How To Avoid The Survivorship Bias Trap In Digital Advertising

joel-nierman"Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media

Today’s column is written by Joel Nierman, global marketing and media director at Critical Mass.

Advertising has always carried a survivorship bias -- the logical trap of concentrating on people or things that "survived" some process, while inadvertently overlooking those that didn't because of their lack of visibility.

In our case, we too often exclude data produced by most of our advertising. But the first step to evade a trap is knowing that it exists. Then we must acquire the knowledge needed to avoid it.

A low-rent type of survivorship bias exists in traditional media: taking basic demographic or psychographic targets and indexing those groups against television program viewers or magazine readers to guide media buys. Advertisers would identify people who were generally more likely to purchase a product and find the media those people consumed. They’d then run ads to others with the same characteristics, trying to exclude the population that did not purchase. The nature of traditional media overcame this problem because ad campaigns’  “wasted” impressions were not actually wasted; instead they unintentionally included groups meant to be excluded during the planning process, including the nonsurvivors.

Today’s digital advertising world has taken survivorship bias to a whole new level. It’s even a much larger phenomenon than just a few years ago because of easily placed pixels and the proliferation of technology platforms that can tackle retargeting and lookalike modeling against the data du jour. It is so easy to find great detail about who has been buying, signing up and watching that the natural reaction is to find more of those people, which the technology easily facilitates. To be fair, these tactics do drive relatively good performance, so their use is not unwarranted.

But what of all the people who don’t convert or take further action after being exposed to some impressions? If the digital media is targeted properly then all the exposed people are in the core target, and yet, as we all know, more than 90% don’t convert once they get to an owned hub. More than 99% don’t even visit a hub in the first place. This is a huge survivorship bias.

Here’s the problem: As an industry, we base decisions and learning on such a small portion of our target groups -- the ones who actually convert -- that we limit the data we use and, correspondingly, the efficacy of our campaigns. We willingly exclude the data produced by the vast majority of our advertising.

Trying To Fix The Bias

Fortunately, the same technology that creates the bias can rectify it. Any DSP or DMP that is worth the incremental CPM can compare various groups, such as converters vs. cart abandoners, nonclickers vs. clickers and watchers vs. readers. Even if the surface characteristics of these groups are effectively equivalent due to precise targeting, there will be key underlying variables that differentiate the two. What caused one suburban mother of two to buy a new tablecloth when another didn’t is there, but we just have to dig to find the answer.

Some platforms take this into account. Their algorithms look at the characteristics of nonconverters and nonresponders to better target advertising as the machine learns. But this isn’t good enough because most algorithms filter out these characteristics from the target set. This reduces the universe of targetable people over time as more people in the target group assume the characteristics of nonconverters, with fewer moving into the converter group.  At some point this diminishing target set won’t yield the needed volume or efficiency.

Dig Deeper

To overcome the digital advertising survivorship bias, we as marketers must learn the characteristics that distinguish converters from nonconverters so we can address them. If the suburban mother of two isn’t buying new table linens because her cats will demolish them, this is a situation that can be overcome by a tactical adjustment and a good copywriter. Knowing these key pieces of information allows today’s digital marketers to do what they do best: prospect for new customers and drive results with targeted and relevant messaging by leveraging data and technology.

The real key is to continue to prospect. Traditional media, particularly television, has survived well in part because it can’t help but prospect because of the broad manner in which it is targeted and bought. But continuing to prospect means identifying a consumer need, and then using products and services to fix that need. Advertising delivers the message that the need can be fixed. So to truly defeat the survivorship bias we must go back to the core principles of advertising. To do this in digital advertising means using technology to learn why things happen, and not just that they happened.

As you know by now, knowing a trap exists is the first step in evading it. We as marketers need to know that the survivorship bias exists in all of us. Then we must actively acquire the knowledge to overcome it. Doing so will allow us as an industry to truly unlock the power of all this data and technology that’s received so much recent hype.

Follow Joel Nierman (@FozzieBuyer), Critical Mass (@criticalmass) and AdExchanger (@adexchanger) on Twitter.

1 Comment

  1. Dorothy Higgins

    As a former traditional media planner happily evolved into the modern world of integrated comms planning, I must demur with the statement that we saw "non-core" impressions as waste. Nothing could be further from the truth. The power of mass media is the capacity to reach current and prospective audiences and value the opportunity to build the brand franchise using reach to drive scale among those primed to purchase and frequency to drive perceptual shift among prospects.

    Reply

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>