You can adjust all of your cookie settings by navigating the tabs in this window.
This is the latest post in our “TIL (today I learned)” blog series that spotlights specific digital advertising technologies and strategies. Today we give you the lowdown on Ghost Ads.
Ghost ads are emerging as one of the most efficient ways to measure ad effectiveness.
In short, Ghost Ads monitor a control group and flag when a brand’s ad would have been served to a user in that group. An advertiser then bids on that impression using the Ghost Ad.
The Ghost Ad itself is invisible to the user. He or she simply sees the ad that actually wins the auction. However, the Ghost Ad and its bid data are visible to ad platforms and advertisers. From that they can glean how users who should have seen an ad behave when they don’t see it. Ghost Ads do the job of PSA-based A/B tests, they just do it more precisely and affordably.
Ghost Ads are more beneficial than PSA (public service announcement) and Intent-to-treat campaigns because they create apples-to-apples comparisons of users who were exposed to ads versus users who would have been exposed to ads.
Related post: TIL: Facebook’s Minimum ROAS Bidding
One drawback of Ghost Ads is that because you need to control the ad auction to execute them, only two companies can actually facilitate Ghost Ads: Google and Facebook. Yet there’s an alternative — called Predicted Ghost Ads — which we’ll discuss later in the article.
PSAs have long been a tactic for measuring ad effectiveness: you split target users into treatment and control groups, and show the treatment group real ads and the control group PSA ads (The Red Cross, The Humane Society). Marketers then compare the resultant, downstream behavior of the two groups.
Because Ghost Ads capture the same impression that would've been served to the treatment group, they create an apples-to-apples comparison of “exposed” and “would have been exposed.” Measuring true ad effectiveness depends on these groups containing the same mix of users.
The PSA methodology, however, requires paying for impressions just to know that you could have won. That gets expensive. The PSAs may also be invalid because today’s machine learning models will target people more likely to interact with PSA ads; this group will be different from the people in the treatment group who were shown real ads. The result is two audience types exposed to two different ads — an apples-to-oranges comparison.
Because Ghost Ads capture the same impression that would have been served to the treatment group, they create an apples-to-apples comparison of “were actually exposed” and “would have been exposed.” This is critical because measuring true ad effectiveness depends on these groups containing the same mix of users.
The chart above shows “exposed” treatment group users in the green region compared to their “would-have-been-exposed” control group counterparts. Ghost ads capture the same impression in the control group that would have been served to the treatment group, keeping the two groups the same.
Image credit: “Ghost Ads: Improving the Economics of Measuring Ad Effectiveness” by Garrett A. Johnson, Randall A. Lewis & Elmar I. Nubbemeyer
PSA campaigns simply can’t generate this “sameness.” And neither can Intent-to-treat campaigns that show ads to a treatment group but assume all these users were actually exposed to the ads. The fact is, some users weren’t reached because the ad lost the bid. Including these unreached users in comparisons creates more apples-to-oranges “noise.”
Here are the main ways Ghost Ads target people more precisely and help generate more incremental revenue for advertisers:
As mentioned earlier, advertisers can only really deliver Ghost Ads using Google and Facebook. Enter Predicted Ghost Ads, which many ad platforms can facilitate. Instead of telling you definitively that a Ghost Ad won or lost a bid, a Predicted Ghost Ad provides the probability. You may not know 100% that a ghost ad would have won an impression. But you’ll know that it had a 96% or a 50% or a 7% chance of winning. Knowing the probability of a bid can be just as valuable as knowing the black-and-white result.
Related post: TIL: Facebook’s Retention Optimization
But whether by Ghost Ad or Predicted Ghost Ad, these campaigns deliver much cleaner comparisons of treatment and control group purchase data than campaigns relying on PSAs.
With those comparisons in hand, companies can identify the impression types — based on user behavior, publisher, ad creative size and type, day of week, etc. — that are more likely to lead to a winning bid and purchase, and then optimize ad spend just toward those impressions.
This kind of efficient targeting and spending (called “optimizing for incrementality”) can generate tens of millions of dollars in net-new revenue for a company. That’s the difference between a profitable quarter and an unprofitable one.