Not All Customer Data Is Equally Valuable

All major projects must begin with clear objectives. In the age of abundant data, what is the point of building yet another data depository? Even when the goal is narrowed down to a customer data platform (CDP), the structure of it may widely vary depending on user types and related marketing channels. And it should never be about fulfilling everyone’s wildest dreams, either. The process isn't different from any other major development projects: If it's about an automobile, what kind of car are we talking about? Surely, no company would just design a new car without answering that question. If the company simply takes everyone’s wish list and try to satisfy every demand, they may end up having to build a truck that drives like a sports car, which will be extremely costly. Likewise, I’ve seen too many overly ambitious—and lengthy—database projects that became irrelevant by the time they were completed. Data initiatives are supposed to yield results even during the development phases. That means proper prioritization of user demands must be the prerequisite. Going Beyond Modeling to Find New Customer Acquisition Targets I am not at all saying some user demands are not worthy of consideration; however, the champion of the project must be able to say “no” or “not now.” Otherwise, the project could just drag on, never reaching the finish line. Proper prioritization of functionality is important, not only for setting appropriate development phases but also for data ingestion planning. Granted there are all kinds of customer data, what would be the essential ones for the initial phases? Allow me to share the types of customer data that must be considered—knowing the nature of disparate data sources will make it easier to prioritize them. Personally identifiable information (PII) The “who” is essential, as the keyword in CDP is the customer. Without the PII, establishing an individual ID is not possible, even with state-of-the-art ID graph solutions. Always try to collect and maintain name, address, ZIP code, phone number and various member IDs. With PII, one can readily append all kinds of third-party data to each record for profiling and targeting. Transaction data Customers’ past purchase history is the most powerful predictor of their future behaviors. Any experienced marketer knows the power RFM (recency, frequency and monetary) data. In the hands of statisticians or AI, transaction data becomes even more potent. The only downside is that they would be missing for prospects who haven’t purchased anything yet. Nevertheless, I would put this at the top of the list. For one, they are very useful while relatively small in size, and two, most transaction data is clean out of payment or accounting systems. Definitely the lowest hanging fruit. Product- or SKU-level item data Along with the transaction data, “item” information is very important. Handling them is slightly complicated, as customers may have purchased multiple items in one transaction, and SKU-level data may look like free-form product descriptions. But once they are cleaned up and categorized—hard work, no doubt—SKU-level transaction data becomes a potent driver for personalization efforts. Transaction channel There are many online and offline channels nowadays, so combining data from all transaction sources often becomes the primary goal of CDP initiatives, in the pursuit of true “Customer 360.” The key is to go after the most significant sources in terms of sales dollars and not to wait until you have everything in one place. Besides, some partners may never relinquish marketing rights of collected data. Response history To find out what elements of marketing efforts actually worked, data in terms of “who responded to what offer through what channel” must be maintained consistently. This type of data could be quite sizable, but if you match the promotion history to actual transactions, they turn into powerful predictors of future responses. Unfortunately, many CDPs fall short of this, as companies give up without even trying. Online behavior Web visitors may come as responses to outbound promotions, but click data originating from channels such as organic and paid search, social media, affiliates and referrals must be connected to buyers as well. Again, the main issue would be the size of such data, and “what they browsed” may not be as powerful as “what they bought.” But click data is great directional guidance with high data coverage and an essential ingredient for proper channel attributions. Qualitative data In the age of Big Data, there are lots of unstructured data, such as online and offline interaction logs, usage and movement data, sentiments and survey results. Major social media companies may never share the identity of individuals, though, and what people say they do and what they actually do may differ vastly. But, if they come with some identifiers, survey data can be very useful for attitudinal modeling. I’ve listed these in a rough order of data potency and ease of manipulation, but depending on marketing goals, the rank may change dramatically. The first thing is to develop a data roadmap in phases. Then, go after one data source at a time, starting with the low-hanging ones with greater usefulness. Don't try to take a giant leap all the way to the fictitious finish line, as completing just the first phase should yield immediate benefits. Why You Should Care More About Customer Satisfaction Than You Currently Do