Never trust algorithms to spend your money; always use FouAnalytics to monitor

Today comes news of "Meta’s ‘set it and forget it’ AI ad tools are misfiring and blowing through cash." Some glitch caused Meta's algorithms to blow "through roughly 75 percent of the daily ad budgets ... in under a couple of hours. [Also], A usual CPM of under $28 had inflated to roughly $250, way above the industry average. Small businesses have seen their ad dollars get wiped out and wasted as a result. Meta hyped Advantage Plus shopping campaigns during earnings calls as a carefree, “set it and forget it” automated solution to online ads. [Advantage Plus] has overspent on numerous occasions and ignored the cost caps we have in place on it,” he said.

You had been warned (by me). June 2021, Forbes -- "You are entrusting your livelihood to algorithms, specifically your digital advertising budgets to ad tech algorithms. Ad tech algorithms are designed to separate marketers from their money as fast as possible."

Algorithms optimize based on available signals

You might think that algorithms can do better than experienced marketers at optimizing digital campaigns. There's a remote possibility that could be true. But that is not the case in 99% of the existing cases, even with "custom" algorithms. Why? Algorithms optimize based on available signals. What signals are available? Right, clicks. We've seen countless examples of algorithms allocating more budget to fake sites that have higher clicks and click rates than real sites. That's because the fake sites use bots and bots click ads more than humans do. The algorithms think that more clicks means more performance, so they allocate more budget to the fake sites.

You might argue that the algorithms you use optimize for sales. But if you were truly honest, you'd realize that the number of data points on sales are SO few and sparse there's hardly enough data for the algorithms to use to optimize. What is happening instead is the algorithms are claiming credit for sales that had already occurred or would have occurred anyway; this is a form of misattribution, if not outright fraud (i.e. deliberately falsifying attribution). Remember the Uber lawsuit from 2017? Not only were app installs being falsified so fraudsters could get paid the "cost per install" ("CPI"), some fraudulent mobile exchanges were simply falsifying the attribution (claiming to have caused an app install, when the Uber app was already installed, or fabricating the transparency reports, when they didn't even run any ads.

Finally, for some of the largest advertisers like FMCG/CPG, automotive, and entertainment (movies), their sales occur offline. So the signals for optimization are not readily available for the algorithms to use to automate the placement of the ads and the allocation of budgets. To be more specific, a large FMCG advertiser sells goods to wholesaler, regional distributors, or large retailers like Walmart. Over the last 20 years, I have witnessed their struggle in getting granular sales data back from these parties, even for their own sales planning and forecasting for manufacturing. If they were struggling with that, do you think they got the sales data from local stores to feed their programmatic algorithms? Of course there may be rare occasions or "one-offs" where this happened and they did get the data to use; but those are the exceptions not the rule.

See: Why FouAnalytics is NOT automated, NOT set-and-forget

Always have analytics in place to monitor campaigns

Hopefully, you now understand that you shouldn't trust algorithms to spend your money. If you have no choice but to buy on large platforms or programmatic channels, then at least have analytics in place so you can monitor your campaigns. That way, you can spot problems earlier and make adjustments, rather than wait till the campaign is over, or the algorithms have blown all your budget. This exact use-case happened this week in a CTV campaign. The team did not make any changes but the nature of the CTV campaign changed completely on April 24 (see the large increase in red in the FouAnalytics chart below). Since they had FouAnalytics in place, it took less than 3 minutes to see what happened. Before the change, the "custom_domain" data grid showed typical CTV apps, and the corresponding supply sources. After the change, you can see most of the custom_domains were websites and not CTV apps. The supply sources also changed. The reason it was marked as dark red in FouAnalytics CTV measurement is because the CTV ads were not run on large screen connected TVs, but on crappy websites like dmv-practice-test .com. Are you OK with paying CTV prices for ads that ran on websites and mobile apps?

In the above case, the client turned off the supply sources that were not delivering real CTV impressions. And remember the following, where 80% of the impressions did not run on CTVs but on websites? Get in touch if you want to use FouAnalytics to monitor your expensive CTV campaigns to ensure you are getting what you paid for -- CTV ads.

Further reading: https://www.dhirubhai.net/in/augustinefou/recent-activity/newsletter/