Thought Leaders in Big Data: Interview with Bill Simmons, CTO of DataXu (Part 5)

Sramana Mitra: Let’s talk about the optimization bit. Once you have the data measurement and analytics infrastructure, what kind of optimization are you able to do? Give me some use cases, please.

Bill Simmons: The typical use case for traditional media is called direct response advertising. If you want to place ads and drive people to your site to purchase things, that is our core, and we do that very well. In newer use cases a customer typically buys guaranteed buys from a large site. One million impressions a day for $10,000, for example.

But what they typically do is rotate the ads behind the scenes to a simple business rule that is selecting which product to show to the user. One of the ways we are able to help them – since we are collecting all the data across all their advertisements, we also have third-party data from our data brokers – is to add decision making to these guaranteed buys. Instead of just rotating and putting a simple business rule [in place], customers can use the DataXu system to proactively select the right message.

Getting back to the example of auto manufacturers: If somebody has a history of being interested in convertibles, we can show that person a convertible ad rather than an ad for a randomly selected car model. Our studies have shown that this results in a significant uplift in performance. Our goal is to become a decision engine across all [a customer’s] digital media buys to help them improve their performance and by leveraging all their data.

SM: Google has this kind of optimization – both the optimization and the learning capability in their AdWords and AdSense products. Is that correct?

BS: Not exactly. Google has auto-tuning on a campaign-by-campaign basis. That helps you get better performance. The difference with DataXu is that we can integrate with a CRM system, with your internal data and your in-house analytics team to build a custom solution. Google has a good offering. We find that there interest in the marketplace to work with a company like DataXu because they don’t really want to share their deals with Google. We can get additional performance out of the media because we have additional data.

SM: What I am observing is that in search, if you offer 25 different keywords and 25 variations of ads in the whole Google self-service AdWords product, it will test the combinations to figure out which ones get the maximum performance. What I am saying is that they are capable of optimizing campaigns. You are doing that at a much larger scale.

BS: That is correct. Google does that on a campaign-by-campaign basis, and they are very good at doing it with keywords. We take a similar approach, but we apply it globally across the entire media allocation.

SM: Exactly. So, that is something they wouldn’t do. They would do it within their systems, but they wouldn’t be able to do it across all ad exchanges, which is what you are doing. You are doing that same kind of optimization, but you are doing it across the entire media scope.

BS: That is right. And we do it without a conflict of interest, too. Google is optimizing to sell more Google media. We are only going to buy Google media if it performs the best.

SM: What percentage of your work is automated? Is there any portion of it that has to be done manually?

BS: What we have learned developing our product is that every customer is different. Some want more automation and some less. You have the option to let the system optimize on its own. Then you have the option to add your human intuition to guide the system. We find that this performs well. There are things that a person may know and that the computer can’t always put that well. For example, if you are working for a retailer that is going to drop prices the next weekend, your system can’t always predict that. But a human operator can say: “I am going to tell the system to ramp up and buy more aggressively starting on Friday because I know prices are going to drop and conversions are going to happen.” I would say it is half and half, and that is probably the way it should be. I am a person who loves automation, and I am a geek dedicated to that as my career. But I also believe that an automation system should be able to be overwritten – the ability to add your human intuition to make it even smarter.

SM: How does that reflect on your organization’s design? Does every client need a certain professional service to cater to them?

BS: We have a services organization here. We find that just throwing a black box at clients is just doesn’t work.

SM: These kinds of quantitative skills don’t really exist inside of marketing organizations.

BS: Some of them very sophisticated groups, but there are a lot of skills needed. We offer a full managed service, co-managed, and what we call platform services, which is our lightest service offering. We find most clients fall somewhere in the middle, in the co-managed version, where they pay for the platform service and an additional fee for an additional service when they need it. That works out well. Our business model is to charge technology fees to support our products. But the service offering is very important.

This segment is part 5 in the series : Thought Leaders in Big Data: Interview with Bill Simmons, CTO of DataXu
1 2 3 4 5 6 7