Custom Audiences & Unintended Consequences: How Misused Data Products Can Hurt Democracy

Our third podcast (Data Brokers & Democracy) explores some of the ways data brokers can impact democracy. As of early 2018, important details are emerging about how certain organizations used data gathered from data brokers combined with illegally obtained Facebook data to micro-target Americans with various messages during the 2016 U.S. presidential campaign.

This post touches on a somewhat different portion the 2016 election story: Russian interference in the 2016 U.S. election; but even more so, the technology that allowed it to happen. To fully understand what Russian state actors were up to during the 2016 campaign, we have to jump back a few years earlier, when Facebook launched a product known as Custom Audiences. This is a cautionary tale about the importance of considering how data-based products and services can be manipulated to have an effect well beyond the creator’s intention. Facebook formally launched Custom Audiences in September 2012. At the time, it was a marketer’s dream come true. Prior to its arrival, marketers had several methods of targeting Facebook users with ads, but they were somewhat limited. For instance, one could display ads to all

Facebook users who had “liked” one’s page. One could also display ads to individuals based on characteristics such as age, gender, location, and interests. But if you wanted to target people who had already made a purchase from your store in the past, and only those people, you were out of luck. Custom Audiences promised to close that gap.

Social media companies fall outside of our data broker definition for the purposes of this project, but in a lot of ways, social media and data brokers are connected. For instance, in recent years, Facebook forged partnerships with some of the world’s biggest data brokers to help advertisers further refine their targeting on Facebook (a practice it recently halted). Snapchat lists eight “Audience Match Partners” on its site. But social media companies also offer services that closely resemble those of data brokers. Many are nearly identical.

Custom Audiences is one such service. It is an idea lifted directly from the data broker industry. In data broker-speak, the concept is known as “onboarding.” Onboarding allows marketers to reach offline customers or contacts online. To “onboard,” a marketer provides their customer list to a data broker. The broker then locates those customers online through their various partnerships with other brokers and websites. Once located, the broker can serve the customers with ads as they browse the web. Custom Audiences is, in essence, a self-service onboarding tool. Advertisers upload a client email list to Facebook’s ads interface and Facebook finds those clients among its users. Facebook then groups those users into a Custom Audience that the marketer can serve with ads.

As the years went on, Facebook added new ways of creating Custom Audiences. One method allows marketers to upload Facebook code to their website. The code recognizes the Facebook users among a site’s visitors and adds them to a Custom Audience. Marketers can even create Custom Audiences for specific pages within their sites. Marketers can also expand the reach of their message by creating a “Lookalike Audience.” Lookalike Audiences are groups of Facebook users who share characteristics with a Custom Audience. This feature is useful for marketers whose Custom Audiences have a smaller-than-desired reach, or marketers who want to expand into new geographic areas.

As we are now learning, Custom Audiences was a weapon of choice used by Russian state actors during the 2016 U.S. election campaign. The Russians used it to repeatedly send messages to American voters identified as being “susceptible to propaganda.” According to the Washington Post and other outlets, Russian operatives set up a network of sites and Facebook Pages resembling those of U.S. political activists to influence voters’ political behaviour. When a Facebook user landed on one of these sites, they were added to a Custom Audience that allowed operatives to target those users with ads and messages across Facebook’s news feed and vast ad network. When the user clicked through an ad on Facebook’s network, they were sent to an external site with more Facebook Ads code and tracking tools. This allowed operatives to further refine their targeting and messaging to certain American voters. Those viewing the ads were never aware they had been “singled out or that the ads came from Russians.”

Custom Audiences and Lookalike Audiences are powerful marketing tools. They facilitate hyper-targeted marketing to Facebook users based on known interests, affiliations, habits, and actions. But when Facebook unleashed the power of Custom Audiences on the marketing world, it does not seem they gave much thought to how people might abuse the technology, let alone how it could be used to interfere with democratic elections.

It is still unclear to what extent Russia swayed the results of the 2016 U.S. election, but one thing is clear: data-based services enabled the Russians to have an outsized impact on the messages Americans saw in the lead-up to the election and its aftermath. Those who develop and regulate social media and other online networks failed to anticipate how these services could be misused in the elections context.

For its part, Facebook is now undertaking a transparency initiative to allow users to know more about who is advertising to them and why they see certain ads. But maybe it’s also time for regulators to consider reasonable limits on how and what data may be used to target Canadians with advertising on social media and other networks. Our democracy may depend on it.