Depending upon your experience, it’s likely that the advent of artificial intelligence — and its most commonly deployed subset, machine learning (ML) — has either been a boon or a boondoggle. I would proffer that the pole at which you reside depends not on your technological savvy, but rather the manner in which ML has been integrated and executed. While the number of use cases is staggering, dissecting several common marketing applications provides an easy path toward enlightenment and appreciation.
If you market a service or product (or merely draw breath in 2019), there’s an overwhelming chance you’re aware of how ads are placed on Google and Facebook. While both platforms are hurriedly building out ML capabilities in an effort to entice more budgetary allocation from marketers, their approaches to doing so have been a study in contrast and have led to decidedly different outcomes for early adopters in the marketing community.
Machine Learning And Google, Part I
From a user experience (UX) perspective, Google has long been at the forefront of machine learning. Consider the personalized nature of its search results, which are informed by the inputs you provide during each interaction, coupled with the physical environment surrounding you.
However, Google’s value proposition to advertisers has historically been vastly different. Enter ad tech providers such as Kenshoo, Marin, Acquisio and the like, which have created a highly profitable and effective boutique industry by creating “bidding algorithms” that score results from marketing placements and dynamically adjust the amount an advertiser is willing to pay for each keyword (in other words, deploying simple ML). Clearly the good folks at Mountain View took notice, hence the onset of …
Machine Learning And Google, Part II
If you like acronyms, you’re going to love Google’s recent product updates. Building upon the prior success of remarketing lists for search ads (RLSA), dynamic search ads (DSA) and responsive search ads (RSA) have now joined the fracas. The intent of DSA, it seems, is to reduce — if not eliminate — the need to set up advertising campaigns, instead allowing Google to scrape each advertiser’s website and build dynamic outputs. For RSA, the goal is to quickly test as many variables as possible within ads, utilizing machine learning to understand which combination of headlines and body copy provide the optimal mix. There’s also Smart Bidding, which is Google’s answer to the algorithm.
Machine Learning And Facebook
While Google’s updates have been couched as product enhancements, Facebook has gravitated toward the stick instead of the carrot. Facebook has access to more than 1 million gigabytes of consumer information on a daily basis, which provides incredibly fertile ground for machine learning to mine this information and enable ad-targeting precision. That said, whereas Google has been a performance marketer’s dream due to the sheer magnitude of variables that can be manipulated, Facebook has long operated as more of a black box. With results that rival and often surpass those of Google, though, it has been able to succeed with this approach.
With the recent rollout of the Power 5, Facebook has all but commanded advertisers to entrust its ML to place advertising in the most effective manner possible. This includes five core components: auto advanced matching, account simplification, campaign budget optimization, automatic placements and dynamic ads.
So, what’s the gist? Whereas search engine marketing gained popularity largely due to the granularity at which its placements could be bought, social marketing has mandated the opposite. The only way to maximize efficiency on Facebook is to create broad campaign structures (i.e., account simplification) that will supply its tech with as many “signals” as possible, thereby empowering its ML to optimize when, where, how and to whom advertising is served on its platform.
How This Has Impacted Marketers
Now that we’ve taken a walk down machine learning memory lane (MLML, if you will), the varying results marketers have seen from ML adoption across both platforms illuminates the pros and pratfalls.
For Google, utilizing all of its ML capabilities to the fullest extent will often diminish search engine marketing performance. This is due to a combination of the aforementioned granularity with which campaigns can be optimized, plus the fact that machine learning simply isn’t capable of understanding the context behind the data it’s processing. Imagine you’re a retailer on Black Friday, and searches for the products you sell spike several hundred percent, with competitors manually bidding up on relevant keywords to capitalize. If you rely on ML to process this new data and adjust in real time, it’s highly likely you’ll lose at least several hours’ worth of sales volume, which could comprise a material amount of potential revenue for your business.
Juxtapose this with Facebook, which cedes auction priority to advertisers who follow best practices, thereby allowing its ML to process the “signals” being provided. In my opinion, going all-in on ML with Facebook is more advantageous than not doing so. In large part, this is because there’s no edge to be gained by manually overriding the system, thus encouraging more advertiser effort to be placed against strategy, measurement and content production.
The Bottom Line: Man And Machine
This doesn’t mean that machine learning works for marketers on Facebook but not Google. It’s simply a microcosm of the larger paradigm, in which understanding how manual oversight interplaying with machine learning becomes vital. Google’s RSA feature is valuable in terms of conducting ongoing multivariate testing to improve efficiency, while DSA provides additive scale beyond manual campaigns. Smart Bidding, if applied pragmatically, can find long-term efficiencies.
Extrapolating this out further, machine learning provides significant value by processing vast quantities of data to find patterns, drawing conclusions no human is capable of and enabling incredibly precise, data-driven activations for marketers. That said, it isn’t a silver bullet, nor is it a replacement for smart strategies, testing constructs and customer and competitive intelligence. Put simply, the best marketers will continue to find intelligent applications for this technology, neatly integrating machine learning within their larger efforts.