Facebook is constantly experimenting with consumers, and even its creators don't fully understand how it works. Relying on very precise knowledge of the preferences of small groups can also have adverse effects on advertisers, reports the International Journal of Research in Marketing.
Users of social media platforms like Facebook, Instagram, and TikTok may think they’re simply interacting with friends, family, and followers while also viewing ads. But according to research from researchers at the Sauder School of Business at the University of British Columbia (UBC) in Canada, these interactions are part of ongoing marketing experiments that are often not fully understood even by the companies behind them.
The authors analyzed all known published, peer-reviewed studies on Facebook and Google’s use of A/B testing—when different consumers are shown different ads to determine which ones are most effective.
Study co-authors Dr Yann Cornil and Dr David Hardisty say that at any given moment, billions of social media users are being tested to see what they click on and, most importantly for marketers, what they buy. You might think that advertisers are testing which messages are effective and which are not; but as it turns out, it is not that simple.
By using A/B testing on Facebook, you can reach a huge audience and observe real behavior – and because participants are unaware they are part of an experiment, their responses are perceived as more authentic and trustworthy.
The problem is that highly complex algorithms decide which consumers will be shown different content and ads; as a result, no one—not even those who created the algorithms—can fully understand why specific consumers were targeted by an ad, or determine why some of them chose to click on it. Dr. Cornil says this comes down to a lack of something called “random assignment”—for example, when experimenters randomly present two different ads to selected groups.
“You can’t say that changes made to an ad increase click-through rates because each ad will have an algorithm that selects participants who are most likely to click on it. If the algorithms are different, that means there is no true random assignment,” he said. “It also means we can’t say with certainty that an ad generated a higher click-through rate because it’s a better creative ad. It could be because it’s associated with a better algorithm,” he added.
What's more, people are often shown ads based on their search history, but if they have already decided on a particular product and then an algorithm shows them an ad for that product, researchers may wrongly conclude that the ad prompted them to make a purchase.
“(AI) selects people not only based on observable things like age, gender or location that we can easily know, but also on unobservable things like past behavior, interests, and even parameters that Facebook itself can’t quantify because those are determined by machine learning and AI,” Dr. Hardisty said. He added that the target groups may seem similar in some ways, but the algorithm may have selected them for completely different reasons.
“It’s basically a complicated model that somehow figured out that a certain type of person — we don’t know which — is more likely to click. So even if we asked the people at Facebook, ‘Why did they pick this group of people?’ they wouldn’t know the answer,” he said.
This matters for a variety of reasons. First, many marketers rely on Facebook A/B testing to determine what to advertise and how; but perhaps even more importantly, different audience segments can be left out of important information, which can reinforce divisions.
“There is one article that explains why women are not targeted in STEM (Science, Technology, Engineering, Mathematics) education ads solely because of algorithms,” Dr. Cornil said.
“Women are more expensive to target (targeting is precisely directing an advertising message to a specific group of recipients – PAP) on social media, and these algorithms will try to generate as many clicks as possible at the lowest cost. So if women are too expensive to target for STEM education, then they are not targeted,” the researcher described.
What's more, algorithms amplify what works and what doesn't, so if women don't click on certain ads, they'll be even less exposed to them.
The UBC study focused on Facebook and Google, but the authors believe that all major social media platforms, from Instagram to TikTok, have similar practices. At a conference, a Facebook employee once told Dr. Hardisty that at any given moment, every Facebook user is an unwitting participant in an average of 10 different experiments. With the advent of AI-generated content and ads, that number is almost certainly growing.
The authors warn that marketers should be careful not to read too far-reaching implications into the results of Facebook A/B tests.
“If you have an ad that’s getting a lot more clicks, it could be because Facebook has managed to identify a small, specific group of people who really like it,” Dr. Hardisty said. “And if you change your entire product line or campaign to match that, that could be off-putting to most people. So you have to be very careful not to draw broad conclusions from one Facebook study,” he said.
Dr. Cornil added that the algorithms used are so complex and precise that social media platforms can “microtarget” people down to the individual level: “It’s about choosing the best possible ads for a certain segment—and that segment isn’t even a group of people.” So the advertiser is unaware of what the AI really knows.
Paweł Wernicki (PAP)
pmw/ bar/
