Response Modeling: A Strong Beginning for Predictive MarketingData mining technology has undergone dramatic advancements in the past decade, significantly increasing its value as a strategic tool for delivering results.
Harnessing the technology to develop a stronger marketing plan has proved to be not only a viable option for companies to increase profitability but also a necessary one. With the ever-increasing competitive nature of the marketplace, corporations are scrambling to put the information they have "mined" over the past decade into action.
Corporations engaging in this technology have a wealth of unique information that will help them transition into a "customer-centric" organization in order to gain a competitive advantage. A wide variety of readily available methodologies and software tools can tap into this customer data to provide meaningful intelligence for strategic and tactical marketing decisions.
Improving ROI With Response Modeling
Response modeling is one of the primary areas in which predictive data mining techniques can be easily deployed to produce immediate improvements to marketing ROI.
Response modeling improves customer response rates by targeting those prospects most likely to react to a particular treatment, campaign, advertisement, media or promotion. This means that instead of mailing or telephoning every prospect on a list, marketers can select only those individuals with a high probability of responding positively. Profits from each campaign are significantly higher because marketers are contacting only likely respondents.
To illustrate this potential in numerical terms, response can increase by 50 percent, while the size of the mailing can decrease by 70 percent. In another instance, campaign response may stay the same, while the size of the mailing is reduced by 50 percent, still saving potentially hundreds of thousands of dollars in mailing costs for each marketing campaign.
These results are achieved by building a model that predicts how a given individual is likely to respond based on demographics, lifestyle, psychotropic data and purchasing history. The point of response modeling can be summed up simply: By analyzing relevant customer data, you can predict and influence customer behavior.
Response Modeling From a Technical Viewpoint
Data-driven algorithms are commonly used for response modeling. Data-driven algorithms learn the relationship between a set of input, or independent, variables and an output, or dependent, variable. Learning occurs during a process called training, in which the model is presented with examples of input-output pairs.
In the case of response modeling, inputs are the customer attributes that may influence the response rate. Output is the propensity of a prospect to respond. Marketers must send a test treatment and collate responses. The model is built on the results of the test treatment and is used to score the remaining list of prospects, refining the list to include only the most likely candidates.
Algorithms commonly used for response modeling include linear regression, logistic regression and neural networks. Because a better model results in a better bottom line and higher ROI, marketers should consider all major modeling approaches. Technology is now available that permits rapid application of all approaches, allowing a marketer to choose those that best represent its data and objectives.
Marketers often use a lift chart to visualize and measure response-modeling performance. The X axis represents the percentage of prospects -- for example, in a mailing list -- to whom a marketer sends product literature. The Y axis indicates the percentage -- relative to all the potential responses -- of responses achieved. By using a model to score prospects, businesses can sort them by the predicted likelihood of their responses. This usually generates a "lift curve."
Automated Intelligent Search
The process of developing a good model is iterative and complex. It typically involves building many models and testing multiple algorithms to find the best set of customer attributes to use for prediction as well as the best set of algorithm parameters for learning. While an expert modeler can do this search manually, it is labor-intensive and prone to reflect the modeler's biases.
While it is impractical to perform an exhaustive search, automated intelligent searches are an excellent alternative. These methods can search huge, high-dimensional spaces efficiently, focusing on solutions that show the greatest promise. These search methods build and compare different models, letting the data determine which one is best, thus eliminating bias. Evaluating hundreds of models instead of a handful vastly increases one's chances of finding a better model and greater bottom-line savings.
Response Modeling to Reduce Churn
Attrition, the defection or turnover of a portion of a company's customer base, is a problem for all industries, especially telecommunications. For example, global wireless telecom companies face annual attrition rates estimated as high as 35 percent. Attrition, or "churn," rates are expected to peak at 40 percent in the United States.
One of the most effective ways businesses can combat this problem is to employ data mining for customer-centric marketing. A business can use its existing customer database along with data mining tools to build models that predict which customers are most at risk for churn and why. Armed with this knowledge, the business can intercede, making those customers an offer likely to entice them to stay.
Predicting which customers will leave is necessary for actively managing a company's customer base, but it is not sufficient in itself. A generally accepted adage applies here: Twenty percent of your customers generate 80 percent of your revenue. Marketers need a way to identify their most profitable customers, as well as customers or prospects that might become high value. This enables marketers to target them, as well as to design incentives to maintain their loyalty. Occasionally, marketers might even want to encourage low-value customers to defect to the competition.
Building Better Models: A Key Advantage
Building better models can be a key competitive advantage in marketing efforts. In a world where more and more consumer data is collected and readily available, data mining and response modeling are becoming everyday weapons in the marketer's arsenal.
Current purchasing trends indicate a strong familiarity with the benefits of response modeling and the approaches behind it, including predictions of propensities to purchase or the likelihood of attrition. However, data mining encompasses many more theories that apply to industries across the board. Data mining is in the growth stage at most companies.
In the near future, we will see more cross selling, customer valuation and profiling/segmentation to better identify buyers and customers. As we know from the example response modeling sets, each tactic has benefits and rewards clearly worth the investment if a data set is available in any form. Because the returns on investing in predictive technology are fast and quantifiable, businesses in a competitive marketplace cannot afford to be without some degree of predictive data mining.
Companies that master and refine these abilities will bring customer relationship management and marketing ROI to unprecedented heights.