The occasion of Sheryl Sandberg stepping down from her position at Meta is inspiring due reflection on the advertising products she helped build. Most commentators seem focused on implications for Meta and its stock price. For advertisers, a better question might be: Are those products good? Is the way Facebook and Google marry advertising and machine learning the best way? If the Sandberg coverage is any indication, it’s far from common knowledge that it’s not the only way.
Indeed, the runaway success of Facebook’s ad offering stemmed from disciplined adaption of Google’s business model, which Sandberg also helped design. After a series of missteps and a disappointing IPO, Facebook under Sandberg became the first company to achieve device-level tracking across millions of phones. As a result, it quickly became profitable.
Unique tracking capabilities gave Facebook an early lead in mobile display advertising — a then-new category that still drives most of the company’s revenue. While other ad companies struggled with app-based media environments with no browser, and therefore no cookies, Facebook figured it out. The company demonstrated superior ability to predict app downloads and purchases. This proved almost as valuable as Google’s unmatched ability to predict browsing and buying on the commercial Web.
It’s uncontroversial to say these operating models create value from surveillance. What you think of Facebook and Google depends on how much you trust them with intimate knowledge of almost everyone’s activity.
Both companies point to their noble missions. Google’s is to inform. Facebook’s is to connect.
But to apply the missions of the companies’ innovative, free utilities to the products that monetized them is and always was a stretch. Google.com might need to track every click to improve search results. The Facebook app might need to track every share to improve its network. But extensive user-level tracking never served the goals of advertising. Advertisers need predictions of what is broadly influential. We need signals of messages changing minds. Actionable ad analytics predict the difference an ad makes, across a population, over time.
This point is worth belaboring: Though to inform and to connect are advertising’s prime directives, the ad offerings of the companies claiming those missions were built with neither in mind. They were designed to maximize credit for sales.
At conception, this was no big deal — all ad offerings are built to get credit for sales. It’s by being lousy predictors that most ad models get a chance to be broadly influential. You target women but also reach some men who buy diapers. Gradually, though, Facebook and Google gathered and processed enough data to get good at predicting credit for sales. Then they invested heavily in machine learning, without altering the goal of their models, and they got great at it. (It’s worth mentioning: both companies mandate their own proprietary sales tracking and attribution, which is like predicting the winner of only the games you referee.)
Being excellent at answering the wrong question is a surmountable challenge for a business. Just shift to a better question. Despite advertiser pleas, neither company spent much energy trying to predict anything more useful than credit for sales. Instead, Facebook joined Google in a mean-spirited effort to coerce the ad industry into pretending influence doesn’t matter. It worked. We are expected to believe that algorithms so accurate can only be a miracle.
For 15 years now, any advertiser concerned that attributed credit might be a poor proxy for effectiveness gets sales attention from Facebook and Google, but no solution. If an ad pro outright rejects attributed credit as the goal for automated ad-buying, as the science of influence dictates we must, that pro will bring friction everywhere her career takes her.
It’s no secret that clicks are a bad signal in advertising. The way everyone knows Google and Facebook pollute publishing is the phenomenon known as clickbait. Clickbait happens because Google’s and Facebook’s algorithms can’t assess the quality of media. Remarkably, Facebook and Google execs have claimed clickbait thrives because media quality is difficult to ascertain. It’s as if they want us to believe their algorithms built the companies, instead of the other way around.
Advertisers, like other humans, have little difficulty assessing media quality. We look at our log files and visit websites where ads showed, because we know quality predicts lift. We consider the design of ad placements and the messages of ads. We note how ads are rendered within the mosaic of a page. We think about the time and place the ad reached us, and whether it was optimal. We pay attention to the quantity and quality of attention an ad captures. These are factors that influence audiences toward or away from a brand, a candidate, or a cause.
But to elevate data on any or all of these above personal data is to be out of sync with what Facebook and Google built. Their click machines barricaded the mindset of an industry against its own interests. References to the predictive power of ad platforms have taken on a farcical aspect. It is not miraculous to forecast my next purchase after tracking everything I’ve done on my phone for a decade. Too many advertising people sound like they’re hoping to watch that one trick over and over for a few more years.
How influence is exerted through media is a burning mystery Facebook and Google are deeply committed to ignoring. When their platforms are hijacked by bad actors, they literally claim media influence is a myth. Businesspeople so willfully blind will never reckon with having sold their customers a harmful product.
Algorithms are just business decisions. To use the factors of influence in ad-buying at scale is entirely valid, though entirely different from what the duopoly sold. It makes sense that it’s being built by entirely different people.