Segment Customer Information for Increased ROIAs the economy continues to struggle, marketing budgets are being further reduced and companies are asking direct marketing departments to do more with less. We no longer have inexhaustible resources like we did in the dot-com boom.
Relying on nearly forgotten best practices of optimized targeting and constant learning, DM is reverting to its baseline tenets of accountability and profitability.
Direct marketing was built on three necessities: list, offer, creative. If you have a list of customers, a compelling offer and a means of delivery, you can execute a program. But in today's environment of squeezing every bit of efficiency out of campaigns, simply blasting out a message, even through an inexpensive channel such as e-mail, is not acceptable. Nor should it be. With data-driven segmentation, cost efficiencies are easily realized and marketing ROI improved. I will discuss four methods for segmenting customer data to better generate targeted lists, increase response rates and decrease marketing costs.
1. Directed queries. The simplest form of segmentation occurs when a domain expert (i.e., the marketer) provides a description, usually based on experience, of the various unique targets in the customer/prospect base. The analyst queries a database, identifies the various audiences described by the domain expert and provides a data profile of each. The various audiences then can be targeted with the message/offer that best corresponds to their unique characteristics.
For example, a nationwide clothing retailer may think there are strong differences in buying patterns based on region, gender and socioeconomic status. The analyst can query the customer database along each of these dimensions to better quantify how the company's customers are distributed within each variable. The first pass of the queries may provide the following:
o Counts for the various values/value ranges within each variable (e.g., percentage of customers living in each of six regions of the country, percentage of customers who are male).
o Cross-tabulations of each variable. After analyzing the output, the analyst may find interesting correlations, such as the number of high-income males in region X is disproportionately high.
o Corporate financial overlays with analyses done on each segment to identify opportunities, such as females in region Y are the most profitable to the company.
With each pass of the data and subsequent increase in knowledge, the marketer may ask deeper questions. The analyst can quickly query the data to provide an increasingly granular view of the customer base. With this new information, the marketer can outline a strategy for each segment in the base, providing a relevant experience for each.
2. Exploratory data analysis. An automated approach to analyzing the data often lends better insight into customer segments than manual querying. EDA lets the analyst feed all data elements into a statistical algorithm to determine which variables are most indicative of successful marketing results or customer behavior patterns. Analysts commonly uncover data that contradict deeply held beliefs on best customer segments in an organization.
For example, queries at a Web hosting company may produce a report showing that young, hi-tech companies are a majority of the customer base. This would lead to the conclusion that young, hi-tech companies are the best targets for the service. But this finding may result more from previous targeted marketing efforts (they may have focused on this segment in previous campaigns) than reality.
An unbiased look into the data might reveal that small retailers in business more than two years have higher response rates to acquisition campaigns, and medium-sized wholesale companies that do $2.5 million to $5 million in sales are more likely to sign up for premium services and are the most valuable customers.
When producing an EDA, ensure that the segments you create are actionable. Often, segments are very small or do not represent true marketing opportunities. For example, an EDA may reveal that customers with 10+ years of tenure are the best targets for a new product. But if fewer than 1 percent of customers have 10+ years of tenure, it may not be feasible to develop a campaign around this segment. The marketer needs to work with the analyst after the publication of an EDA to craft actionable segments from the findings.
It is important to note that EDAs are not completely automated. Algorithms used to sift through the data require little user input, but analysis of the results and subsequent crafting of segments are done manually. Most EDA reports supply bi-variate correlations; these correlations describe the influence of one variable on another. Manual input from the analyst is required to paint a deeper picture of how multiple variables interact.
3. Undirected segmentation. The terms "cluster analysis" and "undirected segmentation" refer to segmentation techniques that are completely automated. Clustering algorithms were designed to uncover correlations and trends involving multiple variables that are difficult to detect through manual analysis. These algorithms can analyze large amounts of data and identify patterns (usually behavioral) that are repetitive in a significant percentage of the customer base. Clustering is extremely powerful for organizations with large, information-rich databases and many product/service offerings.
For example, a wireless telecommunications company may be struggling to package its array of products into logical bundles. The wireless industry is highly competitive, and often customers switch providers because they are unaware of existing products from their current provider that meet their needs or they are presented with competitive plans that reduce their costs and/or provide increased capabilities.
To combat these competitive threats, an undirected segmentation study can identify various customer segments based on similar use patterns. The marketer then can match the appropriate product mix and service plan to each segment. Assume the study produces these segments:
o Busy salespeople. These customers spend at least 5,000 minutes monthly on their mobile phone, most of their calls are long distance, fully use advanced features and spend most of their time outside of their home area.
o Wireless preferred. These customers spend more than 2,500 minutes monthly on their mobile phone, have a mix of long distance and local calling, rarely leave their home calling area, and many of their calls are to a small group of numbers.
o Weekend warriors. These customers spend at least 95 percent of their time on their mobile phone during off-peak hours.
o Emergency users. These customers use less than 50 percent of their allocated monthly minutes, rarely have calls longer than two minutes, and 90 percent of their calls are to or from a single number.
By using the above schemes, marketers and product managers can understand the segments in the population and target offers and/or products more effectively. Many EDA studies would have difficulty identifying the differences between the first two segments. Both groups are highly valuable and spend much time on their phones. But the two groups differ greatly in their use and perceptions of wireless products. Plans and/or bundled offers can be made to each that could increase their lifetime value to the company.
4. Predictive modeling. Predictive modeling has proven its value in almost every facet of customer communication. Modeling is rooted in the collections/risk management industry and has blossomed as a targeted marketing application. Unlike the first three methods, which focus on describing current customer behavior, predictive modeling forecasts a customer's future behavior. The output is the probability that a customer will demonstrate the behavior in question, so the marketer can elect to contact only customers with a high enough probability to produce a positive campaign ROI.
Let's take a software company whose revenue model includes selling upgraded packages to existing customers. It could develop a predictive response model for its annual upgrade campaigns. The model would rank customers based on their likelihood to buy an upgrade.
Only those most likely to upgrade would be selected for future campaigns. By eliminating customers with the lowest likelihood of responding, valuable marketing dollars can be used for better offers, more waves of communication, more expensive mail pieces or saved altogether.
It is critical to realize that an analysis is only as good as the execution of the program that uses it. The four methods of segmentation discussed here require input from both the analyst and marketer. That is the beauty of direct marketing; analysts can hone their craft creating data-driven tools that are wielded by marketers looking to maximize ROI.
It is a symbiotic relationship with the result being stronger customer relationships, increased revenue, optimized campaign ROI and, of course, better direct marketing.