Marketers and brands need to continue pushing for transparency as machine learning technology develops.
Machine learning is getting a lot of buzz in the digital and search marketing world, not all of which has been good. In one camp, you have marketers excited for a future where machine learning and automation drive everything. In the other, those predicting doomsday scenarios in which human marketers have been made obsolete by infinitely faster and more capable robots.
In reality, the future of machine learning will likely proceed along a middle road that splits the difference between the tremendous benefits that result from well-managed A.I., and the total failures that can result from improperly deployed A.I.-driven solutions. During my decade working in search marketing, I’ve experienced real-life examples of both.
As it exists today, machine learning is far from perfect, and it’s the agency’s responsibility to ensure that it works both for us and for our clients — not just for its developers — as it continues to mature.
Ultimately, the goldilocks solution will seamlessly integrate human insights and machine-learning-based automation. But in order to get there, we’ll need to continue testing and challenging machine learning technologies, and ensure we’re getting the level of transparency that’s needed for marketers to retain control over their clients’ interests.
The Argument for Machine Learning in Search
As things stand today, search marketers are rarely able to harness the full potential of the data to which they have access. The increasing complexity of the search marketing landscape has made it difficult for marketers to find time to translate data into real-time action. The sheer volume of data would be hopelessly overwhelming were it not for machine learning solutions designed to ingest and analyze it.
From a practitioner’s workflow perspective, nearly every step of the campaign management process — from creative generation to bidding — could do with some streamlining.
As the machine learning trailblazer, Google has made significant progress in creating functionality to address these challenges across their ecosystem. Responsive Search Ads and Ad Suggestions help accelerate creative testing, Data Driven Attribution helps disperse impact beyond the last click, and advanced bidding models consider hundreds of user-specific data points in real-time decision making.
Machine learning can take over tedious, time-consuming tasks and perform them better than human marketers can, leaving marketers with more time for strategic work — that’s game-changing stuff.
The Challenge for Search Marketers
At the same time, blind adoption of machine learning solutions isn’t in anyone’s best interest. Today’s machine learning toolkit from Google defines success as it can be tracked by digital analytics – more efficient clicks or conversions as defined by a conversion pixel or tag. But for advertisers who define business success by offline actions, online KPIs such as Click-Through Rate or Cost per Download only tell part of the story. Although it’s possible to ingest and track offline data against digital metrics, this can be difficult to accomplish, particularly for smaller advertisers who do not own the point of sale.
Scale represents yet another limiting factor. Campaigns supporting a small population size (such as a rare disease), websites with low engagement or incomplete conversion tagging will have difficulty finding success via machine-based algorithms, which require large sample sizes to inform real-time optimizations.
In addition to these functional gaps, today’s machine learning toolkit has philosophical gaps as well, and ultimately doesn’t offer full visibility into the underlying factors driving performance. Insights derived from automated bid optimization, which could theoretically help inform media mix, targeting and creative strategy, can be lost without sufficient transparency. In short, marketers need a window into what’s going on inside the “black box” of their machine learning tools.
The Path Forward
Marketers are right to be optimistic about machine learning’s potential, but optimism alone isn’t enough — we need to continuously test, learn, and push for more.
To this end, PHM is taking a proactive, methodical approach in the implementation of machine learning technologies. We’re implementing bid rules and strategies only where KPIs are well defined, tagged and at scale. We’re partnering with brand and regulatory teams to approve responsive search ad format concepts, and creating new submission templates. We’re observing Data Driven Attribution models to understand variance and potential impact.
As we continue to move down this path along with clients and technology partners alike, PHM will continue to advocate for more transparency and ensuring that the inputs driving the machine learning’s decision-making align with the success measures our clients strive for.
For machine learning algorithms to reach their full value potential, they must empower marketers to observe and understand their logic.
That’s the bright future that machine learning promises. Now it’s up to marketers to make it happen.
. . .
Pete Levin, VP, Search, is responsible for driving approach, best practices, and partnerships across PHM’s paid search practice of over 70 paid pearch marketing professionals.
Pete started managing paid search campaigns focused on lead generation and eCommerce in 2006 after graduating from the University of Massachusetts’ Isenberg School of Business.
Since joining the PHM team in 2011, Pete has supported enterprise clients in both CPG and Pharma verticals, with a passion for fostering innovation, experimentation, and team-building.
Connect with Pete Levin on LinkedIn.
. . .
Questions? Thoughts? Ideas? Contact us.