Hello dear readers! James Hercher. We’re back from vacation to bring you the latest and greatest in commerce news.
This week we cover a new topic, but it has grown so quickly that it already feels like an old hat. And it’s the machine learning technology that consumes the open web’s advertising budget as large walled gardens integrate media, data, technology and audiences within their own fortresses.
PMax phenomenon
Two weeks ago at the We Make Future conference in Remini, Italy, Pierpaolo Morgante, senior client solutions manager for Microsoft Advertising, said during a presentation that the group would “build its own version of We are building P-MAX.” (H/t Recorded by Mike Ryan of consultancy Smarter Ecommerce. )
Perhaps Google, while its product name “P-MAX” has become a popular abbreviation for the black-box advertising product that walled gardens use to centralize media and data, has advertisers wanting data. They probably don’t like having little or no reporting.
However, almost every large platform has its own version of “P-MAX”.
TikTok is called Smart Performance Campaign and Meta has Advantage+ Shopping Campaign (ASC). Microsoft Advertising is in a testing phase and has not announced a product name other than a “proprietary P-MAX version.”
why are they all doing that? Because the “P-MAX” approach takes the walled garden model one step further by letting the platform control targeting, measurement and creative, and consolidating multiple media channels into one black box. Advertisers’ campaign controls should be set against CPA metrics, creatives uploaded to the system, and possibly broad audiences and geographic targets, rather than constantly tailoring campaigns based on specific audiences or behavioral niches It will be charged.
“You just click a button to turn a switch on in your ad group settings and Microsoft will use your existing assets to find the right users for your campaign,” he said, introducing predictive targeting, the group’s first ad. Technology products with native AI integration, according to a Microsoft Advertising blog post last week.
The major platforms are clearly fully committed to machine learning-based advertising. The question is whether machine learning technology is ready for prime time.
machine unlearning
Google’s PMax is a pioneer in advertising platform AI products, but the highlight is Meta ASC.
ASC is a foreshadowing of whether other platforms (everyone claims they already have their own “PMax”) can do the same trick.
That’s because Meta is going full steam ahead with ASC despite setbacks and major red flags, and machine learning is taking over its advertising platform.
Last month, one of Meta’s new machine learning optimization products was bugged, costing advertisers tens, if not hundreds of millions of dollars, most of which Meta returned as ad credit. Did.
The incident may presage the future of advertising in the age of machine-learning platforms.
Marketers who spend tens of millions of dollars a year on Meta were unable to reach human account representatives for days, and automated systems were unable to fully resolve issues and customer complaints.
Meanwhile, more than a dozen meta-advertisers I spoke with say their ad platforms are still reeling, with wild swings in performance and measurement. Problems with ad serving and platform accounts, such as ads being stopped for no good reason or failed campaign budget entries, are now the norm.
AI takes the wheel
Meta hasn’t stopped to reflect on its recent debacle. Rather, e-commerce and product sellers are being pushed even harder towards automating machine learning.
For example, the Facebook Shop has benefits such as Shopify Attribution and web-to-app optimization integration that are only available if you agree to Meta’s machine learning controls. Without web-to-app functionality, iOS privacy rules prevent measurement when someone links to your website from the Facebook or Instagram app.
Three advertisers from a specialized social media ad buying agency who closely track daily CPMs and performance have seen their ad platform’s He said he doubts the cost-per-click has increased dramatically. .
In other words, the cost of generating clicks on product ads and brand posts that drive people to shopping URLs doubles. Three times according to one source, four times according to another. This is because ASC is driving demand for machine learning for advertisers and her Meta. Based shop and web integration.
However, the Facebook Shops product appears to be buggy, especially with a long-awaited update to its glitch-prone machine learning controls.
Last November, just in time for the Thanksgiving shopping holiday, Facebook Shops was having problems with false fraud reports, shipping delays, and customer service as brands were busy with holiday sales.
and still Basic shop integration Customer service is out of sync and email replies to complainants fail. Marketing strategies like buy-one-get-one promotions and discounts on in-store purchases also fail.
new big dilemma
Some might argue that glitches are natural because they are growing pains. New Facebook shop. Google PMax is new.
Shops doesn’t have all the little ecommerce tools that Shopify does. Also, Google still exited the beta of PMax early and is still in the back 1 stage of the feedback and upgrade cycle. And Google has shown its readiness to make concessions to PMax and change the product based on feedback.
All true.
But it’s hard to reconcile the argument that these products are new and underdeveloped with huge sums of money to advertiser bank accounts and brand development and extensive licensing platforms.
According to many retail executives, agency buyers and e-commerce consultants, this year may be the first time many e-commerce and retail companies are spending more on PMax than directly on Google search and YouTube. .
Again, in case it didn’t catch on, for many high-spending marketers, PMax budgets in 2023 will be larger than standalone YouTube and Google Search.
But if Google and Meta are going to put a lot of advertiser money into PMax and ASC, especially if the product is new or experimental, without transparency and full creative control over the platform. We cannot rely on that excuse.
