One of the hottest storylines in ad tech this year has been artificial intelligence (AI). Proffered as potential panacea for effectiveness, brand safety and transparency, AI has grown from niche discussion to industry obsession, a promised key to smarter digital ad targeting and trading. I’m sure you’ve seen the pronouncements: it can determine the bids most likely to succeed, it can use historical performance data to tailor campaigns and it can even swap out creative based on audience data in real-time.
There’s one issue: a lot of what’s being hyped isn’t actually AI. It’s just tools and technologies being marketed as AI in order to differentiate within a complex and competitive arena. With true AI, a machine imitates intelligent, and maybe even sentient, human behavior. And while much of what we see today looks like the computer is thinking for itself, it’s really just following very specific, pre-programmed paths using simple rule-based actions, and/or predictive analytics or machine learning. While all subsets of AI, even together they don’t add up to real AI. They’re more like “artificial AI.”
It may seem like semantics, but there are important differences. With predictive analytics, patterns in existing data are used to predict probable results and trends in the future, typically using statistical models and methods. Then, there’s machine learning, a branch of artificial intelligence where machines learn and adapt through experience, without the need for predetermined rules and human intervention. With machine learning, models and techniques will change themselves over time as more classifiers enter the system and improve the description of the data to be learned. Examples of machine learning classifiers are K Means Clustering, Linear Regression, Logistic Regression and Decision Trees. These techniques are being used in technologies today for things like facial, voice, music and handwriting recognition.
While not using true AI, the market-available technologies for programmatic advertising that we have today are still sophisticated. They effectively use machine learning and data science-based systems to predict the likelihood of desirable outcomes. They’re certainly valuable for advertisers looking to optimize their media budgets. Further, the automation they allow creates huge efficiencies. They do not, however, fulfill the promise of a set-it-and-forget-it system that gets better or more accurate without any human intervention.
When real AI is finally applied to advertising, it will be transformational. It will intelligently enable desired outcomes to be produced by calling on not one, but a collection of interrelated sciences, techniques and data processing. One day, a truly intelligent AI-based advertising system will enable buyers to seamlessly construct their entire campaign, complete with optimized buys and evolving tactics, just by specifying their goal(s) and budget. Once the algorithms take over, the system will leverage historical data about similar campaigns to make predictions and changes on the fly.
This is all achievable. But supporting it requires highly complex systems to come together, and we still have a long way to go based on today’s fragmented, disconnected assortment of pseudo-systems that look fantastic in isolation, but, in aggregate, don’t add up to the holistic system that buyers need.
Eventually, AI will evolve to where it can improve programmatic media and create a better user experience. And we will eventually get to the point where technology can drive ad campaigns that, without human interaction, achieve campaign KPIs through a virtuous circle of measuring, analyzing and acting on campaign spend, allocation and outcome variables.
In the meantime, platform vendors intent on presenting today’s “artificial AI” will accomplish more by being open about the realistic expectations of their products and the fact that their capabilities are that of early-stage, partial expressions of AI. At the same time, we as an industry should raise our level of thinking and education, so we gain an accurate understanding of AI — not just what it is, but what it can do for us. It is with that knowledge that we can begin to see the true potential of real AI.