reading 5 Flashcards
A popular approach for advertising effectiveness experiments is delivering
public service announcements (PSAs) to the control group
PSAs identify baseline purchase behavior among the subset of control consumers reached by the ads: the experimental ad effect estimator compares outcomes between people who see the focal advertisers ad and those who see the PSA
PSA bad because
prone to error because they require coordination among advertisers, publishers and third party charities. Worse, PSA experiments are rendered invalid when marketters use computer algorithms to optimize ad delivery in campaign separately for the PSA and focal cammpaing
Ghost ads
like PSA ghost ads identify ads in the control group that would have been the focal advertisers ad had the consumer been in the treatment effect on the treated. Unlike PSAs ghost ads are compatible with campaign optimization tech and avoid the costs of PSAs. Instead of PSA ads, the control group consumer sees whatever ads the ad platform chooses to deliver when the focal ad is absent. The ghost ad methodology is implemented at the ad platform level
predicted host ads
many online display ad platforms only suggest an ad to a downstream web publisher which can reject the ad for many reasons. A new approach is needed because the naive ghost ad approach cannot observe control group cases in which the platform suggests the focal ad but the publisher refuses it. This slippage breaks the symmetry between the focal ads in the treatment group and the ghost ads in the control group. To offset this shortcoming, the prdicted ghost ad methodology preserves this symmetry by running a simulated auction across both treatment and control group consumers to determine whether the platform intends to deliver an experimental ad before determining the real winner in both cases
Problem with A/B testing in online ads
traditinoal AB testing involves splittting a group of users into a treatment and control group. By comparing outcomes between these groups advertisers estimate the incrementality caused by the ads
But its expensive and imperfect targeting: the people in the control group might behave differently just because they didnt see any ads, leading to potential bias
logistic issues: A/B testing disrupts the normal ad delivery system, which can lead to additional overhead for advertisers and platforms
Heres how ghost ads work
Control Group with Simulated Ads: In the Ghost Ads method, users in the control group are not shown the actual ad, but the system keeps track of when and where the ad would have been shown had they been in the treatment group. These “would-be” ads are called ghost ads.
Tracking Post-Ad Behavior: Although the users don’t see the ghost ads, the system still tracks their behavior afterward, just as it does with the treatment group that sees the real ads. The idea is to understand how people would behave with no real ad exposure, but with their potential exposure being accounted for as if they had seen the ad.
Incrementality from Ghost Ads: By comparing the outcomes of users exposed to real ads (treatment) versus those exposed to ghost ads (control), advertisers can accurately measure incrementality—how much of the revenue, engagement, or conversion can be attributed to the ad campaign.
Advantages of ghost ads
Minimizing revenue loss: In traditional AB testing, suppressing ads for the control group results in a loss of potential revenue. ghost ads remove this issue because they simulate the ads rather than withholding them, so the system can more accurately represent the scenario where everyone receives ads
2) preserving the ad ecosystem: unlike traditional experiments that disrupt the regular ad delivery system, ghost ads maintain the platforms normal functioning. The ads are delivered as they normally would be, adn the only difference is that the control group doesnt see them, though the system still tracks the potential ad slots
3) Improved Targeting and Accuracy: Because the system still allocates ad slots for control users as if they were receiving the ads, it avoids the bias that comes from simply not showing them any ads. This leads to more accurate results in measuring ad effectiveness.
4)
Lower Cost of Measurement: Ghost Ads reduce the inefficiency of ad suppression, making measurement much more cost-effective. By simulating the ads for control users, advertisers can measure impact without losing potential revenue from withheld ads.
5)
Flexibility for Different Campaigns: The method works well for a wide range of ad campaigns, from direct response ads (aimed at immediate conversion) to brand awareness campaigns. Ghost Ads allow advertisers to better understand which campaigns drive measurable outcomes and which are less effective.
Ghost ads also offer economic advantages over traditional methods
More Efficient Resource Use: Advertisers can measure effectiveness with fewer lost impressions, preserving ad delivery to more users. This means better use of resources and more accurate allocation of ad budgets.
Better Data for Decision-Making: Because the method captures the normal ad delivery process, advertisers can make better decisions about which ads to run, which audiences to target, and which campaigns are most profitable.
Incrementality-Based Budget Allocation: The ghost ads method gives a clear picture of the incremental value generated by ads, enabling advertisers to optimize budgets around the campaigns that provide the most lift in sales, engagement, or other desired outcomes.
Paid Search Ads Generate Limited Incremental Sales:
The study found that a large proportion of clicks on paid search ads were from consumers who would have made a purchase regardless of whether the ad appeared. This suggests that many paid clicks are not truly incremental but are instead from consumers who were already planning to visit the website or make a purchase.
For branded keywords (e.g., “eBay”), the incremental effect of paid search ads was particularly small. Consumers searching for “eBay” were highly likely to click on the organic search result or directly type the URL into their browser, meaning the paid ad had little to no additional effect on their purchasing behavior.
Heterogeneity in consumer behavior
The experiment revealed significant heterogeneity in how different types of consumers responded to paid search ads.
Existing customers, who had purchased from eBay before, were much less responsive to paid search ads. Many of these consumers would have found their way to the website without the ad, either through organic search results or by directly navigating to the site.
New or infrequent customers, however, were more likely to be influenced by paid search ads. For these consumers, the ads played a more meaningful role in driving traffic and purchases, providing incremental value.
channel substitution
The data showed that paid search ads often acted as a substitute for other marketing channels, such as organic search results. When ads were not shown, many consumers simply clicked on the organic link that appeared in the same search engine results page (SERP). This substitution effect was particularly strong for branded keywords.
This raises questions about the necessity of paying for branded keywords when the organic listing is already prominent and likely to capture clicks.
n branded keywords vs branded keywords
Paid search ads for non-branded keywords (like “buy shoes online”) were more effective at generating incremental sales than ads for branded keywords. This is because non-branded keywords are often used by consumers earlier in the purchase funnel when they are still exploring options. Paid ads for non-branded keywords helped eBay capture traffic from users who might have otherwise gone to a competitor
