Setting Realistic Expectations For Targeted Campaign Delivery

  • by , Op-Ed Contributor, December 15, 2009
On Dec. 7, I had the opportunity to participate in a panel on the future of online advertising and measurement at the UBS Media Conference.

One of the key points of debate and discussion at the session was the seemingly irreversible trend toward real-time, exchange-based, impression-by-impression transaction of advertising -- essentially moving the buying and selling of advertising to a bid-and-ask type of system.

Let me talk for a moment about the theory of technological inevitability. It's my theory, so don't go looking it up on Wikipedia. The theory states that when technology makes a development possible, that development becomes inevitable -- regardless of customer demand or marketplace impact.

In 1938, Enrico Fermi published a paper on nuclear fusion. In the U.S., scientists immediately recognized the technological implications: a terrible bomb could now be built. The fact of that possibility made the development of the bomb technologically inevitable. There was no "customer demand" for a terrible killing technology. But of course we had to build it, because we knew that across the ocean the other guys would be.



And so I wonder if the direction that ad targeting is taking online isn't akin in some way to the development of the bomb. We have a tendency on the Internet to believe that -- because we are, after all, the most measurable medium -- things can and should be measured and automated, made as efficient as possible. But too often, we tend to accept measurement at face value, without questioning the underlying imprecision (of, for example, cookies) and the implications that arise therein. We think, for example, that click-throughs are a good measure of ad campaign effectiveness, because a click is such a hard, tangible metric -- despite the fact that 8% of Internet users account for 85% of the clicks (and chances are, that other 92% is buying most of your products.)

Lately, we at comScore have had the opportunity to look under the hood of the business of selling online advertising based on demographic guarantees. I've written in this space before about the difference between cookies and persons, and the effect of cookie deletion on Unique Visitor counts when those counts are based on cookies. Most publisher demographic guarantees are based on cookies, and thus the imprecision of the cookie as a surrogate for a person injects significant error into the delivery of an ad to a specific (demographically targeted) person. But too, we've learned that even when publishers guarantee impression delivery based on registration -- knowing that this session was preceded by a specific user account signing onto the site -- there can still be some dirt in the assignment of the session to an actual person.

I'd like to refer you to the comScore blog post by our President/CEO Magid Abraham on the topic, but let me summarize some of the key findings here.

Ultimately the issue is that impression-level targeting is cookie-based, and a cookie on a machine may in fact tie back to multiple users of that machine. In a household with two parents and two teenagers, for example, in some cases four different people can share a single cookie. So even if a publisher knows that I'm a (young) 50-year-old male, because I registered at their site, really they've tied my age and gender to the cookie associated with my browser. That publisher cannot know that today it is my wife, not me, visiting the site from my computer. The cookie just isn't that smart.

In the comScore panel, we are able to see cookies passing between Web sites and panelists' computers; and we are able to see and identify the different persons using those computers at those points in time.
We recently conducted a study across some major online ad sales entities to determine how often the identifying cookie is associated with the behavior of a single user. Across 17 major entities -- publishers, ad networks, and third-party services -- we found an average of 44% of cookies were associated with a single user. Among the 56% of cookies that point to multiple users, if we assume that the wrong user is identified 50% of the time (which strikes as a conservative assumption), then on average, 28% of the time (half of 56%) the cookie involved in direct audience guarantees is pointing to the wrong person.

But it gets worse. In the market research and online metrics spaces, we like to assume that the basic demographic information provided to us when we ask for it is, a priori, correct. If many different Web sites ask me my age and gender, we like to assume that I'll tell them all the same thing, and that I'll be honest across the board. But there are in fact many reasons for persons to be less than truthful about, for example, age when registering at a Web site. Teenagers may want to be older to access adult content. Some people might fudge the truth a bit on a dating site. Others may simply not want to share their demographic data accurately because of privacy concerns.

In our panel, we can observe the demographic data that precise individuals provide to different publishers. We have seen, for example, a teenager tell one social network they were 17; another they were 36; and still another that they were 39.

All this is not to warn you off buying or selling guarantees off demographic composition of impressions. Rather, we began doing this research as an outgrowth of work we were doing with some major agencies, who wanted to understand what a reasonable expectation was of target delivery of a campaign in a post analysis. Basically, we did a post-buy evaluation for an agency that showed that a publisher's campaign delivered what I thought was a pretty darned good concentration of the (very narrow) target audience -- especially compared to the concentration one might expect to be able to buy from other media. But the publisher had touted a 100% target delivery, and the result of the disparity led to buyer concern.

It is important for buyers to have realistic expectations about the delivery of campaigns, because -- let's face it -- when advertisers are happy with the delivery of their online ad campaigns then they will come back for more, and everybody wins. Calibrating those expectations to ultimate delivery can only increase that happiness. Even given the issues raised herein, I still profoundly believe the Internet can do a better job than other media of delivering a minimum of waste (impressions outside the target) and thus a maximally efficient delivery in a targeted campaign. But, promising a level of targeting accuracy that can't be delivered benefits no one.

3 comments about "Setting Realistic Expectations For Targeted Campaign Delivery".
Check to receive email when comments are posted.
  1. Kevin Horne from Verizon, December 15, 2009 at 2:46 p.m.

    It's interesting that the % of single-person HHs plus % of single-parent HHs is about equal to your single-cookie stat of 44%. This implies that every other HH in the US has multiple people in the family using a single machine. I have a hard time believing that when i see the data on # of consumer PCs shipped annually and personal observations of kids holed up in the bedroom using their own machine for gaming, IM, facebook, whatever.

    Any supporting observations out there about multiple users on a single HH machine? 100% multi-use seems too high to me.

  2. Joshua Chasin from VideoAmp, December 15, 2009 at 2:55 p.m.

    Keep in mind that it was 44% of cookies, not computers. The more people using a computer, the more cookies that computer is likely to receive (because different users visit different sites.)

  3. John Grono from GAP Research, December 15, 2009 at 8:59 p.m.

    As always an excellent thought-provoking post Josh.

    You referenced the example of the 17 year-old using different ages at different sites and different times. I'm sure we all have anecdotal evidence of this happening, but I am extremely interested as to whether you have any empirical data as to its incidence.

    That is, is this 'fudging' done only on occasion or with some degree of regularity? Is it only certain segments of the online population or is it wide-spread - maybe a gender or age skew?

    I realise that there may be privacy/confidentiality issues, but any guidance as to the likelihood of this happening at a macro level would be wonderful.

    By the way Josh, this is not the 54 year old male researcher from Sydney that you know, but an attractive 26 year old Florida bikini-babe (only kidding).

Next story loading loading..