Marketing

IBM just released the results of its ad targeting bias audit

After analyzing the Ad Council’s massive Covid vaccination campaign, it found that ads showed a preference for certain age groups, education levels, and more.
article cover

SOPA/Getty Images

· 3 min read

Get marketing news you'll actually want to read

The email newsletter guaranteed to bring you the latest stories shaping the marketing and advertising world, like only the Brew can.

Yes, the algorithms marketers rely on to show ads probably have an unwanted bias. That’s according to IBM, which just released the findings of a five-month study on bias in advertising.

First announced in June, IBM partnered with the Ad Council to audit a subset of its massive Covid-19 vaccination campaign using an artificial intelligence tool that could help advertisers find bias in their digital targeting.

A reminder: Marketers want to reach relevant audiences, obviously. But, assumptions about the right audience can overlook potential customers. For example, nail polish has traditionally been marketed to women, despite a growing audience of men with actual style.

  • A study from a few years ago examined how algorithms can perpetuate biases in advertising; after finding that “people in wealthy areas responded more strongly to e-commerce discounts than those in poorer ones,” researchers explained that an algorithm could essentially take that information and run with it, offering “lower prices to higher-income individuals going forward.”

Here’s what IBM found:

  • The Ad Council’s test was set up to target “liberal- and conservative-leaning” designated market areas (DMAs), which were then divided into age groups. According to IBM, analyzed ads ran programmatically and across its owned properties, like The Weather Channel.
  • Pouring through 10 million impressions and 108 unique pieces of creative, the tool found that the campaign favored women, people aged 45–65, and those with incomes of less than $100,000.
  • It held a particular bias against those without a college education, meaning the model didn’t try and assemble the right creative to “convert” that group, or to get lower-educated audiences to click on advertisements.

Though IBM says it doesn’t exactly know why this particular campaign was biased, it did show that specific audiences were underrepresented. “The question becomes…what’s happening elsewhere in the campaign that we’re missing,” Robert Redmond, head of AI ad product design at IBM, said.

Takeaway: IBM’s report concludes that “bias can exist in the data and algorithms that are employed for digital advertising, and that bias is not always immediately observable to the human eye…there is more work to do to understand which external forces drive these inequalities.”

In the meantime, IBM is looking for more campaigns to audit for further research—the tool isn’t yet available for use.—RB

Get marketing news you'll actually want to read

The email newsletter guaranteed to bring you the latest stories shaping the marketing and advertising world, like only the Brew can.