In the world of real-time, personalized customer experiences that drive revenue, it is fashionable to think of segmentation as a relic of a bygone era, kicked to the dustbin of history by the concept of one-to-one marketing.
While it is true that customers expect omnichannel personalization and for brands to recognize them as individuals with unique preferences and behaviors, segmentation is still a valuable tool that allows marketers to target an individual customer with a next-best action based on what the data says will produce the desired result as opposed to basing an action on gut instinct or a hunch. An organization might, for example, want to understand how likely a customer is to churn. A machine learning propensity model trained to find commonalities in the data will produce segments that reveal this insight. Segmentation does not preclude personalization; even customers with the same likelihood to churn may be presented with a different next-best action.
Adieu, “Batch and Blast”
Before expanding on the role of segmentation to derive actionable insights from customer data, first some background on why segmentation became a popular marketing tool. Historically, the adage of “the more the better” held true as far as customer data in an organization’s database. “Batch and blast” campaigns were the order of the day; for email, direct mail or other channel-based campaigns, responses were a numbers game – the more you sent, the more responses you’d get.
As the volume of data and the number of channels increased, however, drawbacks emerged – beginning with cost and inefficiency. A response rate may have stayed static at, say 10 percent, but the company paid more for the same return, with more unresponsive customers. In addition, more channels and different ways for a customer to move through a customer journey meant more customers who weren’t being reached at all, further diluting the effectiveness of a channel-centric campaign.
Limitations of Basic Segmentation
For a simple, illustrative example, consider a neighborhood pizza parlor that sends an email to every customer promoting a free dessert for any party of three or more patrons. The email’s hero image shows a meat-lover’s pizza next to a frosty mug of beer and a bowl of ice cream. Without segmentation, the offer is immediately irrelevant to any customer who regularly dines alone or as a couple, or for those who are lactose intolerant. Maybe a party of three – a couple with their child – would love the free dessert, but the parents are non-drinkers, or they’re vegetarian and they’re put off by the hero image. The promotion loses more luster when the establishment opens another location or adds delivery and take-out options.
Even basic manual segmentation will mitigate many of the drawbacks associated with a blanket offer. The free dessert promotion goes to customers who have previously purchased desserts. A different 2-for-1 large pizza offer goes to those with an average order of $40 or more. Segmenting out an audience using only transactional data, the company is more relevant and more cost-effective.
Recognizing that a small pizza shop will all but certainly not utilize advanced segmentation techniques, the concept itself applies equally to enterprise businesses. Marketers experimenting with segmentation quickly realize that customer data has endless stories to tell. Working on a hunch, or gut instinct, marketing teams can come up with any number of ideas for how to manually segment an audience, and target a segment with tailored content. College students are offered a free appetizer on karaoke night. Seniors are offered an early-bird discount.
The more data, the more stories – and the more difficult it becomes to decipher those stories manually. A hunch, after all, can only be based on a finite number of data points. With more data points to consider, a hunch has less chance of hitting the mark. The pizza parlor will be less confident, in other words, that a manually created segment reveals what’s most meaningful about the customer.
Enter Machine Learning
Trusting that an audience segment reveals what’s meaningful about a customer is where unsupervised machine learning models come into play. Unsupervised machine learning models are used to discover data correlations that reveal something interesting about an audience, and are a primary use case for audience segmentation.
An unsupervised model will segment an audience based on what – according only to the data – is important, interesting or unique about a particular audience. The model exceeds the threshold of what’s realistic for marketers to divide an audience manually, finding patterns in the data that may not be apparent to the naked eye. In the pizza parlor example, machine learning models might be used to find a correlation between a customer’s physical address and the time of day the customer visits the establishment, the frequency of visits and order size, of the number of pizza toppings and how many soft drinks a customer buys.
The use of machine learning to find patterns in customer data does not preclude a personalized experience, or what is often referred to as one-to-one marketing. Rather, it allows a marketer to deliver a hyper-personalized experience the marketer can trust is meaningful and relevant to the customer because the data says it’s meaningful and relevant.
Editor’s Note: The next blog in a series on segmentation will focus on why dynamic audience segmentation is important for creating segments that are not tied to a channel.