When a state-level government agency in the United States set out to advertise grant programs and consulting services for entrepreneurs, it wanted to reach everyone who might benefit—including women, who have historically been underrepresented among business owners. Yet precision ad targeting, it turns out, can unintentionally leave some of the people who need help most out in the cold.
Researchers at Cornell Tech have developed a method to fix that. Isabel Corpus, a doctoral student in information science, and Allison Koenecke, an assistant professor at Cornell Tech and the Cornell Ann S. Bowers College of Computing and Information Science, have created a multi-campaign approach that reduces "skew"—the underdelivery of ads to certain demographic groups—while remaining cost-effective. Their work was inspired by a troubling reality: online advertisers often miss their mark, and when government programs meant to help people fall short, it can amount to inequity in who gets access to vital resources.
The problem lies in how advertising platforms categorize users. Google, for instance, offers advertisers three gender labels: male, female, and unknown. Using only the first two leaves out a measurable chunk of potential targets. "Unknown" users tend to be people with low socioeconomic status or nonbinary gender identities—groups already facing systemic disadvantages. "As a government advertiser," Corpus said, "you're not looking to exclude anyone from potentially being shown these resources."
To address this, Corpus and Koenecke designed a campaign with four target audiences: male; female; male and unknown; and female and unknown. Their approach split the advertising into two waves, first targeting "female" and "male plus unknown" users, then pivoting to "male" and "female plus unknown." The result was a measurable reduction in skew and greater cost-effectiveness compared with simpler budget-splitting methods.
The stakes are significant. According to a 2023 Census Bureau report, just 39 percent of the 36.4 million U.S. businesses were owned by women—a disparity that targeted government advertising could help narrow, if it actually reaches the intended audience. The researchers will present their findings at the ACM Conference on Fairness, Accountability, and Transparency (FAccT '26) in Montreal this June.
Koenecke emphasized that algorithmic tools have a long history of biased distribution, whether intentional or not. But she sees reason for optimism. "Increasingly, organizations are realizing the value in quantifying these potentially hidden biases and taking steps to ameliorate them," she said. The hope is that this method could eventually extend beyond a single agency, helping ensure that public resources reach the people they're designed for—no matter their gender, income level, or where they fall in an algorithm's blind spot.