Sign up ... it's free. Join Now
Marketers Know AI Is the Future, But Do They Understand AI Today?
This article is part of an occasional series from leading voices about key issues facing marketing today.
Here's a quick reality-check for the next artificial intelligence (AI) pitch you hear: Ask what the company's solution optimizes for. If the answer is along the lines of "anything you need," that should raise a red flag.
I began working with AI as a teenager, taught in the field at Harvard and MIT, and wrote books on the subject. Breakthroughs in the field since I wrote my first book at age 16, How to Build a Computer-Controlled Robot, have been extraordinary. In many ways, our present is a version of the future described in the science fiction novels I read as a kid.
But innovation never moves at the pace of fiction. And reading today's breathless headlines about how AI will completely transform marketing and advertising overnight makes me worry that advertisers are being taken for a ride.
So, let's put aside the fiction and focus on the facts that matter to the industry today.
I keep hearing about AI, machine-learning and deep-learning. Explain.
AI began with the idea of programming a machine to demonstrate intelligence. Today, AI has become the umbrella term for many kinds of algorithm-based solutions to finding patterns in data. For example, you could write an algorithm that describes the features of a cat and then program a machine to recognize cats.
Machine-learning, which is a subset of AI, is about showing patterns to a machine and deriving algorithms that allow machines to learn from those patterns. So, instead of programming rules into a system, you create a learning framework whereby the computer finds patterns. In that scenario, the machine discerns the nature of a cat by looking for patterns in cat pictures.
Deep-learning is very similar to machine-learning, but with a notable exception: Instead of giving the machine the answer (this is a cat, and here's why), the machine looks for deeper patterns, which may not be obvious to people. Here, the machine learns what a cat is by identifying, testing, and learning abstract patterns in images of cats and non-cats.
As you might have guessed, I picked cats for a reason. Though machines are indifferent to felines, humans love to watch videos of cats. No news there, right? But the important takeaway is this: Today's AI cat-recognition capabilities are the result of more than a decade of innovation. Indeed, it's no accident that the history of this narrow but deep capability grew as did YouTube, which began operations in 2006; that capability has enabled the platform to filter and give users the content they want most (like cat videos).
More-powerful computers and faster connections are the enabling technologies in this case.
Got it. But robots can drive cars, so why aren't they running today's advertising industry?
AI is good at solving narrow and deep problems. Identifying a cat is a good AI problem because the goal remains clear and consistent over time, allowing the machine to learn and optimize. On a much more complex level, the same is true for self-driving cars. Once considered the pinnacle of AI capabilities, self-driving cars are really just thousands of narrow and deep problems organized into a single framework. In that sense, a self-driving car is the result of thousands of cat problems being solved in real-time, which means exponential increases in processing power (Moore's law) have been the key to self-driving cars.
Advertising challenges largely relate to issues of culture and human psychology. These aren't the narrow and deep problems of cat recognition and automated driving—because people are complicated, and our goals are both fluid and difficult to articulate. So, although the technology exists for a computer to create a short film, there's no reason to believe that an AI creative director is on the horizon. Nor will CMOs be replaced by software.
But advertising does present some goals that are ripe for AI disruption. The key is that you need a clear, unambiguous, and consistent business goal to tie to AI. Purchases are the ideal business goal because they're binary—people either buy, or they don't. With purchases as a fixed starting point, machines can discern patterns that elude humans, and from those patterns advertisers can mine a wealth of actionable insights. Taking it one step further: purchase intent also represents a good AI challenge; but, of course, there are more variables in this instance than in measuring a purchase, so it quickly becomes more complex.
Engagement, which is the key to an attention-driven practice such as advertising, can also benefit from AI. But it's critical to understand that the benefits of any AI optimization depend on how you currently measure engagement. For example, if an impression is your proxy for engagement, you can optimize with AI all you want, but you're not necessarily going to better understand how people are truly interacting with (or feeling about) your content. That requires more sophisticated measurement, which can't be attained simply by counting impressions, so proceed with caution. Because the more you use AI to optimize the wrong metrics, the further off course you'll go.
Point taken. We're using the right metrics. How can AI help our campaigns today?
The success of an advertising campaign depends on a strong understanding of business objectives, a great match between the ad creative and the intended audience, and a clear understanding of the context in which consumers see your ad. Failure in any of these areas will lead to the failed use of AI. However, AI can optimize for clearly defined business objectives, but only if they can be articulated with precision, and measured and optimized over time—meaning that AI won't change your campaign overnight.
Let's take the interplay between creative and audience as an example. Great creative can be in the eye of the beholder, and AI can be used to objectively validate creative's effectiveness as a campaign runs. That's a good thing. We need to move away from HIPPO (the Highest Paid Person's Opinion) and make creative choices based on real-time engagement data. That said, a typical campaign is short, putting a hard limit on optimizing the interplay between creative and audience to drive engagement.
Simply put, AI needs time to work, which means the biggest insights will come from optimizing engagement over months—even years—of rigorous testing.
But there's another important factor to consider—context.
An example of context: if a self-driving car doesn't know if it's in the US or the UK, there's no way to optimize for safety; without context, the car won't know if it's supposed to drive on the right side or the left side of the road. Similarly, in advertising, if the goal is to optimize around engagement, the AI needs to know the media context—something that's impossible to do inside the walled gardens that attract the majority of today's digital ad dollars, since the content is user-generated and not vetted by editors.
To be blunt, that means you can't use AI to optimize context inside walled gardens, even if you've clearly defined your business objectives, articulated them with precision, found a way to measure, and given AI the time it needs to work. Therefore, it's important for marketers to think outside of the walled gardens if they are to take advantage of AI for campaigns; in doing so, they have more control over the campaign's placement and context, which leads to more positive business outcomes.
So, context is key?
Yes! In fact, AI without context isn't intelligence at all. But AI without context does explain how advertisers end up with clunky placements and embarrassing brand-safety issues.
As with any tool, the key to AI is how and where advertisers put it to use. If well executed, it can be a powerful tool that revolutionizes the way advertising works. If not, it will lead to the continued spread of irrelevant ads and breakdown of consumer trust. The choice is ours as to which way we decide to go.