The move, lauded as a win for users’ privacy, ultimately cost big tech platforms billions of dollars.
But when Kellogg marketing professor Anna Tuchman examined the issue, she found very few studies that showed the effect on businesses who advertised on these apps.
In fact, no one had actually quantified the value of user data—such as browsing activity and online purchases—that is tracked and shared across applications to help deliver ads on many major advertising platforms. This type of data is called “offsite data,” because it’s used to target ads on a particular platform but is generated off the platform.
So how effective was offsite data in helping to harness new customers through advertising, and if it went away altogether, how much would it cost businesses?
To study the issue, she teamed up with Nils Wernerfelt, an economist at Meta who will join Kellogg as an assistant professor this fall, along with Robert Moakler of Meta and Bradley Shapiro of the University of Chicago.
The team conducted a large-scale experiment with more than 70,000 advertisers on Facebook and Instagram to find out just how much companies benefit from using offsite data to deliver targeted ads.
They found that removing the ability to use offsite data would result in a 35 percent increase in the cost to gain each new customer—and would disproportionately hurt smaller advertisers.
“A lot of the discussion around privacy regulation has been focused on the consumer side, and for good reason,” Tuchman says. “But we found that removing offsite data makes it harder for smaller advertisers to acquire new customers and compete with incumbent firms. That could have implications for entry into the market, and it could ultimately have downstream negative consequences for consumers if that leads to less variety and higher prices.”
Helping advertisers find the right customers
Digital advertising now constitutes a majority of advertising spending, precisely because businesses can tap detailed user data to match ads to potential consumers. To get the data from other websites that provides the basis for this matching—called offsite data—social-media platforms encourage businesses to install “pixels”: small pieces of code on their websites that collect information about users and send them to platforms for use in ad delivery. For example, if a customer buys shoes on a retailer’s website that has a Meta pixel installed, that purchase information would be sent over to Meta. This helps the platform understand who the retailer’s actual customers are and helps it target ads for that business.
When businesses create ad campaigns on Meta, they set a specific objective, such as driving purchases. Meta then uses offsite data in its algorithms to show ads to people who are most likely to purchase that product. For example, if pixel data indicates a product is purchased only by women, the platform’s algorithm knows not to show ads for it to men.
“Advertisers often don’t have a clear sense of exactly who they should target,” Wernerfelt says. “One of the great powers of digital advertising is that Meta and other platforms can dynamically show ads to people, see who is buying the product, and update the target audience.”
Pixels are ubiquitous. Most social-media platforms have variants of pixels that businesses can install, and millions of companies use them. But many consumers have voiced concerns over their information being tracked and exchanged in such ways. Both companies and governments have responded with new policies and regulations. Apple’s “Ask App Not to Track” feature made it easier for users to opt out of being tracked by such methods. The European Union’s General Data Protection Regulation law regulates such personal data collection, and countries around the world are considering similar policies.
“The digital advertising ecosystem has grown so rapidly that it’s ahead of regulation,” Wernerfelt says. “But regulations are being put in place with very little evidence to inform trade-offs. We wanted to provide one piece of the puzzle that shows the cost on the advertising side.”
Estimating the value of offsite data
To find the value of offsite data, the team conducted a large-scale, randomized experiment with more than 70,000 advertisers on Facebook and Instagram over one week in the fall of 2021. For each campaign, the team randomly chose 5 percent of the target audience and allocated 5 percent of the campaign’s budget toward it. Within that 5 percent, the team randomly withheld advertisements from 10 percent of users who would normally see those ads. They then compared outcomes from those who saw the ad with those who didn’t.
From that data, they could estimate the cost per incremental (new) customer for each campaign. The results were wide-ranging: while the median cost per incremental customer was about $42, it ranged from $5 to more than $200 for campaigns between the 90th and 10th percentile of effectiveness. That’s because the campaigns ran across verticals, from e-commerce to consumer-product goods, in countries around the world.
“This experiment was on a massive scale and really allowed us to feel like we were getting a hold on ad effectiveness across industries,” Tuchman says.
With this data in hand, the team wondered what would happen if offsite data went away. With algorithms unable to access browsing behavior and previous purchases, ads would likely be optimized for clicks. That means they would be shown to users who are most likely to click on the ad, rather than those who are most likely to buy the product.
“But when you show ads to people who simply click a lot, you’re missing out on a lot of customers,” Wernerfelt says.
To determine the precise effect of this change, they repeated their previous experiment—only instead of optimizing ad delivery using offsite data, they targeted ads from the same campaigns at users who would be most likely to click on them.
If advertisers were to optimize for clicks instead of purchases, the team estimated that the median cost per incremental customer would rise to $57—about a 35 percent increase from using offsite data. The median total ad spend for the week was about $1,250. That means instead of gaining 30 new customers that week, the companies would only gain 22 new customers.
“We showed that losing this data would have a big effect,” Wernerfelt says. “And it affected the vast majority of advertisers negatively.”
But one group was affected more than others: smaller advertisers. The team’s analysis showed that the effect of losing offsite data was five times greater for these advertisers versus larger ones. That’s because small advertisers are often little-known and rely on such offsite data and ad targeting to bring in niche incremental customers.
Targeted advertising: good for small advertisers and consumers?
The team also looked at purchase data six months after the experiments to better understand the long-term effect on customers. They found that ads that use offsite data to target customers are much more effective in bringing in repeat customers than click-optimized ads.
This finding suggests that offsite data may have benefits for both advertisers, who prefer long-term customers, as well as for consumers, Wernerfelt says.
“There is a concern that these ads might be persuading people to buy products they don’t need, or that they will regret later,” he says. “But if consumers are still buying this product in the future, it seems likely they are benefitting and deriving utility from these products.”
These results show that policies regarding tracking consumer data should be nuanced and thoughtful, the researchers say.
“You can protect privacy in this manner, but it might have real costs on the advertiser side,” Wernerfelt says. “There’s no privacy ‘free lunch’ here.”
To protect small businesses and ensure a competitive environment, perhaps platforms, advertisers, and users could find a way to allow advertisers to use the data while compensating individuals for any privacy losses. Some companies are already conducting research into privacy-preserving ad delivery and measurement solutions, as well.
And digital advertising is not just limited to for-profit companies. Wernerfelt has also studied the effectiveness of digital advertising for public-health interventions.
“This is a fascinating area with such far-reaching implications,” he says. “There’s a lot of excitement, but there are also a lot of unanswered questions, and privacy regulation is high among those.”