Thumbnail

Opening the Black Box

USC Marshall Researchers Develop First Public Release of High-Speed Approach to Large Scale Internet Advertising

October 11, 2018
Thumbnail

These days, when companies decide to advertise on the internet, they generally seek guidance from consultants who help them discern optimal sites and how much money to spend. Agencies take a set of requirements—ad dates, budget—and run it through, “…black box-like software that has group-based algorithms,” said Lan Luo, USC Marshall associate professor of marketing. “They help the client, but their method is secret.”

Now Luo and her research partners, Gareth James and Courtney Paulson, have published a paper in the Journal of Marketing Research (JMR) that provides a nonproprietary high-dimensional algorithm for targeting ad campaigns. The methodology is not only openly available, it evaluates 100 times the number of websites as current open source analytics programs.

“In marketing,” said James, professor of data sciences and operations at Marshall, “they hadn’t been able to optimize over more than a few websites at a time. This program can scan tens of thousands of sites in the same timeframe as it would take to search 10 sites using currently available methods.

“We have opened the black box." -- Lan Luo, associate professor of marketing

“This is a real advance and it’s not just the algorithm,” he said. “The key to our approach is that our criterion can easily and efficiently be optimized over thousands of websites.”

The paper, Efficient Large-Scale Internet Media Selection Optimization for Online Display Advertising,” was published in the August issue of JMR. Software to implement its methodology has been designed by Paulson, a former DSO doctoral student, now assistant professor at the University of Maryland’s Robert H. Smith School of Business, and is publically available at the CRAN software site.

“We have opened the black box,” said Luo.

More than Marketing

James, the E. Morgan Stanley Chair in Business Administration, is an expert in statistical methodology. His work has been particularly useful in marketing settings, and he knew as he developed the high-dimensional function in this algorithm that it would lend itself to Luo’s work in product development.

Together they created a methodology that not only has the potential to advance online marketing strategy, it can add value in other high-dimensional contexts—the subject of the team’s next paper.

“Our method can take unique requirements into consideration,” said Luo. “Within the context of programmatic advertising, the algorithm is a model of flexibility, efficiency and functionality.”

In their paper, the team focuses on programmatic ad buying with real-time bidding, the system that currently comprises about 50 percent of the U.S. display ad market. Their methodology provides prescriptive bidding guidelines for campaigns involving a large number of websites. “This is the first opportunity to see how to use this high-dimensional optimization in practice because our code is nonproprietary,” said James.

“The algorithm hand picks the right places to advertise for the best results,” said Luo. “This solves the problem of choosing where to place an online display ad among thousands of websites. The impact of this code on efficiency is huge.”

The code has been developed into software that campaign managers can calibrate prior to placing bids. Based on results from the optimization, they can then decide where to place bids, as well as the price they should bid in a real-time setting.

“Because our optimization runs efficiently over a range of budgets, campaign managers can use it to determine an optimal campaign budget,” said James. “It can accommodate common internet display advertising considerations and match them with content websites. We believe our code and software will add considerable valuable to internet display ad campaign managers.”