Not All Data Types Are Equally Useful

Big Data is big, in part, because no one throws anything away. It’s also worth noting that "big" doesn't always spell "better" (it may spell “clunky” though). That’s why the concept of Data Lake, a system for storing raw information, came about some time ago, just to keep “useful” data handy. However, determining usefulness of information is a subjective matter. Data that are critical for supply chain or inventory management could be useless for target marketing. The size and color of shoes are important for the former, but for targeting, they may not be so critical. For the prediction of future buying behavior, variables such as style, type, brand, or price would be far more useful. Further, I’d say a summary of personal and categorical behaviors—such as life-time total spending, number of transactions, average amount per transaction, number of product categories touched, number of weeks since last purchase, number of days between purchases, number of opens and clicks in past 3 months, etc. – are very predictable (hence useful). Keep your data close Creating such “individual level behavioral indicators” is often an extra—and often overlooked—step after selecting useful set of data. I call it “data refinement” step, which is a worthy subject for multiple articles. Often, I get questions like “Why would anyone need data refinement steps when well-established Data Lake is in place?” To which I’d say, “Well, because the lake water is still not drinkable.” Data refinement is probably the most important step before analytics regardless of the data source. Before that discussion, nonetheless, let’s talk about what types of data deserve to be kept close for the purpose of targeting and personalization. As I pointed out many times before, personalization isn’t just an option for modern-day marketers, and effective personalization starts with the customer data. Now, there are so many kinds of data that “describe” your target customers. Some could be collected and some could be bought. If I may list a few noteworthy data types: Transactions (RFM) Data / Products & Services PurchasedDemographic / Firmographic (from 3rd-party data providers)Promotion / Response HistoryChannel Usage / Online BehaviorLifestyle / Survey ResponsesPayment / Delinquency History (for subscription/continuity businesses)Call / Communication LogMovement DataSocial Media / Survey / Sentiments Now, I know this list could be overwhelming. Just in the category of CRM data, I can list a few hundred data points. Commercial data providers sell hundreds of variables in categories of demographic, geo-demographic, lifestyle, behavioral, and ethnic data. Just for the top three types listed above, the number of variables may exceed 1,000 easily. And we didn’t even get to online behavior (such as browsing history), and campaign promotion and response data (which could be really big). Winning behavior Therefore, not doing anything until a magical database with all these elements is completed would be silly. Your customers and prospects need to be entertained now, and they really do not care about your data challenges. That means you have to make the most of what you have. Worse yet, if you are still stuck in the data collection stage, it would be great to know what data types are most useful beforehand. From a cheat sheet drawing from over 30 years of examining model documents and profile reports, behavioral data are clear winners. Simply because what people actually do are quite different from what they say they would do or what they look like. For instance, most survey respondents claim to support “Green” causes, as in “Sure, why not?” But when we examine what they have actually purchased in the Green category, personal portraits may look very different. Now, behavioral data in itself is a large category: Transaction data (conversions and purchases)Response data (responses, such as opens and clicks, to marketing campaigns)Online behavior (such as clicks, views, browsing, etc.)Call/communications data (communication logs, positive or negative)Social media behavior (what they say or share) Of these, what would be the most predictable data type? Transaction data—records of monetary exchanges—is the king. Of many ways to cast votes for products and services, monetary payments matter the most. Talk is cheap in most cases. Then, what makes up such transaction data? At the risk of oversimplification, remember these six data types: Who boughtWhat productFor how muchWhen (date/time)Where (channel, store, or device)Using what payment method That’s it! And why do I state the obvious here? Because too many marketing databases do not contain these basic elements in clean and usable forms. If your transaction data is scarce for some reason, then add commercial demographic data (from third party sources) to your data menu. “What they look like” is a passable proxy for “what they’ve been doing.” That is why these two types of data are number 1 and 2 on my list. Once you have these two categories solidly covered, then venture into other types of data and slowly expand your data menu. Remember, not all types or data are equally useful.