(In case you missed it, check out chapter one where Dstillery defined these terms.)
All marketers hold a set of beliefs as to who their customers are and how they behave, supported by years of studies and focus groups. Market researchers have long recognized the limitations of this self-reported behavior and worked within the constraints of customers’ honesty, memory, and introspective abilities. For instance, people aren’t always comfortable answering a pollster’s questions honestly, a phenomenon exemplified by the polls leading up to the 2016 presidential election. People don’t always remember exactly why they took specific actions or made certain decisions, and we are notoriously poor witnesses to our own behavior.
In our digital age, self-reported behavior is a small part of a very large picture. Observed behavioral data allows marketers a view of actions that are taken both online and offline – captured from anonymous web and mobile device IDs. Combined, this data paints a robust picture of the customer journey, giving marketers the power to more accurately predict how people will behave.
The sets of rules that govern a computer’s behavior are known as algorithms. All computer programs follow algorithms of varying degrees of complexity. Machine learning algorithms are a specific subset of algorithms which instruct a computer to comb through data and draw insights.
The strength of algorithms lies in their ability to learn from raw information and determine which data is relevant. There are many different machine learning algorithms, you may have heard of some, like deep learning, neural networks, or k-means clustering. Each solves different problems in unique ways. It’s the machines, not humans, that decide on the weight to assign properties.
Supervised vs. Unsupervised Machine Learning
Two broad categories of machine learning algorithms are supervised machine learning and unsupervised machine learning. Generally speaking, we use supervised machine learning when we require the data to answer a specific question, whereas unsupervised machine learning is designed to uncover patterns that we otherwise would not.
If you have a smartphone you’ve already experienced both types of machine learning first hand. Smartphones use supervised machine learning when you teach it to recognize your thumbprint in order to unlock it. Your phone employs unsupervised learning when it automatically groups your photos by the people who most frequently appear in them – all without any direct input from you.
By understanding when to deploy these algorithms, marketers can effectively discover and scale new consumers.
Supervised Machine Learning for Marketers
With supervised machine learning, a data scientist must provide the algorithm with numerous examples of what the marketer wants to learn. This trains the computer to perform the specific machine learning task.
Let’s say you wish to identify high-value prospects. This happens to be an ideal application of supervised machine learning. An algorithm can examine attributes and behaviors of your existing customers. Looking back at their online and offline behaviors, it could then identify relevant patterns that may not be obvious signs of being in-market for a particular item or brand affinity. Once the machine identifies patterns that indicate near-term interest for a product, it can understand what makes someone likely to be a new customer for your brand.
The algorithm needs a data scientist to teach it the difference between “this is a consumer who purchased sports apparel” and “this is a consumer who did not purchase sports apparel.” In AI terms, the examples the data scientist inputs are known as “training data,” and the consumer’s actual behavior are “outcomes,” which can either be positive or negative. The algorithm uses this training data to assess the various attributes of a consumer — sites visited, for instance — to determine the probability of a consumer converting.
We can address this with a supervised learning algorithm because the business problem “find potential converters” can be translated to the yes or no question “is this individual user likely to convert?”, and because we have data available to teach the machine what a “yes” and a “no” look like.
But sometimes the question is more complex. Sometimes you want the data to go beyond the questions you know to ask. This is where unsupervised machine learning comes in.
Unsupervised Machine Learning for Marketers
With unsupervised learning, the algorithm looks for patterns in the data, without the explicit direction of what in particular it’s supposed to be looking for. This allows marketers to segment a population of existing and prospective in-market consumers and uncover unexpected opportunities. The strength of this approach lies in the machine’s ability to withhold any preconceived notions of who marketers think their customers are. By analyzing consumers’ observed behavior data and detecting which users are most similar, the algorithm can uncover and group these act-alike users — the same way your phone can group similar images of faces.
As a result, unsupervised learning can be used to either validate or disprove expectations that marketers inevitably bring to their research. By answering questions marketers didn’t know to ask, it can find micro-audiences they couldn’t know existed. For example, a sports apparel brand might discover that, in addition to the obvious athlete subpopulations, they also garnered significant attention from video game enthusiasts.
By harnessing the power of machine learning, market researchers drastically increase their understanding of who their customers are and how best to engage them. Doing so will not only improve marketing efforts, but also corporate strategy and customer experience. In our next chapter, we’ll show you how.