Talk about timing. I set out to write this column before the article “
Marketers Question Quality of Ad Targeting Data
Providers”
appeared in the
Wall Street Journal last week. In fact, I planned to write my post specifically so misinformation like this wouldn’t be published. Let me
explain.
One media-buying exec is quoted in the WSJ article saying, “If you’re doubling your costs by adding in data, the performance has to be at least twice as good. Often
the cost of the data won’t justify itself.” That’s a widely believed truth in the industry. I believed this initially as well. This begs the question, “Why buy data at
all? We’ll get to that in a second.
Nearly every ad server and DSP today use last-ad-seen or last click as the default attribution methodology. Most agencies and marketers do
as well, at least within the display/video ecosystem. If your campaign is 50% user-targeted data and 50% non-user-targeted data, the only time the user-targeted data gets credit for the conversion is
when the user-targeted impression was the last ad served/clicked before the conversion. If you’re targeting dozens of segments, the chance any one is the last ad seen or clicked is
minimal, and is much more likely to contribute to a multi-impression path to conversion instead.
In one study, we took nearly 1,000 campaigns across industries, all of which had KPI-driven
conversion goals, and analyzed every impression in the path to conversion among each user that performed a qualifying KPI activity. Then we removed all users who only saw one impression prior to
conversion to allow a better gauge of how a blend of user- and non-user-targeted impressions influence performance. Here’s what we found:
1)
Baseline: Users who were only shown ads with site- or content-based targeting.
2) Blend: Users that were served an impression within both a
content/site-based targeting ad group and a user-targeted ad group using third-party data. Result: A 4x lift in performance.
3) Data only: Users who were
served multiple impressions each time using user-targeted third-party data segments. Result: An 8x lift in performance.
The question of data accuracy still exists, but it’s
an unnecessary worry. Audience discovery may reveal that despite your desired target audience being “cold and flu sufferers,” “cat lovers” performed best. So what? Buy
cat lovers! At least you’ve identified the inaccurately described segment that you really want! Don’t get hung up on “what should be.” Buy “what is” and what
works.
Seeing results like this means there is great value in data and the data business model. Discovering and understanding that value is 100% dependent on selling data as a
pure-business model. Those providers who insist that they’re the only ones who can marry their data with media to enable good performance are misguided and prevent buyers from discovering
and valuing the data accurately. There are dozens of these companies now, providing an unfortunate obstacle to accurate data valuation.
I believe that bundling and “data
doesn’t work” misinformation are two things holding back user-targeting and data usage today. I hope this post is a start toward solving the latter problem -- and that data companies
unbundle to solve the former, enabling a brighter future for themselves and their industry.