31 Mar 2013 dmarti   » (Master)

The targeting game

We have an information gap for discussing the ad targeting problem. There are papers that, if you put them together, help substantiate the argument that adtech is bogus. But they're behind paywalls.

Here's a good one. "I'm not a high-quality firm, but I play one on TV" by Mark N. Hertzendorf. RAND Journal of Economics, vol. 24, number 2, summer 1991. $24 to download. I have a copy because I helped Doc Searls with some research for his book, The Intention Economy, but you probably don't. (I promise I'll get to the Open Access rant some other time. Yes, “closed data means people die” but we'll talk about that later.)

Advertising is a form of signaling. Rory Sutherland, vice-chairman of Ogilvy Group, said, To a good decision scientist, a consumer preference for buying advertised brands is perfectly rational. The manufacturer knows more about his product than you do, almost by definition. Therefore the expensive act of advertising his own product is a reliable sign of his own confidence in it. It is like a racehorse owner betting heavily on his own horse. Why would it be “rational” to disregard valuable information of that kind?

But advertising can break down as a signaling method when the medium is noisy enough that the probability of an individual user seeing an ad is low enough. Hertzendorf writes, Furthermore, the noise complicates the process of customer inference. This enables a low-quality firm to take advantage of consumer ignorance by partially mimicking the strategy of the high-quality firm. That's in an environment where the presence of many TV channels makes it harder for the audience to figure out who's really trying to signal. Noise helps deceptive sellers.

But what happens when we introduce targeting? Let's give the low-quality seller the ability to split the audience, without the audience members knowing, into marks and bystanders, with marks receiving the ad at higher probability. In that case, marks receive the signal of a high-quality seller, and the bystanders receive the signal of the low-quality seller.

Something that I just figured out from going over this paper again is that the splitting of the audience doesn't have to be accurate in order for adtech to work. Rebecca Lieb, at iMediaConnection, points out that her BlueKai profile is largely false, and writes, If ad platforms aren't delivering the targeting that advertisers are paying for, the emperor has no clothes.

Au contraire. Ad platforms are doing their work just fine. Targeting works even if it's inaccurate, as long as it can reliably split the audience. Even the most basic cookie scheme will do that. An ad network can randomly call some users left-handed and others right-handed, or divide them by height, or whatever. The only important thing is to split the audience persistently, so that some have a higher probability of receiving an inaccurate "high-quality" signal from a deceptive seller.

Where sellers in Hertzendorf's scenario must rely on increasing noise in the medium in order to deceive, targeting lets them make the first move.

We're still in the early stages of the game, though. If an individual is aware that targeting is possible and doesn't know if he or she is mark or bystander, the signal is lost. So you get the effect that I think is happening in web advertising, with the value of the entire medium going down, even for advertisers who do not target.

However, some buyers are still unaware of the extent of targeting. One politician saw an ad for a dating site on a political party press release and attributed it to the party, not to the Google ad service used on the site where he read it.

From my point of view inside the IT business, a lot of the adtech stuff looks old and obvious, but some of the audience is still figuring it out. People already detest and block the email spam that the Direct Marketing Association worked so hard to protect, because that's obviously "addressed to me." Understanding web ad targeting is taking a lot longer, which is understandable because it's so complex. (see bonus links below for introductions to the current state of the art.)

The signaling power of an ad campaign is the seller's advertising expense as estimated by the buyer. Advertising that is itself costly, such as celebrity endorsements or signs in high-cost areas, has what you might call "creative signaling power." Advertising that is attached to a high-cost medium, such as Vogue magazine or the Super Bowl, has "media buying signaling power". And there's a multiplier effect from the quality of the ad itself, since some ads are more memorable than others and tend to make people think that they've seen them more often. (so quality does not map directly to "informative" or "entertaining".)

When an ad appears in a medium that facilitates targeting, the media buying signaling power tends to go away, depending on the accuracy of the targeting and the audience member's knowledge of the extent of targeting.

Brand advertisers, who Doc Searls splits out from direct response advertisers, seem to have an understanding of the targeting problem. John Hegarty, founder of the ad agency Bartle Bogle Hegarty, said, I'm not sure I want people to know who I am. I find that slightly Orwellian and I object to it. I don't want people to know what I drink in the morning and what I drink at night. I think there's a great problem here - throughout history we have fought for our freedom to be an individual, and you're taking it away from us. I think there'll be a huge backlash to that and Nike will have to be very careful.

And Richard Stacy writes, The great thing about advertising is that no-one takes it personally.

On the audience side, we have the feeling of "creepy", which is hard to pin down, but that I think is an important notification from your inner economist about an information imbalance, which you would be mistaken to ignore.

So here, roughly, are the rounds of the adtech game. It would have been an interesting experiment to play them out in order, but this is a real-time strategy game, not a turn-oriented game. Some players have gotten to round 3, and others are still on round 1, or think they're on round 1 and are getting beaten at round 2.

Round 1: Targeting that partitions the audience without the audience's knowledge. Need not be accurate because a persistent split is enough to attract low-quality sellers. I'm using "low-quality" in the economics paper sense, not the "haha your phone sux and mine r00lz" sense. Most adtech people are not in it to deceive, but from a misguided quest for efficiency that follows from a lack of understanding of signaling.

Round 2: Audience begins to understand targeting. Ad blocking increases, value of web ads decreases. Increasingly crappy web ads drive demand for privacy tech.

Round 3: Privacy tech, such as stricter treatment of third-party cookies, makes targeting more difficult and less accurate. The value of advertising across the entire medium rises, and the sites that pay for original content are able to get more ad revenue and control, at the expense of adtech middlemen.

Just as targeting didn't have to work with total accuracy to give an advantage to deceptive signalers, privacy tech doesn't have to be 100% to push things back in the other direction.

Bonus links:

Retail Surveillance Is About To Make Your Online Targeting Seem A Lot Less Creepy

NewsCred Blog: How magazines can stay relevant in the era of branded content and digital marketing

Jim Brock: Would Do-Not-Track lead to “data oligopolies”?

Adam Lehman: Just Who Do The Data Paranoiacs Think We Are?

xkcd: Instagram

Scott Meyer: Why Is Another Browser Company Forcing Naive Decisions on the Internet?

Mike Daly: Today’s Burning Question: Firefox to Block Third-Party Cookies By Default

Doc Searls: How advertising can regulate itself

Advance look at post-adtech web ads: Village Soup Shows ‘Native’ Ads Can Work on Local News Sites

What everyone should know about ad serving

How and Why We Track: Confessions of an Ad "Tracking" Company

7 things you don't know about ad networks and are afraid to ask

Syndicated 2013-03-31 16:40:06 from Don Marti

Latest blog entries     Older blog entries

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!