Would you entrust your life to an algorithm? Perhaps the advanced algorithms used in self-driving cars or the simple algorithms used in high-speed elevators at modern hotels — the ones with no buttons on the inside. You might assume that the parties
that developed these algorithms know what they are doing; you might even assume they have safety measures built-in. But algorithms are simple if-then statements — “if this happens, then do that.” Even the most elaborate algorithms are just multi-step if-then statements, or nested ones — if-then statements inside other if-then statements. As such, they are limited. What if the input they were looking for was missing? What if there’s a new scenario that was not accounted for in the laundry list of if-then statements?
The algorithms are not sentient; they cannot think for themselves and accommodate previously unforeseen scenarios. Algorithms also do not have morals, common sense or judgment. They slavishly do the calculations they were created to do; but that doesn’t mean they can’t kill you — think self-driving cars and high-speed elevators.
But what does this have to do with ad tech and digital marketing? Simple. You are entrusting your livelihood to algorithms, specifically your digital advertising budgets to ad tech algorithms. You assume the ad tech vendors that developed them know what they are doing. You may even assume the algorithms help you do better digital marketing. But they don’t. Ad tech algorithms are designed to separate marketers from their money as fast as possible. Don’t believe me? Consider the following.
Real Time Bidding Algorithms
Just like with high-frequency trading (“HFT”) on Wall Street, ad tech has fashioned RTB (“real-time bidding”) platforms that purportedly help marketers buy ads by the trillions on sites and apps by the millions. Note that all of this was completely unnecessary if advertisers bought real ads from real publishers with real human audiences in the first place. But let’s suspend disbelief for a few more minutes so we can dig in. In theory, when a webpage is loaded in a browser, multiple ad slots become available for advertisers to place ads into. Each ad slot sends out “bid requests” to auction off that opportunity. Multiple bids, sometimes dozens, are submitted by advertisers. The winning bid wins the right to serve the ad into that specific ad slot at that specific time.
All of this happens in mere milliseconds so the ad can be served within a reasonable amount of time after the person visits the webpage. But since the advent of real-time bidding, many layers of competing algorithms have been added on top to help various parties make more money. The enormous amount of computation takes time; sometimes it takes so much time that the ad is not served in time before the person has moved on — i.e. scrolled further down on their mobile device. That’s why you see blank rectangles marked “ad” with nothing in it. Advertisers won the bid, even though the ad wasn’t served in time or didn’t arrive in time to be displayed on-screen. The RTB algorithms don’t have this feedback loop so they literally don’t know the ad wasn’t served to completion — not “viewable” (no opportunity to be seen). But advertisers still paid for it.
Viewability and Fraud Detection Algorithms
Because of problems like viewability (above) and fraud (ads loaded by bots not humans), a crop of detection companies were created to help advertisers detect these problems. They threw more algorithms at the problems. But the problem was that bad actors’ algorithms were more advanced and able to trick the detection companies’ algorithms. So viewability and IVT (invalid traffic) problems were not detected. Newsweek used off-the-shelf malicious code to alter viewability measurements so that non-viewable ads could be sold as 100% viewable. The malicious algorithm defeated the detection algorithm. Fraudsters also buy traffic specifically designed to evade the detection algorithms of IVT vendors. The fake traffic is made from headless browsers (browsers without screens) and malware on devices. These software programs mimic humans’ behaviors like mouse movements, page scrolling, and clicks and touch events to trick the detection algorithms into not marking them as “invalid.” That’s how the bad guys’ algorithms “get away with it.” Also, note that if you train the detection algorithms on existing bots, that’s what it will look for (and miss all the other forms of fraud that are many times larger). See: Ad Fraud is More Than Just Bots. That’s why marketers should not entrust their digital ad budgets to the protection of viewability and IVT detection algorithms. Not only are those ad budgets not protected, marketers are wasting even more money paying for protection services that don’t work — remember the part about “separating marketers from their money as fast as possible?”
Algorithms Optimizing for Win Rates, Click Rates, etc.
Assuming we ignore all of the above problems with algorithms doing real time bidding, ad serving, and viewability and fraud detection, there’s still the problem of algorithms optimizing for the wrong things — i.e. parameters that are not aligned with marketers’ business outcomes. I have previously written about the easy-to-measure “vanity metrics” that marketers use to judge the performance of their digital marketing campaigns. Obviously higher click rates don’t necessarily mean more sales or better business outcomes, especially if you realize bots love to click ads while humans don’t.
But let’s blame the ad tech algorithms instead of the marketers. The algorithms used to optimize campaigns are “tuned” to optimize for easy-to-measure things like click rates, win rates, etc. More specifically, when algorithms see higher click rates, they increase bids or budget allocation to those sources; when algorithms see win rates that are too low, they raise bids to try to win more auctions. As you can imagine there are countless other ways the algorithms can be tuned — for example, “if the campaign is lagging behind in spending, then increase pacing so the money can be spent faster.”
Ad tech algorithms optimize for win rates and click rates because those are the data it can see and use in if-then calculations — if click rate is higher for site A, allocate more budget to site A. Sadly, site A uses bot traffic; bots click more than humans, so the ad tech algorithm just allocated more of your budget to the fraudster operating site A. Similarly, fake sites exhibit higher win rates, so ad tech algorithms faithfully allocate more budget to fake sites, unbeknownst to the marketers.
Hopefully marketers reading this are starting to realize that the algorithms are not tuned to help them do better digital marketing, but instead are tuned to help them spend as fast as possible (so ad tech companies can make money as fast as possible). Advertisers are entrusting their ad budgets to algorithms created by ad tech companies. These ad tech companies answer to their investors and their top priority is to make money and give a return to their investors, not to help you do better digital marketing.
Even if they keep telling you their algorithms are designed to help you improve your digital marketing, how could they? They don’t even ingest your sales data to use in their if-then statements and optimization algorithms. (I realize a handful of ad tech companies do). The vast majority of ad tech algorithms are tuned for optimizing towards higher win rates and click rates, which means they are sending more of your money to fake and fraudulent sites and apps — deliberately. By design, those algorithms are separating advertisers like you from your money as fast as possible. Would you entrust your livelihood and your digital ad budgets to ad tech algorithms?