Closing The Gap Between Marketing And AI Systems

As a professor of machine learning, I often get asked how I think artificial intelligence (AI) will change the world we’re living in. To answer this question, I usually find it easier to speak about one of the faster-moving fields of AI—digital advertising, in which AI has been around for quite some time but has always sat on the side of big tech. Now, as technology is becoming more available, you can see how it can potentially change the way we do things.

To paint a picture, in the world of marketing, programmatic systems are a part of everyday life. Through them, the likes of Facebook, Google and TikTok have allowed marketers to reach billions of people, providing these opportunities for businesses both large and small. Today, thanks to AI, even a small business can achieve great success through a global audience, democratizing this opportunity for those willing to work for it. This has given rise to an industry called performance marketing, in which the goal is to use these systems as efficiently as possible to maximize a company’s marketing ROI.

But there's a paradox in the world of performance marketing. For an industry that aims to give the best performance, we tend to do it in a very inefficient manner. There are various business-related reasons for this that are hard to fight against but often brought about by a misunderstanding of how AI works.

One common misunderstanding is the term "audience." It’s widely known that big tech profiles every person who uses their systems—a crucial component of how the internet is able to function in today’s world. But when it comes to understanding those audiences and reaching them, there always seems to be a gap because the way a machine understands a person isn't the same way people understand people.

The human mind is amazing, as it can take very abstract concepts and simplify them. It’s an innate ability that allows us to understand the world around us. So, if we, for example, were to describe a friend who recently had a child, we could easily condense all our experiences with that person and probably sum it up as “the family got a cute baby boy.” This impression would then affect how we would interact with that person.

In advertising, we build communications by applying a similar way of thinking: We condense abstract concepts of human truths and use that as a way to guide what we say. When it comes to media targeting, we then simplify this abstraction into a “person with a child,” hoping the system will find the right people. But, a machine can't tell if a person has a child. Rather, he or she is just someone who has pictures of a baby, a small person, a monkey (because machines tend to mistake monkeys for children), diapers, a boat (because inflatable boats can look like folded diapers to machines), a duck (a common toy and design on baby clothes) or tiny clothes. Because of how we think of our audiences, among the seven of those I mentioned, only one or two are very good indicators and just a couple more could make sense in retrospect. Our abstractions don't work well with the systems that we use today. As a matter of fact, it’s counterintuitive for a computer.

This is a problem with performance marketing because if we keep falling into the logic of how we describe people as humans to computers, we start describing our audiences the same way as everybody else and don’t fully utilize its capabilities. This can create unnecessary competition and drive ad costs up. So, to truly understand how we should target people, we should think like a machine, which, of course, is easier said than done. It's hard for us to comprehend the dimensionality of big data, but we could use machines to help us make decisions based on data.

For example, our client, Sportsmaster, wanted to run a Facebook campaign with two iterations: a version made by a well-established international media agency and another made and run through AI. After conducting stringent research on the audience, the agency placed them into 86 groups—a high mark compared to many other campaigns. This drove about 23,000 app installs, which, based on benchmarks, was quite good. The AI campaign, on the other hand, required less manual labor and created 345 audience groups, which garnered 140,530 app installs—over six times the results.

For the AI strategy, we used long-tail interest targeting—a machine-assisted strategy by which AI deciphers a business' existing social media audiences to create and recommend potentially thousands of interest combinations. This method can automate campaign management, reveal more granular information about audience campaigns and help increase brand performance.

To properly handle the systems that we have today, businesses need to understand and use machine learning themselves. This, in itself, is a paradigm shift in how some industries work because communications isn't just an art field as it previously was but also very much a science field. Knowing how to understand how a machine thinks is a skill that might be as important as knowing how to work well with your coworker.

What different industries are doing now is figuring out how AI sits within their cultures and ways of doing things. Of course, there are many who think that AI will replace jobs, but what AI is really doing is freeing up time so people can do the things they do better.