Advertise Smarter Online

Women Leading Research: Courtney Paulson

SMITH BRAIN TRUST – For firms advertising online — for example, a luxury vacation company — is it better to target audiences with specific interests or aim for the widest possible reach? That’s a question companies are grappling with more and more as they migrate their advertising budgets to online ad campaigns. 

In new research, the Smith School’s Courtney Paulson and two co-authors develop a method to compute how to optimize both goals — pinpointing people with specific interests, while also targeting a general audience.

For the cruise line company used as a case study for the research, the untargeted consumer was indeed an important source of revenue. The power of suggestion from advertising appears to have an important persuasive effect, says Paulson, assistant professor in the department of Decision, Operations and Information Technologies at the University of Maryland’s Robert H. Smith School of Business.

“Those [untargeted] people will see the ad, in some cases, and say, ‘Oh, I would love to take a cruise. I hadn’t thought about that, but maybe I should take a cruise to Cancun instead of flying,’” Paulson says. “You want to try to reach as many of those people as possible, as long as you’re not losing that target group that you think you are the most interested in.” 

The method, which involves incorporating multiple constraints, proved effective at balancing those competing interests – hitting as many people as possible while specifically targeting the people that the advertiser cares most about.

“Internet advertising can be a high-dimensional problem, since each website represents a unique advertising opportunity,” says Paulson.

In their research, Paulson and co-authors Gareth M. James and Paat Rusmevichientong, both from the University of Southern California, considered the general constrained high-dimensional problem, where underlying parameters satisfy a collection of linear constraints. They developed a Penalized and Constrained (PAC) regression method to compute the penalized coefficient paths on high-dimensional linear and nonlinear fits, across that set of linear constraints.

The researchers conducted extensive simulations and illustrated some of the many applications where optimization problems are demonstrated. Among them was this common marketing challenge: optimizing various metrics, such as reach or click-through rate, across thousands of Internet websites, subject to various budget constraints.

They crafted what they describe as “an elegant and efficient algorithm” for solving the penalized and constrained problem. “Our algorithm is both simple to implement, because it can be applied in conjunction with standard optimization methods, and much faster than standard approaches such as quadratic programming,” the researchers say. 

The research builds upon Paulson and James’s earlier work, which explored the possibility of developing a model that took internet advertising bidding into account, to help advertisers decide how much money to spend on each website for optimal reach.

For decades, companies have relied on well-established marketing metrics such as reach, the fraction of potential customers exposed to a firm’s ad. But today, they might instead consider more specific metrics, such as click-through rates. That change, however, requires new, more flexible budget allocation processes, says Paulson, both because of the many different metrics and to handle common ad campaign constraints.

In the researchers’ latest research, they examined a case study of the constrained internet marketing problem which uses a unique and proprietary comScore Media Metrix dataset, with anonymously recorded webpage usage information from a panel of 100,000 Internet users for an entire year. The researchers chose a one-month subset of the data, examining how often people clicked through an ad on a particular website.

They created a binomial model, which assigned a certain probability of whether a consumer would see or not see an ad. “Basically, there is a coin flip,” Paulson says. “If this advertiser has made a bid on the ad and won it, then their ad will appear to me. If not, it won’t. We formulated our model based on that.”

Then, on top of that layer of probability, there is a probability that the consumer might actually click on the ad. “Maybe it is an ad that is particularly targeted to that consumer; something the consumer is particularly interested in. Most of the ads that pop up in front of people, they just scroll past.”

But the people who actually click on an ad are theoretically far more valuable as customers than the people who just let the ad scroll past. 

As advertisers add constraints to the equation, they do naturally decrease their overall reach, the research finds. However, they do have greater click-through success with the subset of people they are actually targeting. 

And, Paulson says, the sacrifice is likely worth it. After all, she says, they’re not giving up entirely on the general online audience.

“That would be too big a sacrifice, basically giving up completely the chance of reaching the unknown, untargeted customer,” she says. “There are always going to be people that you won’t know to target.”

Read more: James, G., Paulson, C. and Rusmevichientong, P. “Penalized and Constrained Regression.” 2018.

Courtney Paulson is an assistant professor of Decision, Operations and Information Technologies at the University of Maryland’s Robert H. Smith School of Business. She received her B.S. in Statistics from the University of Central Florida in 2011 and her Ph.D. in Business Administration (Statistics) from the University of Southern California in 2016. 

Research interests: Algorithm development, constrained optimization models, and systems architecture. Her work focuses on interdisciplinary data analytics applications, which has led to projects with researchers across domains such as marketing, operations, engineering, and systems analysis. 

Selected accomplishments: Paulson’s work has been recognized with awards from statistical organizations, including the American Statistical Association, and organizations outside the statistics discipline, including the INFORMS Society of Marketing Science, which recently honored her with the 2015-2016 ISMS Doctoral Dissertation Award. She’s also a one-time champion of the TV game show “Jeopardy!”

About this series: The Smith School faculty is celebrating Women’s History Month 2018 in partnership with ADVANCE, an initiative to transform the University of Maryland by investing in a culture of inclusive excellence. Daily faculty spotlights support activities from the school’s Office of Diversity Initiatives, starting with the seventh annual Women Leading Women forum on March 1, 2018.

Other fearless ideas from:  Rajshree Agarwal  |  Ritu Agarwal  |  T. Leigh Anenson  |  Kathryn M. Bartol  |  Christine Beckman  |  Margrét Bjarnadóttir  |  M. Cecilia Bustamante  |  Jessica M. Clark  |  Rellie Derfler-Rozin  |  Waverly Ding  |  Wedad J. Elmaghraby  |  Rosellina Ferraro  |  Rebecca Hann  | Amna Kirmani  |  Hanna Lee  |  Hui Liao  |  Jennifer Carson Marr  |  Wendy W. Moe  |  Courtney Paulson  |  Louiqa Raschid  |  Rebecca Ratner  |  Debra L. Shapiro  |  M. Susan Taylor  |  Niratcha (Grace) Tungtisanont  |  Vijaya Venkataramani  |  Janet Wagner  |  Yajin Wang  |  Yajun Wang  |  Liu Yang  |  Jie Zhang  |  Lingling Zhang

Illustration credit: frender