Maybe the backlash against digital targeting is not because targeting is a bad idea, but because it has been badly executed by the industry.
The quality of targeting data is in the news, which stands to reason since programmatic media rely mainly on data to decide whether to buy an impression. It’s now known that data is the big differentiator, secret sauce, and strategic weapon of media suppliers and advertisers alike. Data is what connects the marketplace.
Online, data is usually derived from behaviors, and online derived segments are basically lists of people who did those behaviors. We massage them, but by the time they go to traffic, they are still just lists. The issue is: What does a behavior mean?
Meaning is an issue because online behaviors are an imperfect predictor of attitudes, and so are demographics. Despite that, we liberally infer attitudes from behaviors and demographics. A viewing behavior that indicates intention to buy a Ferrari might just be a teenager playing around at Ferrari.com.
Discussing the quality of data, the industry is now using a “nutrition label” metaphor, which seems apt. So let’s dig deeper.
Origin, “istic,” Time, Inference
Foodies adore talking about origin. Likewise, modelers and planners want to know the provenance of a defined segment. An Auto Trader view of a 2010 Kia is a good pedigree, while a Twitter view of that same car may be less indicative of a potential buyer.
Also, as with food, data goes rotten with time. Intention is fleeting. “Interest” sticks around for longer, but maybe not much. Demographics, the canned tomatoes of targeting, last a long time.
Behaviors last forever because they are historical facts, but the attitudes inferred from them can last for seconds, or decades, or possibly never existed at all.
Should data have an expiration date?
Buyers also glom on to the terms deterministic and probabilistic. These are soft layers of abstraction like organic, or artisanal. I mean, the teenager checking out the Ferrari would be called deterministic, but a modeled billionaire in Dubai would be called probabilistic. Who is more likely to buy a Ferrari? What matters is the value of the audience, not the method of derivation.
North of Provenance
To work, the data-as-nutrition metaphor needs some important information not normally available: What’s in the package? Food declares what it is on the package, and that can be visually verified (usually!). Data, not so much. In fact, to the user, data is just a big pile of numbers. Gold or snake oil, who can tell?
We need to know whether audience composition is, in fact, what it is assumed to be. Do “auto intenders” intend to buy an automobile? How, indeed, does a planner choose from thousands of possible segment choices? To do that, we must pull back the veil. Did they have that condition? Are they that demographic? Do they harbor that attitude? After that, we need to know if those elements predict purchase.
The key to media quality is what portion of the audience was likely to be persuadable. Call that “target density.” In programmatic delivery, this is mainly a function of data quality.
In the advertising supply chain, moms with babies in diapers quickly become “females 22 to 35” -- or, online, “people who looked at a page about baby stuff.” But half the people who looked at baby stuff were shopping for a baby shower. Even I look like a mother when I shop for a baby shower. This is surrogacy, and sometimes (sadly) it’s all we have.
If our target is people who like ice cream, maybe no targeting is needed -- everybody likes ice cream, right? So the density of a target in the general population is a nice way to understand whether a specific segment was worth it.
In any case, when data segments are certified to contain the audience that they claim to contain, and when campaigns can be efficiently measured against real attitudes, not surrogates, media will become more accountable.
Many things in ad tech are assumptions built on top of assumptions - and data quality is just one of those.
Buyer beware and buyer, please don't waste your entire digital budget on data you haven't independently verified for accuracy -- both in terms of quantity and quality.
Powerful article Ted. My favorite that we're all interested in quantifying through the auriferous process of data mining, refining etc... "Target Density!"