Building Customer Insight

IT'S A SIMPLE FACT: Great direct marketing and CRM programs don't happen by accident. They're the products of insights that target relevant messages to the most receptive customers or prospects.

Unfortunately, while firms spend countless hours creating and executing complicated marketing plans, very few devote the same amount of time and energy to understanding their customers. A company can do this by taking an inventory of its customers, profiling them, building tactical tools and looking at their lifetime value.

The first step in building customer insight is to take a comprehensive inventory of customer data. Before beginning any customer analysis all customer data must be identified and documented. This process will typically start in the information technology department if a company has a customer data warehouse or a marketing database. However, the accounting or finance departments may hold any billing or transactional data not housed in the larger systems. In addition, the sales department may have its own contact or prospect database. This is especially true in business-to-business companies, since they are most likely to store transactional billing and fulfillment data in one system and marketing or contact information in another. Finally, internal or external research groups are likely to have databases that contain analytical data, such as customer demographics or survey information.

After these data sources are identified, take a hard look at the quality and quantity of each. Audit the data for accuracy, examine its timeliness and determine what holes or problems may exist. The data may be too complicated, inaccurate or out of date to provide any real customer insight. On the other hand, there might be a wealth of transactional sales and billing information but little or no descriptive data, such as consumer demographics or business “firmographics.” If that is the case, dollars and time may be necessary to gather such data elsewhere, most likely through third-party overlays.

Once the information has been assembled it's time to get to work. Although it's tempting to build sophisticated statistical models of customer behavior right away, the best next step is to build a profile of customers. To begin, a firm should try to incorporate as many dimensions of customer behavior as possible. It should find ways to creatively combine customer purchase history with demographics, geography and qualitative or psychographic data in order to paint a rich picture of customers. While it may be helpful to know customers' average age and household income, it's much more valuable to understand, for example, the demographic differences between high- and low-value customers, repeat vs. new or lapsed customers, or multiproduct vs. single product buyers. Companies with many products should also build profiles of the buyers of different products or product categories.

This step does not necessarily require sophisticated tools or techniques. A firm should simply segment customers into different, commonsense groups like the ones described above, and then use simple cross-tabulation analysis and queries to compare and contrast the different customer segments. However, a much more robust profile can be built using multivariate statistical techniques. CHAID and decision-tree analysis are especially effective in building simple profiles of different customer groups, while other methods like cluster analysis, factor analysis, and latent class analysis are useful in identifying different customer segments or profiles that may exist within the customer base.

With a solid understanding of key customer profiles, it's time to turn attention to modeling the customer life cycle — customer acquisition, retention or repurchase, cross-sell and lifetime value. By examining each of these behavioral types, models that can direct targeted, data-driven marketing efforts will result.

A response or acquisition model will provide a tool that identifies the prospects most likely to respond to an offer or purchase a product for the first time. Such a model can be a “cloning” model that compares customers with a random sample of prospects, or a response model that uses the results of a promotion to model responders and non-responders. In either case, a company will need to rely only on external data such as demographics or firmographics, since prospects ultimately will be scored with this model.

Once a customer is acquired, statistical models can be built using a customer's own behavioral data to predict their likelihood of retention or repurchase. These models are critical in helping to understand which customers are most at risk of attrition.

Customers may require proactive marketing to retain their business. If a company sells many different items or services, product-specific cross-selling models will help a firm understand who it should be offering additional products.

Finally, by measuring and modeling customer lifetime value, a company can score customers according to their potential lifetime value. Combining this score with the retention model, customers are then viewed in two key dimensions, value and risk. A truly data-driven, customer-focused marketing strategy will result.

We haven't discussed the types of statistical models that should be built, because that isn't the most important thing to consider. Whether CHAID, linear or logistic regression, neural networks, genetic algorithms or some other technique is chosen, predictive modeling's success depends first and foremost on the quality of data. After that, it's the skill and creativity of the modeler that will have the greatest impact on a model's accuracy.

Remember, data and models alone don't ensure customer understanding. At each step along the way, a firm should look for creative ways to incorporate qualitative information about why its customers behave the way they do.

Use customer surveys or focus groups to gain insight into the most loyal customers, most at-risk customers or best prospects. While models and profiles will help a business decide who to market to, these tools will reveal insights that will allow meaningful, relevant customer communications.

JOSEPH M. DeCOSMO is president of DeCosmo and Associates Inc. in Burr Ridge, IL. He is program chair of the Chicago Association of Direct Marketing's DM Days (April 23-24).