Predicting consumer responses to ads and helping marketers build winning campaigns is a major advantage to working with Big Data
Editor's Note: "Genome Decoded" is a series that tells the story of Genome, our data-driven audience-buying solution, from many angles, to bring you a little closer to Big Data.
Big Data gets bigger every day, feasting on an endless banquet of online actions and transactions. Rather than being overwhelmed, some experts are using a technique called predictive analytics to break down Big Data and accurately predict future events. In the 2012 election, ace statistician Nate Silver used Big Data and predictive analytics to correctly call the winning presidential candidate in all 50 states. The Genome from Yahoo! team uses the exact same approach to predict the future for its customers’ marketing campaigns, according to Kira Roytburg, director of Analytics for Genome.
Here’s how they do it: Analyze huge amounts of relevant data with sophisticated models, mix in human judgment, and continually tweak the models based on real-world results. Kira’s team can provide detailed analysis of every stage of a marketing campaign, from set-up to final results, but in this interview we focused mainly on Genome’s predictive analytics capabilities because, well, they’re so cool.
Yahoo! Ad Blog: What can predictive analytics do for your customers---and how do you do it, anyway?
Kira Roytburg: Simply put, we use this capability to predict a wide range of future outcomes for our advertisers’ campaigns. Those outcomes usually involve a customer action or conversion, like clicks, sales, site visits, answers to surveys, almost anything the customer wants to measure.
Advertisers use these insights to make more informed decisions about their campaign strategy, including target audiences, media placements, media channels, and creatives,. Our goal is to help them reach their campaign goals as quickly and efficiently as possible, and to minimize wasted time and expense.
YAB: Where do you begin with something like this?
KR: We usually start with the marketer’s goal, like running a campaign with a certain budget to maximize online purchases, and we build a data model designed to execute that campaign. Our models are built to handle tens of thousands of data points, and we use a lot: Yahoo! Data, the marketer’s own data, and information from leading third-party sources, as well. In fact, we have user data based on more than 700 million users a month, including how they respond to seeing banners, videos and other types of online ads, the sites they visit, the articles they read, plus their social and search data.
We analyze all that historical data to predict how responsive specific users will be to a particular ad or offer. And we use that data to select the optimal audience to reach the marketer’s goals. We not only predict how audiences react, we can also model outcomes for the most effective creatives and media placements, if that’s what the customer wants to know.
YAB: Once a campaign is off and running, is there any more need for predictive analytics?
KR: Yes. Once we’ve started, we fine-tune our campaign models every day. Models need to be updated all the time, because online user behavior never stops changing. And our campaign model is directly actionable, so when we adjust a prediction, we can execute that change on our model within 15 minutes.
YAB: So it’s not only about the amount of data you work with, it’s also how fast you can analyze and use it.
KR: Right. Traditional offline marketing models use a smaller, stable data set that’s geared toward describing an audience segment. Nothing really changes within any given day, week, or sometimes even months. For instance, a mail campaign might take several months between advertiser mailings to customers. No one wants to see an envelope with that offer every day (plus, it’s expensive).
But these days, online advertising doesn’t choose to use Big Data … it is Big Data. And with Big Data, you’re working with massive amounts of information that are always changing. You need to be able to predict your audience behavior as well as describe it. Success boils down to how fast you can extract relevant and useful insights from that data. And we’re fast, that’s for sure.
YAB: What’s the most important thing an advertiser should know about how predictive analytics can help them?
KR: They can gain a whole new level of visibility into what their target audiences really look like. That visibility will give them a clear understanding of the thinking behind our recommendations for audience segmentation and campaign optimization, and they can make more informed decisions on their campaigns. Plus, predictive analytics helps provide what we call the “elasticity of performance.” Usually, campaign performance goes down as scale is increased. But with the insights delivered by predictive analytics, you can increase scale without hurting performance.
YAB: What do you think sets Yahoo! apart in terms of predictive analytics?
KR: That’s a good question. I can’t think of any other company that has so much data, can build a model, deploy it, and measure its performance as quickly and accurately as we can. But I’d say our real differentiator is that we can deliver a very comprehensive set of analytics that spans the whole campaign process, from strategy creation all the way to execution and the measurement of final results.
We have very good algorithms and technology, but it takes more than that. You need smart people who can find creative solutions to help clients solve problems and seize opportunities. We work closely with the Genome Solutions group to understand both the science and the marketing.
