How to Combat "Made for Advertising" Sites
By Ali Manning
In June, the ANA released a preview of its Programmatic Transparency Study, which reported that 15 percent of advertiser budgets are going to garbage sites. We are 20 years into this thing, so why does this keep happening?
Despite extensive industry collaboration to set standards, a competitive marketplace for software to root out fraud and guarantee brand safety, the industry continues to wrestle with low-quality ad experiences gobbling up ad budgets and eroding brand value with subpar consumer experiences. The ANA's latest study focuses on "Made for Advertising" sites, those that spoof real content in the pursuit of heavy ad loads.
It's time to admit standards and monitoring will not solve the problem. The automated solutions amount to weeding your garden with a lawnmower. Without getting to the root of the issue, it will continue to spring up, as it has periodically over the past decade.
The root of the issue is: Digital media KPIs are highly gameable. Clicks and attributed conversions are poor proxies for business outcomes. There's a reason CEOs don't go on earnings calls and credit market share growth to video completion rates. They don't matter.
Advertising is most effective on quality sites. Fake sites do well in campaigns measured on clicks, attributed conversions, or video completion rates.
Marketing starts with intended business outcomes. Marketers thoughtfully plan creative and audience strategies to achieve their desired outcomes.
But when it comes to ad buying, outcomes are not considered. Powerful machine learning is instead trained on those gameable KPIs, which they are only getting better at achieving automatically. No creative or audience strategy proves out if the last mile is controlled by a robot trained to deliver ads to other robots, or to hand out coupons to your customers when they're lined up at the cash register.
Reach is what brands need their awareness campaigns optimized toward. But the algorithms that make most media decisions are trained to deliver high video completion rates at a viewability threshold with low CPMs. The values that drive ad delivery are not the values of the brands paying for those ads.
Even companies who have invested in measurement solutions that are great proxies or even "the real thing" are almost always using that measurement only to "read out" results of campaigns, instead of feeding optimization directly. When measurement does make its way to optimization, the process is usually manual.
While every brand says they value high-quality content, the more pressing question of whether ad opportunities are valuable enough to justify their cost is answered by algorithms. "Scaled" decision tech is not the best tool for brands with their own data, pursuing specific business outcomes. There is no "flight to safety."
Even brands who only buy direct are making decisions through generic processes lacking due consideration of total cost and total value. (Though it never seems to make the ANA studies, big publishers can't serve all the demand for premium at competitive rates while competing with junk, so often mix "audience extension" and other junk into their "premium direct" packages.)
Because this has been going on for so long, and generative AI promises to make it worse, it's now past time to pull up the weeds at the root.
The solution is already under marketer's control: Marketers already own the training data and the outcomes; all they need is their own media buying algorithms.
You can't spoof real business results. Train custom media buying algorithms to achieve your real business outcomes, and AI automation will finally steer you away from garbage.
Control makes all the difference: your AI, trained on your brand values.
Why? Because whether you take control or not, algorithms will decide what publications and content is valuable, and reward it with your budget. To train your own is to make your values the machine's source of truth.
There is no AI without training data, and training data determines what AI learns. For small advertisers, this is a challenge. But brands have first-party data, as well as their media log data (which I'd argue is also first-party), plus data from publisher partners, measurement partners, and licensed third-party data. Brands can choose what's best for AI training. This approach has already proven much more powerful than any shared algorithm, even those built by the most powerful platforms.
Bring real data that matters to a brand to increase the predictive power and get the algorithm predicting the right outcome, and the cream will rise to the top.
The views and opinions expressed are solely those of the contributor and do not necessarily reflect the official position of the ANA or imply endorsement from the ANA.
Ali Manning is co-founder of Chalice Custom Algorithms.