Unraveling header bidding’s problems with user data

“There are some real security concerns about header bidding that aren’t being talked about,” said an ad-fraud researcher requesting anonymity.

Because waterfalling — a technique where publishers move inventory from one market to the next — is inefficient at driving revenue, many publishers adopted header bidding, which simultaneously offers inventory to multiple exchanges before making calls to their ad servers. While waterfalling is clunky, one benefit of this technique is that it limits how much user data the bidders could harvest, said Todd Garland, CEO of digital ad network BuySellAds.

For example, if the highest-bidding network in the waterfall wins 50 percent of the impressions in the auction, then subsequently, half of the impressions for sale won’t get passed along to other networks. With this approach, each network only obtains user data from whoever they sell impressions to, which is only a fraction of the total audience that the total impressions in the auction are sold to, Garland said.

But with header bidding having all calls go out simultaneously, each bidder can get access to data from all the users who were served impressions from the auction. A Rubicon Project spokesperson said, on average, the top 100 publishers in terms of comScore traffic have four header bidding partners. But it also isn’t unheard of for publishers to near double digits with their bidders.

“Many exchanges allow demand-side platforms to take bid requests and basically just ‘listen’ for data without spending money,” said independent ad tech consultant Brad Holcenberg. “So the more [bidding partners] you work with, the more likely the data gets to their customers and can leak out further from there.”

While sources did not share any first-hand experiences, data leaking can be problematic for both publishers and users. When data leaks, it can devalue publisher inventory because retargeters use audience data from premium publishers to target their users while they’re visiting websites that have lower CPMs. Hyper-targeting can also irritate users and turn them onto ad blocking. And in the most extreme cases, leaks of personal data can allow bad actors to hack people’s personal accounts and devices.

Jeremy Hlavacek, vp of programmatic at The Weather Company, said that while header bidding does make it easier for programmatic platforms to obtain user data, this effect is mitigated by the fact that data aggregators are connected to most supply-side platforms, so they can find indirect ways to obtain and sell user data even if waterfalling limits direct access to user data. But Hlavacek pointed out that with multiple partners bidding on all impressions available in the auction, header bidding significantly increases the amount of data points in the exchanges. Other sources said it’s this data deluge that is most problematic when it comes to security.

“Header bidding generates as much as 10 times the amount of calls to get the same result,” said Tom Kershaw, chief technology officer of Rubicon Project. “Whenever you generate that much noise, it is much harder to track [fraudulent activity] because it becomes easier for people to hide.”

Speaking on the condition of anonymity, an exec from an ad tracking firm said that data leakage is not affected by whether a header bidding product is browser-based or server-side. However, moving the auction to direct server connections could theoretically give publishers more control over who is bidding on their inventory.

“Regardless of the setup, it’s about your governance and having control over which vendors are involved with bidding,” the exec said. “It can get out of control if it is not managed properly.”