Digital advertising is one of the biggest new industries of the Internet era. In 2018, USD 273Bn was spent on digital ads globally.
What if I told you it was all fake?
Or more accurately, what if this article told you it was all fake?
The article has several examples of real-life experiments – e.g., when eBay stopped all its ads on Google for 3 months, the drop in online sales was 0%. Yes, you read that right – zero, nada.
But how? Aren’t people clicking on those ads? How can the impact of removal be zero?
Two factors drive this surprising outcome.
- Do you remember the last time you searched for an app on Google? For example, I searched for “Spotify” last week. The first link was an ad, for Spotify.com. So I just clicked on that. And an ad-click counter somewhere in Google went up. This is called Selection Bias. Of the people who clicked on any given ad, you don’t know how many were looking for that item anyway.
- Search engine algorithms are trained to show a given ad to people who are most likely to click on it. This Algorithmic Bias magnifies the selection bias. As the algorithm gets better (and Google’s algorithm is pretty mature now), increasingly, the people who are shown an Amazon ad are the ones who were planning to go to Amazon anyway.
In fact, some skeptics believe that ads don’t work at all. That no one is convinced to change behavior based on ads. Could it be true?
Well, do you remember the last time you clicked on an ad while not looking for that exact item? Me neither. (except for that banner selling N95 masks last week, but it was sadly out of stock).
But surely, you say – if digital ads didn’t work, why are marketers still spending hundreds of millions on them?
As the article says, quoting Steve Tadelis, “marketers actually believe that their marketing works, even if it doesn’t.” Good old Cognitive dissonance.
Yes, cognitive dissonance does play a role. But I think there’s a second element too. You know the saying, “what gets measured gets managed“. Google and Facebook report clicks and impressions, not actual incremental purchases. That’s what is available, so that’s what marketers track and optimize. They don’t track incremental purchases, because there’s no way to monitor that.
In other words, what doesn’t get measured, doesn’t exist.
- Scott Alexander wrote a great article on selection bias among doctors: Why doctors think they are the best
- Scientific American on how to convince someone when facts fail