sometimes annoying.

The key here for the future is that those predictive models, with the capability of handling huge amounts of data in real-time, will at the same time be creating some sort of abstract model with enough speed and the ability to change the model on the fly. Then we can see a completely different game.

SM: Are you plugging into transaction databases and CRM systems to bring that kind of data into your real-time learning machine?

SS: We don’t look at the transactional data, but we look at the third-party data, and then we can input any data. If we have the transactional data we can use it, because of the scalability.

SM: Maybe you should look into talking with your customers about transactional data. I think a huge amount of advertising dollars is lost because of this lack of a bridge between the advertising systems and transactional systems.

SS: That is a good point. It is a young industry. There are many new ideas in this field and one of them is going to be this one.

SM: What is your business model? Do you license your technology to ad networks?

SS: Right now we are using it internally. Mostly we use the power of RTLM to improve the outcome of the campaign. If it is a CTR (click-through rate) campaign – if we don’t use our predictive modeling – we have a rate of 0.4% average. When we use our RTLM model we reach 1.2% to 1.5%, which is unique. Nobody else is reaching that level. Then we have a three- to five-time ROI for us and for our clients.

SM: Who are the clients?

SS: Advertisers. We are DSP (demand side providers). Our clients are McDonald’s, Pepsi, Coca-Cola and Bank of America, for example. They wanted to put ads on mobile devices. We have a contract with them and put their ads on our system.

This segment is part 4 in the series : Thought Leaders in Big Data: Interview with Dr. Saed Syad, Chief Data Scientist at AdTheorent
1 2 3 4 5 6