Share This Page

A Research-Based Method for Improving the Effectiveness of Online Ads

by Rob Kelly Tuesday, May 5, 2015

Measuring online display ads is difficult, and misleading results can cause companies to spend money on ads that don’t work. That’s why Paul Hoban, assistant professor of marketing at the Wisconsin School of Business at the University of Wisconsin–Madison, investigated and identified practical solutions for measuring and improving the effectiveness of digital advertising, which he presented to alumni and business leaders at the recent Alumni and Friends Summit hosted by the School’s A.C. Nielsen Center for Marketing Research.

Current best practices in digital ad targeting and measurement create challenges for managers looking to optimize campaign effectiveness. Typically, digital advertising platforms use algorithms based on a large number of consumer characteristics to target specific audiences. Because these algorithms are proprietary, the company offering the product or service has little precise control over when and to whom an ad is served.

 Paul Hoban, assistant professor of marketing, Wisconsin School of Business, UW-Madison

Paul Hoban, assistant professor of marketing at the Wisconsin School of Business, explains a model-enhanced experimental approach to measuring display ad effectiveness at the Alumni and Friends Summit hosted by the School’s A.C. Nielsen Center for Marketing Research.

Second, digital ads are frequently evaluated based on a “last click” or “last view” attribution. This means that the last ad to which an individual is exposed prior to visiting a site or making a purchase receives all the credit for that activity. In response, ad servers have developed targeting algorithms focused on finding people who are very likely to undertake a given activity. Thus, many of the people targeted have already decided to make a purchase before seeing the ad, which means that the ad is inaccurately credited with convincing them to buy.

Because of these (and other) shortcomings, Hoban and Randolph E. Bucklin, professor of marketing at the University of California–Los Angeles, developed and tested a model-enhanced experimental approach to measuring display ad effectiveness that allows managers to quantify the lift caused by the ad while keeping the control group (and associated cost) quite small. Further, it relies only on data that is already available to firms. To gain additional insight, they broke their analysis out across the purchase funnel.

The experiment found that the ads were least effective on people who had previously visited the site but had not created an account or completed a transaction, which is where companies typically spend most of their advertising money.

“These are individuals who went to the site but didn’t do the thing you wanted, so now we’re going to inundate them with display ads, hoping they’ll do the thing you wanted them to do. In this case, the ads had no effect whatsoever,” Hoban says. “These are individuals who were exposed to your products or services, got there, and said, ‘I don’t want this,’ and left. The display ad is trying to persuade them that they made the wrong choice. That’s hard to do.”

Rather than spending money on trying to convince visitors to come back, Hoban determined that it’s better to reallocate more advertising resources to targeting an audience that is more likely to take action as a result of the ad—users who have already created an account.

Read the full paper, "Effects of Internet Display Advertising in the Purchase Funnel: Model-Based Insights from a Randomized Field Experiment," published in the Journal of Marketing Research in 2014.