Are you doing Audience Targeted campaigns in programmatic?

Audience targeting has come up more and more often, recently. Advertisers are using "audience targeting" as the reason why they think they are not exposed to ad fraud and why they continue to pour money into digital ads (that are audience targeted). After all, they say, we uploaded our email list and are targeting our own, real customers, so how can there be any fraud? Or we supplied a list of doctors' NPI ("National Provider Identifier") numbers and our ads are targeted to individual doctor's cookies wherever they show up online, so how can there be any fraud? Or we are targeting "C-suite" users and even the office buildings where they work, so how can there be any fraud? All of the above advertisers also think that the higher click rates they are getting must mean that audience targeting is working really really well.

Ummm, sorry to burst your bubble. Let me explain why the above assumptions are not "on target" (pun intended) and why you are still exposed to ad fraud, in some cases all of it -- i.e. it's ALL fraud.

Audience segments are derived from anonymous cookie history

Let's start at the high level and then dig deeper. As far back as 2011, Nielsen was already reporting that only 5% of ads were even properly targeted (slide below). Out of 200 million impressions purchased with audience targeting, only 10 million reached the intended target. Nielsen has since removed that blog post and case study, because that reality would have hindered or spoiled the rampant growth of audience targeting revenues for adtech companies. Advertisers just had to believe it worked; that is great news for the ad tech companies offering audience segments for sale and no one would be the wiser.

Another study some ten years later from Neumann and Tucker confirmed that the data used in audience targeting is so inaccurate that even for a single targeting parameter -- e.g. gender -- the accuracy of the targeting (42%) was worse than random; and for two targeting parameters -- e.g. gender + age range -- the accuracy dropped to approximately 1 in 4 (24%). This is because even the most basic targeting parameters are approximated from website visitation patterns of anonymous cookies -- i.e. "anonymous cookie history." Unlike the walled gardens (Google, Facebook, Amazon) where users are logged-in all day long, ad tech companies can only harvest data on NON-logged-in users -- i.e. anonymous cookies -- that visit publishers' sites.

By stringing together the website visitation patterns of such anonymous cookies, data brokers try to approximate who they are and what they like; this is rarely right and often entirely wrong. What gender do you deduce from an anonymous cookie reading news on USAToday.com or shopping on Walmart.com? Of course these data brokers claim to do "error correction" and "data enrichment" and try to match cookies to the cookies that visited advertisers' sites and lists that advertisers upload. But only a fraction of those cookies "match" or "overlap." In other words, even if some match, that does not mean that the entire cookie pool is accurate.

If audience segments are so bad, then why does it "perform" so well?

The advertisers I talked with are quick to defend audience targeting, because they've paid for it for years and it would be embarrassing to realize or admit that it didn't work. The audience targeting "must be working well because we are getting much higher click rates" than if we just placed ads on publishers' sites. Well, it's nice that you got lots of clicks, but have you asked whether the clicks are from humans or bots? Do you even have the right tools in place to see for yourself? Google Analytics doesn't show you what's in the following chart. Your fraud verification vendor doesn't either. The following chart shows the clicks that arrived on the advertisers' landing pages, from "utm_source=programmatic." In case you didn't know already, in FouAnalytics, orange means declared bots and red means bad bots. Dark blue means human clicks. If you squint, you may be able to see the tiny blue slivers in these six donut charts. Between 1 - 4% of the clicks on these audience targeted campaigns are from humans. So what did you say about getting lots of clicks meaning audience targeting was working really well? You certainly got lots of clicks. [period].

I am also sure you remember when I told you that bots can easily pretend to be any audience segment you want to target. Some consumer-facing examples include the following: bots will deliberately look at swing sets in spring, to become part of the "swing set intenders" audience segment that advertisers want to target. The same bots will look at vacation and travel content sites in early summer and look at backpacks around back-to-school. This way they become part of the vacation-intender or back-to-school intender audience segments, respectively, that advertisers pay higher CPMs for. Bots also deliberately visit medical information or journal sites to pretend to be physicians, audience segments that pharmaceutical advertisers are desperate to target. Not only do bots cause the ads to load, they also click on the ads, which explains the high click rates you see in audience targeted campaigns.

Wait, where did my ads serve? Don't worry about it, it's all "targeted" down to the cookie

When you ask the ad tech vendor for a list of where your ads ran, and you might have seen something like the list in the following slide. Notice the casino, gambling, lotto, scratch-off, "smurfs," and criminal apps in the list. The ad tech vendors selling audience targeting services will be quick to tell you "don't worry, we are targeting the doctors by NPI number so even though these apps look strange, we assure you your ads are shown to the doctors you want to target." Really? What about the bots that pretend to be doctors by faking website visits to specific sites? What about the bots that copy off and replay a cookie you think is matched to a doctor based on NPI number? Right, you didn't account for that, did you? Further, even if the ad were targeted to the right doctor, would they be in the right mindset to see and click your ad when they are playing Subway Surfers or Candy Crush? No, they would not.

Finally, if we flip this argument around, we can ask the following: "if audience targeting worked so well why have I rarely seen ads that I considered relevant?" You may have seen "creepy" ads for the exact product you looked at on Amazon moments ago; but that is retargeting. One or more ad tech companies tracked your cookie looking at a product on Amazon or an advertiser's website, and showed you an ad for that product or advertiser moments later on a different website. That is retargeting not audience targeting. Retargeting is often perceived as creepy and is also often too late (I already bought the product). But it is a different issue than the crappy audience targeting being contemplated here.

"retargeting is creepy; audience targeting is crappy"

Approximations of approximations of possible cookies

Now let's get to the heart of why audience targeting does not work as well as advertisers assume.

How many marketers have challenged the claims of the ad tech vendors selling them the magical snake oil of audience targeting? I mean how many marketers have truly dug into the "how"? Exactly, technically, how are these vendors accomplishing this black magic? Well, you don't need to press them because I've done that for you. When pressed, hard, this is what they told me. The advertiser uploads a list of email addresses or NPI numbers that they want to target. The vendor then uploads these lists to one or more "onboarders" like LiveRamp. The onboarder -- i.e. data broker -- does some black magic called "cookie matching" to identify which anonymous cookies in their vast database may be related to the identifiers the advertiser supplied. The mental image of overlapping circles, like Venn diagrams, would be useful here. In some cases, you'll find that portions of two circles overlap; that's called the "match rate." If they are able to match cookies, then when certain anonymous cookies appear, they can be targeted with ads. That is the theory of how audience targeting should work.

The reality is that most of these cookie pools (circles) have little overlap with other cookie pools. And this has gotten even worse over time as browsers like Safari and Brave dump or expire cookies regularly -- e.g. every month, every day, or every time you close your browser. The cookies are not persistent enough to be found when doing cookie matches with onboarders. How often do you run the cookie matching? Right, not often enough. So don't assume that the cookie matching process will find the matches needed in each ad exchange to properly target your ads even if you uploaded email lists, NPI numbers, or other identifiers. The approximations of approximations of possible cookies means that most of your ad are still not targeted to the right audience segments, let alone individual cookies.

To sum up, just like there are fake sites, fake mobile apps, and fake users, there are fake audience segments created to absorb your money. They want you to believe the theory -- you think you are targeting audience segments or even individual cookies and your clicks are higher so you think it is working. But the reality is 1) bots pretending to be certain audience segments by visiting a select number of sites, 2) bots clicking on your ads to make "performance" look better, and 3) bots pretending to be individual doctors by copying and replaying their matched cookie.

So, "are you doing audience targeting?" Are you assuming that it works? Are you actually getting any real business outcomes from audience targeted campaigns? And should you look more closely into it now, and challenge your own assumptions on audience targeting, before your CFO or CEO asks you about it? And, now, does the masthead graphic make sense?

If you need help with this let me know. We can use FouAnalytics to check the clicks coming from these audience targeted campaigns and see if your campaigns are as red and orange as the 6 audience targeted campaigns in the slide above.

Continue reading: How to use FouAnalytics to Scrutinize Clicks from Programmatic Campaigns