“The Sell Sider” is a column written by the sell side of the digital media community.
Today’s column is by Paul Bannister, co-founder and executive vice president at CafeMedia.
Does behavioral targeting make publishers more money? That’s a huge question with many potential ramifications, most significantly the conversations about government regulation of user tracking.
A study released last week examined that very question. The main takeaway: Behavioral targeting doesn’t earn publishers very much money at all, and any gains are likely taken by ad tech firms.
The researchers looked for the presence of a cookie and then controlled for a number of other factors (device type, geography, etc.) to see how cookies contribute to increased earnings (higher CPM) for the publisher. This analysis was based on data from a single large publisher, from a single week in May 2016. The authors used some sophisticated statistical techniques to try to ensure that the comparison was fair. They determined that behaviorally targeted ads – or, more accurately, the presence of cookies – contributed to 4% higher CPMs.
It makes for great headlines, but I don’t believe the study was constructed to be able to reach this conclusion. I saw several fundamental problems with the data set and with how the analysis meshes with the realities of digital advertising.
Soon after the study was released I wrote a thread of tweets addressing the shortcomings and wanted to share a more thoughtful analysis.
The flaws
The study’s authors are aware of some of its limitations and clearly state that their study doesn’t quite answer the behavioral targeting question. They acknowledge (page 28) that they studied whether the presence of a cookie affected value generated for publishers, though they believe there is a correlation between the presence of a cookie and behavioral targeting, which is an oversimplification.
They also note (page 29) that the data comes from only one company and can’t necessarily be extended to other companies – both are fair points. But there are other issues that go unstated.
First, the data set is from May 2016 – three years ago. There has been tremendous change in the industry over the last three years, including Safari’s Intelligent Tracking Prevention, the EU’s General Data Protection Regulation, the migration to first-price auctions, overall programmatic growth of more than 50%, the rise of header bidding, exchange bidding and countless other shifts that impact cookies, targeting and demand. It’s hard to see how data preceding many of those innovations has any applicability in today’s world.
The study’s authors share very little about the setup of the publisher in the study, inevitably raising questions. How are their floors and blocks set up? What header bidding partners were they running? Did the study look at their revenue from AdX and also from header bidder exchanges? Is this publisher running on a Google stack? Does the publisher have direct deals running at the same time? How were unfilled impressions counted or were they not tracked at all? How was the data set filtered?
The answers to these and many other questions could have a huge impact on how the findings should be interpreted and the ability to replicate the study.
What kind of cookie?
Finally, and most importantly, this study is all about the presence of cookies yet makes no mention of first- vs. third-party cookies. In practice, these two types of cookies are extremely different. The study talks about using the unnamed publisher’s data to see “whether the user’s cookie ID was available” (page 13). Since this uses the publisher’s data, I presume the only cookie that would have been available is the first-party cookie of the website’s domain.
There’s no straightforward way for a publisher to see whether a browser accepts third-party cookies or not. You can implement a convoluted two-step process to confirm whether third-party cookies are enabled, but it’s very nonstandard, and since it’s not documented in this research, I believe it’s safe to assume it was not implemented to create this data set. Even if it were done for this research, there’s no way to check for the presence of a particular third-party cookie – you can only check for the existence of cookies for domains that you have access to, and publishers don’t have access to the domains of the third-party firms they work with, directly or indirectly.
Since the researchers were likely looking at first-party cookies, there cannot be any connection between those cookies and advertisers’ behavioral targeting: none at all. The presence of a first-party cookie would show whether the user has visited the site before. There’s no relationship between a user having been on this site before and whether an advertiser would have a cookie (and any inherent behavioral targeting value) for that user or not. Hence, it’s not a surprise that the study does not find a sizable difference in the value of users with first-party cookies than without.
Why does any of this matter?
There are many regulatory discussions around the world about privacy, behavioral advertising, the growing dominance of the duopoly/triopoly and related topics. In-depth research studies like this one will be used by governments to make decisions about the future of our industry, and we want our governments to make decisions based on the best information possible.
If the interpretation of this study is inaccurate and contributes to arguments to completely shut down behavioral targeting in the United States or another country, it could have a negative impact on publishers. Already struggling companies would be hit with another loss of revenue. This would have particularly damaging effects on small- and medium-sized publishers that are the most dependent on programmatic advertising that heavily relies on behavioral targeting.
This would also likely continue to cement the dominance of the walled gardens, which have their own internal targeting systems and don’t need to work with any third parties.
I’m not making the case that behavioral advertising is good, nor am I making the case that ad tech firms aren’t taking an outsized cut of the growth in digital advertising. We just need better research and data that can support or refute these points and lead to the right outcome for the industry and ultimately, for consumers.
Antonio García Martínez, of Wired Magazine and previously Facebook, wrote a great Twitter thread on this topic and summed it up with the excellent point: “Do the [behaviorally targeted] ads in fact ‘pay for the internet’ (as is often repeated)”?
If behaviorally targeted ads are creepy to users and don’t “pay for the internet,” then they’re likely doomed by regulators. We must ensure that we use accurate facts to make decisions that produce far-reaching consequences.
Follow Paul Bannister (@pbannist), CafeMedia (@CafeMedia_) and AdExchanger (@adexchanger) on Twitter.
