We’ve been conditioned as an industry to think in terms of segments, or rote, easily grasped audience breakouts. These are often defined in advance by the advertisers or publishers and are very limited in terms of reflecting everything we can know about our audiences, all the traits and qualities that would help us truly target them.
Today, especially in mobile, there is a better way, one that considers audience beyond cookie-cutter segments.
A Quick Look at Segments
Before we get into all the inferiorities and downsides of using segments, let me firmly establish I’m actually a huge fan of segmentation. As an experimental social psychologist, I studied and observed that people are predictable.
Part of being predictable is exhibiting correlated and categorical characteristics and behaviors. And, by nature of exhibiting categorical attributes, within the context of the human experience, people can be logically clumped with other people (segments) and uniformly described and anticipated. Even more interesting, at least to the psychologist (and statistician), is just how easy it is to segment, or cluster, humans.
So, if segmentation is easy to do, because humans are so predictable, and if very smart practitioners have consistently validated numerous clustering algorithms, why shouldn’t we use segments in mobile targeting?
Why the Segment is Inadequate
The shortcoming of segmentation is most easily described if you understand how segments or clusters are formed and defined. While there are many different techniques, statistical and otherwise, used to create segmentation systems, for the most part, they all include two sets of variables: clustering variables and describing variables.
So, for example, age and gender might be the variables that determine to which cluster someone is assigned, dining and shopping preferences might be the characteristics by which we evaluate the meaningfulness of the assignment, and how and where we make use of the information.
With this basic premise in mind, you can begin to think about and evaluate the power and usefulness of segments.
Unfortunately, broadly defined segments also have a much lower likelihood that folks assigned to each cluster will be truly homogenous on relevant, usable descriptive characteristics.
For example, do my age and gender alone give insight into my dining preferences? Or, would it be helpful to also know about income, neighborhood, lifestyle, and past dining habits? Of course.
Most Segments are Built in a Generalist Vacuum
Marketers want to use segments in very specific settings. And, in mobile, that setting gets even more specific.
Take the example above. Segments designed to indicate or predict dining preferences. To build a segmentation system that can effectively anticipate consumer dining behaviors, you need more than basic demographics. You need specific behavioral insights.
In mobile, the advertiser who is looking for consumers with a particular dining habit or preference, is really looking for a subset of those consumers, or those who will be prompted by a mobile message to engage in a particular behavior. The idea that any pre-built segment definition is going to have precise alignment with each individual advertiser’s goal is obviously unrealistic.
A real-world example: In the context of the buying and selling of digital media, is the fact that many of the segments we try to leverage in the media industry tend to be based on simplified publisher definitions — crudely categorized demographics and basic lifestyle and behavioral points. The end result is essentially targeting broad audiences, a handful of large segments at a time, and taking a rather linear approach to reaching and optimizing the best audience.
Marketers Leave Opportunity and Money on the Table
Segments, including those used widely across the industry, are by design somewhat static and defined in broad strokes. This means they lack specificity and precision, which in turn means they aren’t superstars at identifying unique audiences, much less even more dynamic mobile audiences.
A more advanced approach to audience targeting embraces the idea of dynamic audience discovery and development, and the ability to simultaneously consider tens of thousands of attributes and characteristics, independently varying across location, time of day, content and device, the key triggers in mobile. So, how does this work?
It Starts with Real-time Learning and Multivariate Models
Real-time learning allows you to actually gather the data you need, relevant to what you are trying to accomplish right now, versus relying on previously built, static segments. Real-time learning with predictive scorecards (versus segment assignments) allows you to continually evaluate and adjust which combination of characteristics are the most important in identifying a target audience.
The result is the ability to save money by only communicating with likely prospects, at a time and place they are most interested, and maximizing return on investment by driving bottom line behaviors with the precision afforded when you can evaluate and modify the most effective targeting and messaging, on the fly.
Take a marketer of organic groceries, who’d like to use mobile to promote seasonal berries and a new store location.
- What are the most important characteristics in identifying the best audience?
- Is there a segment for berry-loving organic shoppers? Is the segment most interested in the grocers message at 9 a.m. on Tuesday, the same as the segment most interested at 6:30 p.m. on Thursday?
- Is the difference between these two segments best predicted by income or occupation?
And, what if the answer to all of the above questions is “it depends.”
Real-time machine learning, mixed with rich data sets and multi-variate scorecards, allows the marketer to ask and answer all of these questions (and more) at once, and then take action (or not) according to all possible answers. In other words, the organic grocer can learn, in real-time, who is actually engaging with his berry and new store ads, and then only spend money on sending messages to other engagers.
If the consumers are engaging at different times, days, or locations vary, no problem, his campaign can adjust and optimize accordingly. Segments can’t ever be that accurate, that precise, or that dynamic.
Header bidding is a programmatic technique that allows publishers to offer their inventory through multiple ad exchanges before they serve up ads from their ad server.
Whatever approach you take to your m-commerce project, one thing is certain: if you want it to deliver the results you’re expecting, context should be front and centre of your design.
As Facebook keeps changing its news feed algorithm, one constant factor is the domination of video content and so brands keep experimenting with ... read more
How are mobile payments, bitcoin, blockchain and other financial services technologies enhancing the consumer purchase journey?