Connecting Online and Offline Customer Profiles to the Real World

When it comes to precision targeting, there simply is no substitute for modeling. Many marketers have descriptions of their customers in their minds, and some might have completed extensive segmentation work to back up their assumptions. Some will even go further and try to select target prospects based on such descriptions, which may read something like: “Our target is affluent young millennials who are highly educated, environmentally conscious, and digitally savvy early adopters.” Such descriptions are always helpful when you are developing marketing narratives. Gaps between segment descriptions and targeting rules Trouble ensues when trying to convert such qualitative statements into actual selection logics. Millennials? How old are they exactly? What about the folks born in early 1980s who barely got classified as such? This is a wide age range—will you break the category up into young and old millennials? What would be the dividing line between the two? Affluent? By what standard? If we simply choose to use household income, what value would be considered high enough? $100,000 per year? $150,000? Why not $300,000? What about their expendable income, which may vary greatly from region to region? What about wealthy retirees with limited income? What about their home values? Shouldn’t we consider socio-economic status while at it? How Personalization Helped Alex and Ani Boost Conversions 65% It gets even more challenging when you encounter descriptions like “environmentally conscious,” “digitally savvy” or “early adopters.” How would you define them only using first- and third-party data (that are commercially available) without surveying everyone in the country? Such qualitative descriptions should be translated into model-based personas to be actionable. Rules have serious limitations Translating target descriptions into selection logic is a priority in the marketing industry. When forming your plan, follow these three points: We have over 1,000 variables in our database, and over 500 variables in the third-party data source. Which one would you pick to describe your target? Surely not just age and income?If you do choose age and income as the starting point, how young is young enough, and how rich is rich enough? Where would you draw the line and based on what guiding principle?Let’s throw a few more variables into the mix, such as children, dwelling type, length of residence, etc. Are all these variables equally significant? (They never are.) Using only a few popular variables in this age of abundant data is an oversight. But even if you only have between 10 and 20 variables to choose from, you'd already be in need of modeling. Can you simultaneously consider even just four and five variables and intuitively comprehend interactions among them? When it comes to hundreds of variables, that is clearly a job for computers, which would yield algorithms, then produce scores. Marketers could then go after highly rated individuals first, then move on to the next score group. Continue to expand the contact universe until marginal predictive power diminishes. Any gains chart would clearly show where the line should be drawn. With selections rules? You wouldn’t even know how big your initial target universe would be until you actually run the counts, which often turn out to be too big or too small. And when you eventually exhaust manually manufactured lists, good luck creating a new set of rules that would be as effective yet distinct from the previous ones. Separate targeting and messaging There is of course a need to maintain consistent segment descriptions. Shared by all participating teams, they are essential to communication strategies. But do not insist on creating selection rules that exactly reflect such descriptions. For one, the test environment in which those segments were created may immensely differ from the general population. More importantly, available variables in the larger universe could be immeasurably different from the ones in the controlled environment, unless you are willing to spend big money to survey the whole country. Instead, employ bridging models, which would fill the gaps between the two worlds. Once you define the target in the test environment, build a model only using commonly available data (such as third-party data). The resultant model will project “likelihood to resemble the target” to much larger universes. You may feel not agree with the resulting description. But marketers should really differentiate targeting and messaging, because usable data may be in vastly different shapes in each stage. Always remember this two-step approach: Define who to talk to first, then decide what to say to them. Of course, all based on data. 7 Digital Marketing Strategies to Boost Lead Generation