As a computer science student many years ago, I learned a bit about artificial intelligence and its applications. At first, I was quite excited by the idea of computers thinking like humans. Then, I quickly realized: computers couldn’t really “think”— at least not the way that a human brain could.
What if—I thought—we gave a computer large amounts of data and compute power. Could it do something that resembled thinking?
I soon learned that it couldn’t—at least not at that time when I was introduced to our largest campus computer, an IBM 1130 that was one of the first somewhat user-friendly mainframe computers built by IBM. This computer that came with a whopping 16k of memory filled the entire room! It needed punch cards to enter data. After lots of late nights at the computer center attempting to get a simple program running, but having to redo it over and over on punch cards (mostly correcting typos and syntax errors), I was convinced the day when computers could rival human brains was not going to be anytime soon.
Fast forward a few decades. The significant evolution of computing power is reaching a tipping point where computers’ speed and capacity might rival—and even exceed— the capacity of human brains.
To put in perspective how rapidly computers have evolved, I am wearing, on my wrist, a computer (a Fitbit fitness tracker) that is the size of a key on the IBM1130, with 1MB of memory and processing power vastly exceeding the capabilities of the IBM mainframe.
Still, the success of machine learning in mimicking human learning has been happening in fits and starts. The iPhone voice recognition feature “Siri” that uses machine learning can make you feel like it is the most amazing invention one day, and make you want to throw your phone across the room in frustration on another day.
Hollywood often seems to see the future before we do. Movies like Short Circuit, I, Robot and many others project scenarios involving “intelligent” computers and robots that either help humans or become so smart as to turn humans into their slaves. The movie Minority Report describes a future where a human brain wired to a computer can predict bad events that are going to occur, as a way to warn and prevent them from happening.
While the various versions of the future portrayed in these movies may or may not be our future, we cannot ignore the reality that, increasingly, computers and robots are steadily becoming an integral part of our lives.
In my recent travel to Australia, I was absolutely amazed by the immigration check at the airport—done entirely by a computer-based facial recognition system. I simply put my passport in a slot, faced a camera and, within a few seconds, the computer matched my face to my passport picture and gates opened. No humans whatsoever! What’s fascinating is the machine’s ability to recognize a face, even one that may be tired, jetlagged, happy, angry, sleepy, shaven, unshaven, unkempt hair, with or without makeup. This was, until recently, something only humans could do. But, there are times when even humans hesitate as we see someone we haven’t seen in years and wonder: “is that really you, Natalie?”
You are probably thinking: okay, great, but what does this have to do with digital marketing. Well, everything.
Digital marketing suffers from a significant challenge today (okay, maybe more than one).
Many marketers are still focused on so-called “reach and frequency” goals, causing them to broadcast the same message to everyone over and over again, irritating consumers and causing them to try at all costs to avoid advertisements, for example, using ad-blockers. Why are marketers still stuck with this one-size-fits all mentality? Perhaps many marketers are stuck in the era of TV and print advertising— and just can’t seem to see how digital marketing truly offers an opportunity and indeed an imperative to conduct 1:1 communications with consumers. The other reason may simply be that taking advantage of digital marketing’s ability to connect with consumers 1:1 is a lot harder than it sounds. Programmatic media buying platforms, while they have brought automation and scale to the marketing process and for targeting audiences more accurately, are simply becoming like large speakers attached to really bad music. The creative and messaging side of marketing has had zero to very little automation. In this sense, we haven’t really improved the quality of the music; we’ve just made the music louder.
Understandably, making better music in marketing by creating and delivering relevant and precise messages tailored to the individual is much harder. How is a marketer to know whether or when an individual is likely to respond to a specific message, or even what message or product to deliver or recommend to a consumer?
We could, of course, try and program computers with every possible scenario. For example, a consumer browses a particular outdoor product. It is almost the weekend. The weather forecast is rain. Based on these conditions, and known consumer interest (like hiking and trail running), the advertiser could then offer an ad showing a pair of boots with a certain message related to hiking in the rain. Another example: It’s the middle of the week. The weather was sunny. The advertiser can offer the individual a different ad with a different product and message for outdoor running in warm weather. You get the idea. There are many, many possible combinations of behavior and context related to a consumer’s purchase intent.
This process of determining the behavioral and contextual profile of an individual and then delivering customized messaging to them can quickly become very daunting and expensive, so the obvious question becomes: “Can we have computers learn how to do this on their own without having to be programmed”?
One way to answer that question is to ask ourselves, how do we humans learn to do the “right” things in the right context? How do we know, for example, when to bring flowers, versus a bottle of wine, on a date? Do we program a child with these kinds of rules? How do we adapt quickly to learn the date is allergic to flowers or wine, and take a cake instead?
The answer is: We learn from others, do it ourselves, make mistakes, and get corrected. This is exactly how machine learning works. Continuous learning from vast amounts of data allows machine learning algorithms to closely mimic the human brain. Machine learning often works best when the computer is presented with very large volumes of data and when the algorithms are exposed to what I call “outcome data,” for example, the mistakes that algorithms make. This is referred to as developing a machine learning “training” model. When done with sufficiently large data sets, the models become so accurate—they mimic and often exceed the capabilities of the human brain.
Machine learning algorithms can allow marketers to process vast amounts of consumer data, learn from it and predict the best-fitting product, message, and timing of delivery in context. The result is the reward of higher consumer engagement. Today, with the power of cloud computing, inexpensive hardware and vast amounts of memory, machine learning’s boundaries continue to expand. Problems we never thought could be solved are now suddenly falling within the realm of possibility.
For example, marketers have for a long time segmented their users into “audience buckets.” This is done because such audiences and campaigns targeting at them are much easier to manage at scale. The problem, of course, is that the audiences themselves are very broad, making this a repeat of the traditional “one-size-fits-all” marketing mentality. Are we effectively saying that all those people in each audience behave exactly the same? Do all Millennial Moms behave the same? Of course not.
To identify individual moms, their preferences, and things they are likely to respond to is hard work for humans. Not so much for machines, especially aided by machine learning. Machine learning algorithms have the potential to process the massive amounts of data needed to find the nuanced differences between people—much like facial recognition algorithms do to create micro-audiences—via a process called clustering.
If machine learning can identify such micro-audiences and further identify their preferences and habits down to individuals, imagine the results we can achieve by combining that with the machine learning’s ability to determine the creative or message that such micro-audiences are responding to best. This creates a powerful combination that is, in essence, defining the future of marketing: data-driven, personalized and relevant.
We recently witnessed the power of machine learning in digital marketing when the Jivox IQ machine learning algorithms (dubbed Neuron) outsmarted a CPG brand manager. Neuron demonstrated that Time-of-Day-specific messaging outperformed audience-based messaging, as consumers showed higher engagement when they saw, in the morning, ads with cereal paired with messages about healthy breakfast.
As intelligent as machines could potentially become, would they take over marketing and put all marketers out of work? The short answer is no. Marketers would, instead, be spared from the drudgery of analyzing mountains of data and poring over piles of reports in an attempt to deliver the right message to the right person at the right time. Marketers will instead spend more time on creative strategies, media strategies, and understanding consumer behavior. Data, which is the fuel for machine learning, will become a marketer’s friend. Very soon we will have a lot less of the drudgery and a lot more golfing, while the machines slave away at our offices, delivering brilliantly performing campaigns.